Health Technology Update — Issue 22

(October 26, 2022)

In This Issue

Issue 22 — October 2018 — Artificial Intelligence Issue

This issue of Health Technology Update features brief summaries of information on a range of artificial intelligence technologies — from chatbots to systems for the detection of cognitive impairment and dementia. These technologies were identified through the CADTH Horizon Scanning Service as topics of potential interest to health care decision-makers in Canada.

  • IDx-DR: Automated Screening for Diabetic Retinopathy
  • Chatbots: AI-Based Delivery of Therapy or Coaching for Mental Health Conditions
  • Using Artificial Intelligence for Stroke in the Emergency Setting
  • Detection of Cognitive Impairment and Dementia with Artificial Intelligence
  • Focus On: Artificial Intelligence in Population and Public Health
  • Mini-Roundup: Recent Horizon Scanning Reports from CADTH and Other Agencies


Have you heard of a new health technology you think will have an impact on health care in Canada? Please let us know!



Issue 22: Artificial Intelligence

Artificial intelligence (AI) is a branch of computer science concerned with developing programs to perform tasks that would usually require human intelligence.1-4 Recent advances in computing power combined with declining costs have led to an expansion of AI research and applications, including in health care.4,5 Common approaches to AI include: machine learning — training an algorithm to perform tasks by learning from patterns in data rather than performing a task it is explicitly programmed to do;5 support vector machine — a type of machine learning often used in the diagnosis or disease prediction to classify the absence or presence of a condition;6 artificial neural networks — a method of mimicking the way the human brain learns and is used for solving complex problems where relationships are unclear;4,6 and deep learning — a more recent form of artificial neural network that has many hidden layers of decision-making between input and output.6

CADTH’s publication An Overview of Clinical Applications of Artificial Intelligence7 serves as a starting point for individuals seeking a better understanding of AI concepts and their emerging uses in health care.

In this issue of Health Technology Update, we present readers with five articles about new and emerging health technologies and clinical areas that use AI:

  • IDx-DR, a diagnostic system designed to autonomously screen retinal images of people with diabetes for early detection of diabetic retinopathy
  • chatbots, AI-powered conversational agents that attempt to mimic human therapists for people living with mental health disorders
  • e-ASPECTS, an automated software enhanced with AI to assist physicians in identifying and interpreting the extent of brain damage in acute ischemic stroke patients from computed tomography, or CT, exams
  • BrainFx, a tablet-based cognitive and neurofunction assessment tool that uses AI to detect early risk factors for mild to moderate brain dysfunction
  • the many ways population and public health researchers are exploring how AI could impact disease surveillance, forecasting, and modelling.

Author: Jeff Mason


  1. Beam AL, Kohane IS. Translating artificial intelligence into clinical care. Jama. 2016;316(22):2368-2369.
  2. Chartrand G, Cheng PM, Vorontsov E, et al. Deep learning: a primer for radiologists. Radiographics : a review publication of the Radiological Society of North America, Inc. 2017;37(7):2113-2131.
  3. Fogel AL, Kvedar JC. Benefits and risks of machine learning decision support systems. Jama. 2017;318(23):2356.
  4. Standing Senate Committee on Social Affairs, Science and Technology. Challenge ahead : integrating robotics, artificial intelligence and 3D printing technologies into Canada’s healthcare systems. Ottawa (ON): Canada Senate; 2017: Accessed 2018 May 24.
  5. Ching T, Himmelstein DS, Beaulieu-Jones BK, et al. Opportunities and obstacles for deep learning in biology and medicine. bioRxiv. 2018.
  6. Jiang F, Jiang Y, Zhi H, et al. Artificial intelligence in healthcare: past, present and future. Stroke and Vascular Neurology. 2017;2(4):230-243.
  7. Mason J, Morrison A. An Overview of Clinical Applications of Artificial Intelligence. CADTH Issues in Emerging Health Technologies. 2018.


IDx-DR: Automated Screening for Diabetic Retinopathy

Diabetic retinopathy is the most common cause of vision loss and blindness in adults, in Canada.1-5 It occurs when high blood sugar levels damage blood vessels in the retina — the light-sensitive tissue located at the back of the eye.3 Deep learning algorithms are being trained, using retinal images of varying disease severities, to autonomously identify early diabetic retinopathy and prevent vision loss.1,2,6-8

How It Works

IDx-DR (IDx, LLC, Coralville, Iowa) is an artificial intelligence (AI) diagnostic system designed to autonomously screen retinal images for signs of diabetic retinopathy.9 IDx-DR is intended for the primary care diagnosis of moderate to severe diabetic retinopathy and macular edema in adults 22 years of age and older with diabetes and no previous diabetic retinopathy diagnosis.9,10

IDx-DR can provide screening decisions without the need for a clinician to interpret the images and results.3 Clinic staff can be trained in four hours to operate IDx-DR, as the device provides guidance for imagetaking.11 IDx-DR is only to be used with the Topcon TRC-NW400 non-mydriatic retinal camera.9 Two 45-degree field of view colour images of the retina per eye are taken by either clinical or non-clinical staff trained to operate the camera.3,9 The camera images are then uploaded to the IDx-DR client user in the local computer and submitted to the IDx-DR analysis system located on a secure server.3,9 IDx- DR uses deep learning detector algorithms to look for diabetic retinopathy-specific lesions in the images.9 The algorithms evaluate each image for the disease and image quality.9 If the image quality is insufficient, IDx-DR software will indicate that immediate reimaging is needed.9

The AI test results are provided in less than one minute and indicate if the images are negative for more than mild diabetic retinopathy (requiring retesting in 12 months), or positive for more than mild diabetic retinopathy.9 Based on the results, the clinician determines if an individual requires a referral to an ophthalmologist for further evaluation and treatment.9 An associated report is also produced, with care instructions for the provider.12

Automated systems have the potential to offset high demands for screening, and reduce rates of missed diagnoses and intra-expert variation in screening.2,6,13 These systems may also reduce the screening workload because of efficient automated analysis of large numbers of retinal images, and could improve clinic-level workflow efficiency and coverage, allowing for increased patient interaction.1,2,14

Who Might Benefit?

In 2013-2014, approximately 3 million Canadians lived with diagnosed diabetes and about 200,000 were newly diagnosed with diabetes.15 An estimated 11.4% of the Canadian population will be diagnosed with diabetes by 2025.16

All individuals with type 1 or type 2 diabetes are at risk for diabetic retinopathy.17 In Canada, the disease affects about 500,000 people, and is more common in Indigenous populations.4,17 In 2014, 66% of Canadians living with diabetes reported receiving an eye screening exam in the past year.16

IDx-DR is not intended to be used to screen individuals with laser eye treatment history; injections or surgery in the eye; floaters; blurred vision; persistent vision loss; severe non-proliferative, proliferative, or radiation retinopathy; retinal vein occlusion; previously diagnosed macular edema; or pregnant individuals with diabetes.3

Availability in Canada

IDx-DR is currently unavailable in Canada.9

The system received FDA clearance on April 11, 2018 under the De Novo premarket review pathway for novel low to moderate risk devices having no prior legally marketed devices.3 IDx-DR is being used for patient care in the Diabetes and Endocrinology Centre at University of Iowa Health Care, and will be introduced into several other US health care systems in 2018.18

IDx-DR received the CE certification from the Underwriters Laboratory in 2013, and a Class IIa Medical Devices classification for sale in the European Union on April 23, 2016.9 IDx has partnered with Medical Workshop in the Netherlands, and IBM Watson Health for the rest of Europe, for device distribution.9

What Does It Cost?

The costs of purchasing and operating IDx-DR were not identified. In the Harris Health System (Texas, US), the cost of a similar automated screening system for diabetic retinopathy, without AI features — the Intelligent Retinal Imaging Systems (IRIS) — is approximately $55 US per person.19

Current Practice

The 2018 Diabetes Canada diabetic retinopathy guidelines recommend regular, comprehensive, dilated eye exams for the early detection of treatable diabetic retinopathy.4 If retinopathy is not detected, annual screening is recommended for people with type 1 diabetes, and every one to two years for people with type 2 diabetes.4

Screening can be performed with stereoscopic-colour fundus photography, digital stereoscopic retinal photography, digital ophthalmoscopy, or optical coherence tomography by “qualified vision care professionals (ideally optometrists or ophthalmologists).”2,4,6,8,20 Screening services based on trained human graders (non-ophthalmologic or ophthalmologic graders) are a suggested first-line assessment in order to reduce the number of people requiring further specialized ophthalmological assessment.1 Telemedicine programs relying on fundus photography and ultra-wide field imaging are used widely in Canada to manually identify and triage people with diabetic retinopathy.4

In Canada, automated screening for diabetic retinopathy is not recommended by guidelines for clinical practice because of present limitations including technical failures in retinal vessel identification.21

Published Information

Published Studies

In a validation study in the Netherlands, retinal images from 1,371 people with diabetes in the Hoorn Diabetes Care System were graded by IDx-DR, and independently by three specialists, to determine the accuracy of automated retinopathy screening in primary care.22 In a US observational study,23 retinal images from 892 people (aged 22 to 84) with diabetes from 10 primary care clinics were evaluated to determine the accuracy of IDx-DR in detecting more than mild diabetic retinopathy.

Conference Abstracts

Results from the following studies are available within conference abstracts:

  • A validation study in the US of 528 people with diabetes from five ophthalmologic centres compared the accuracy of IDx-DR to ophthalmologist examinations in detecting diabetic eye disease.26
  • In a study in the Netherlands, 1,500 people with type 2 diabetes in the Diabetes Care System West-Friesland were screened for retinopathy using IDx-DR.27 IDx-DR was evaluated for accuracy in comparison to manual grading by three retinal specialists, and workflow changes resulting from the use of IDx-DR were measured.27

Registered Clinical Trials

A multi-centre observational study of 600 participants was reported as complete in 2014 but published results were not identified. The study compared the performance of IDx-DR to ophthalmologistperformed dilated eye examinations in its ability to distinguish between no or mild non-proliferative diabetic retinopathy without macular edema and more than mild non-proliferative diabetic retinopathy with or without macular edema.28


The letter issued by the FDA following a De Novo classification of the system identified risks to health with using IDx-DR including false-positive results, which lead to additional unnecessary medical procedures; false-negative results, which delay further evaluations and treatments as a result of diagnostic algorithm or software failure; and operator failure in providing images meeting input quality specifications.10

Issues to Consider

The ability of deep learning algorithms to detect referable diabetic retinopathy and reduce incorrect readings depends on the amount and quality of data available for training.2,14 Automated systems may produce incorrect diagnoses when algorithms are unable to detect lesions because of variations in image lighting, contrast, clarity, background texture, and fundus camera quality.14,29,30

The cybersecurity of medical images is a growing concern, particularly with image-sharing over networks and storage in electronic databases.31 Digital watermarking can be used to hide patient identity in medical images; however, it can make slight computational changes to the retinal images, which may affect automated diagnosis.31 New algorithms are being developed to ensure that embedded watermarks maintain an individual’s health information without causing information loss or compromising diagnostic accuracy.31

IDx-DR accepts most image formats and uses common output formats for compatibility with electronic medical record systems and software.9

Related Developments

AI Diagnosis of Eye Disease

Additional pilot studies for the IDx-DR software are pending in Finland, the Czech Republic, and Italy. IDx-DR plans to conduct trials to expand its indication for use with other camera manufacturers (Laura Shoemaker, Director of Marketing Communications, IDx, LLC, Coralville, Iowa: personal communication, 2018 Jun 28).

IDx, LLC is developing algorithms for detecting disease across multiple imaging tools, with a focus on fundus imaging and optical coherence tomography.9 Algorithms are being developed for detecting stroke risk, cardiovascular disease, Alzheimer disease, signs of age-related macular degeneration (IDx-AMD), and glaucoma (IDx-G) indicators from retinal images.9

Other automated diagnostic systems for diabetic retinopathy screening include EyeCheck, iGradingM (Medalytix (Group) Ltd., Manchester, UK), EyeArt and EyeApp (Eyenuk, Woodland Hills, CA), and Retmarker (Retmarker, SA, Taveiro, Portugal).1,7,32

Automated teleretinal systems for diabetic retinopathy screening include the cloudbased SELENA automated software, and IRIS.7,30

DeepMind (Alphabet Inc.) has developed and is testing an automated deep learning algorithm designed to detect 50 eye diseases (including diabetic retinopathy) using optimal coherence tomography scans.33

Looking Ahead

In Canada, in 2007, the estimated cost to the health care system for vision loss due to diabetic retinopathy was $205 million.34

There is interest in enhancing the precision of automated retinal screening by combining algorithms from multiple imaging tools, and by developing deep learning algorithms that could infer disease progression patterns, determine reliable predictors of diabetic retinopathy, and combine image data with other health data for more comprehensive information, including risk of systemic disease.8,35 Using deep learning algorithms for screening could also contribute to the universal standardization of retinal image grading across multiple populations.8

AI may also be used for teleretinal screening programs, which may help overcome geographical barriers and wait times to increase access to screening.30,35

Because IDx-DR is currently unavailable in Canada, the impact of this technology on human resources for diabetic retinopathy screening in the Canadian health care system is unclear.

Author: Humaira Nakhuda


  1. Norgaard MF, Grauslund J. Automated Screening for Diabetic Retinopathy - A Systematic Review. Ophthalmic Res. 2018;60(1):9-17.
  2. Orlando JI, Prokofyeva E, del Fresno M, Blaschko MB. An ensemble deep learning based approach for red lesion detection in fundus images. Comput Methods Programs Biomed. 2018;153:115-127.
  3. US food and Drug Administration. FDA permits marketing of artificial intelligence-based device to detect certain diabetes-related eye problems. 2018 Apr 11; Accessed 2018 Aug 15.
  4. Altomare F, Kherani A, Lovshin J. Retinopathy. Can J Diabetes. 2018;42 Suppl 1:S210-s216. Accessed 2018 Aug 15.
  5. Fraser CE, and D'Amico, D.J. Diabetic retinopathy: Classification and clinical features. In: Post TW, ed. UpToDate. Waltham (MA): UpToDate; 2016 Oct 28.
  6. Powers M, Greven M, Kleinman R, Nguyen QD, Do D. Recent advances in the management and understanding of diabetic retinopathy. F1000Res. 2017;6(2063):2063.
  7. Rajalakshmi R, Subashini R, Anjana RM, Mohan V. Automated diabetic retinopathy detection in smartphone-based fundus photography using artificial intelligence. Eye. 2018;09:09.
  8. Somfai GM, Gerding H, Debuc DC. The Use of Optical Coherence Tomography for the Detection of Early Diabetic Retinopathy. Klin Monbl Augenheilkd. 2018;235(4):377-384.
  9. IDx -DR. Coralville (IA): IDx, LLC; 2018: Accessed 2018 Aug 15.
  10. Center for Devices and Radiological Health (CDRH). Letter to IDx. SIlver Spring (MD): US Food and Drug Administration; 2018 Apr 11: Accessed 2018 Aug 15.
  11. Doheny K. First In-Office Screening for Diabetic Retinopathy Cleared by FDA. Montclair (NJ): EndocrineWeb, Vertical Health, LLC; 2018 Apr 12: Accessed 2018 Aug 15.
  12. FDA Permits Marketing of IDx-DR for Automated Detection of Diabetic Retinopathy in Primary Care. 2018 Apr 12; Accessed 2018 Aug 15.
  13. You Z, Hu X, Shi K. Will artificial intelligence replace ophthalmologist in diabetic retinopathy screening? Biomedical Research (India). 2017;28(15):6920.
  14. Rahimy E. Deep learning applications in ophthalmology. Curr Opin Ophthalmol. 2018;29(3):254-260.
  15. Public Health Agency of Canada. Diabetes in Canada. 2017 Nov 14; Accessed 2018 Aug 15.
  16. 2015 Report on Diabetes: Driving Change. Toronto: Diabetes Canada; 2015: Accessed 2018 Aug 15.
  17. Eye Damage (Diabetic Retinopathy). Toronto: Diabetes Canada; 2018: Accessed 2018 Aug 15.
  18. Press release: University of Iowa Health Care first to adopt IDx-DR in a diabetes care setting. Coralville (IA): IDx, LLC; 2018 Jun 26: Accessed 2018 Aug 15.
  19. Garoon RB, Chu Y, Gupta S, Weng CY. Automated teleretinal screening for diabetic retinopathy in the harris health system: Cost analysis & economic impact. Invest Ophthalmol Vis Sci. 2016;57 (12):1585.
  20. McCulloch DK. Diabetic retinopathy: screening. In: Post TW, ed. UpToDate. Waltham (MA): UpToDate; 2017 Jun 16.
  21. Hooper P, et al. Canadian Ophthalmological Society evidence-based clinical practice guidelines for the management of diabetic retinopathy. Can J Ophthalmol. 2012 Apr;47(Suppl 1):1-30. Accessed 2018 Aug 15.
  22. van der Heijden AA, Abramoff MD, Verbraak F, van Hecke MV, Liem A, Nijpels G. Validation of automated screening for referable diabetic retinopathy with the IDx-DR device in the Hoorn Diabetes Care System. Acta Ophthalmol (Oxf). 2018;96(1):63-68.
  23. Abràmoff MD, Lavin PT, Birch M, Shah N, Folk JC. Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices. npj Digital Medicine. 2018;1(1):39.
  24. Abramoff M. Artificial Intelligence for Automated Detection of Diabetic Retinopathy in Primary Care. 2018 Feb 22; Accessed 2018 Aug 15.
  25. IDx LLC. A Multi-center study to evaluate performance of an automated device for the detection of diabetic retinopathy. Bethesda (MD): U.S. National Library of Medicine; 2016 Nov 15; updated 2018 Jan 17:
  26. Maguire MG, Daniel E, Niemeijer M, Pistilli M, Folk JC, Abramoff MD. Identifying diabetic eye disease: Comparison of clinical examination by ophthalmologists to automated detection from retinal color images. Invest Ophthalmol Vis Sci. 2015;56 (7):2014.
  27. Nijpels G, Abramoff M, Niemeijer M, Talmage E, Van Der Heijden A. Real-world workflow effects of automated diabetic retinopathy screening in a primary diabetes care setting with the IDx-DR device. Eur J Ophthalmol. 2015;25 (3):e20.
  28. IDx LLC. Computer detection of diabetic retinopathy compared to clinical examination (CDDR). Bethesda (MD): U.S. National Library of Medicine; 2012 Jun 21; updated 2014 Apr 22 Accessed 2018 Aug 15.
  29. Dai B, Wu X, Bu W. Retinal Microaneurysms Detection Using Gradient Vector Analysis and Class Imbalance Classification. PLoS One. 2016;11(8):e0161556.
  30. Sun JK, Cavallerano JD, Silva PS. Future Promise of and Potential Pitfalls for Automated Detection of Diabetic Retinopathy. JAMA Ophthalmol. 2016;134(2):210-211.
  31. Singh A, Dutta MK. Imperceptible watermarking for security of fundus images in tele-ophthalmology applications and computer-aided diagnosis of retina diseases. Int J Med Inf. 2017;108:110-124.
  32. Tufail A, Rudisill C, Egan C, et al. Automated Diabetic Retinopathy Image Assessment Software: Diagnostic Accuracy and Cost-Effectiveness Compared with Human Graders. Ophthalmology. 2017;124(3):343-351.
  33. De Fauw J, Ledsam JR, Romera-Paredes B, et al. Clinically applicable deep learning for diagnosis and referral in retinal disease. Nat Med. 2018.
  34. Liu S. CLEAR SIGHT: a trial of non-mydriatic ultra-widefield retinal imaging to screen for diabetic eye disease. Bethesda (MD): U.S. National Library of Medicine; 2015 Oct 20; updated 2018 May 3: Accessed 2018 Aug 15.
  35. Chee RI, Darwish D, Fernandez-Vega A, et al. Retinal Telemedicine. Current Ophthalmology Reports. 2018;6(1):36-45.


Chatbots: AI-Based Delivery of Therapy or Coaching for Mental Health Conditions

Chatbot technologies — artificial intelligence (AI)-based conversational agents — marry mobile and Internet-based therapies with simulated human conversation to help address mental health problems. Available 24/7 and to anyone with an Internet connection, chatbots are positioned as an adjunct to conventional therapy, and as a way of increasing access to therapeutic education for individuals in need.

How It Works

Conversational agents are a form of AI that mimic real-life human interactions through a conversational approach, much like talking with a therapist or friend.1 This software is activated by natural language input (text, speech, or both) and uses a pattern-matching algorithm to link user words with topic categories. It then executes goal-directed commands through these topic-linked inputs.

One type of conversational agent, chatbots, are commonly employed in customer service or social media interactions. Another, embodied conversational agent (ECA), is like a chatbot but with the addition of a virtual avatar that embodies the AI computer system.1 All modern chatbots trace their origins to ELIZA — the original chatbot created by a psychotherapist — and have now broadened out to a variety of different applications and uses. Chatbots can deliver psychological techniques modelled on real psychotherapy, such as cognitive behavioural therapy (CBT) or positive psychology, through this conversational interface. Although some chatbots may mimic human interaction, they are generally not intended as a replacement for human-led therapy. Chatbots provide responses that can lead patients to videos, mood tracking, and core concepts in psychological interventions.2 Chatbots may also have machine learning, which allows them to adapt to new information.

Some chatbot technologies have security systems to help prevent breach of data. For example, Woebot attempts to eliminate security issues with user data by encryption, compliance with data privacy legislation, and informed consent (Dr. Athena Robinson, Chief Clinical Officer, Woebot, San Francisco, CA: personal communication, 20184 Jun 4).

Who Might Benefit?

Individuals with mental health disorders or individuals experiencing distress could potentially benefit from chatbots. Mental health conditions are common: one in three Canadians will meet the criteria for having a mental health disorder during their lifetime.3 There is also a high societal, economic burden of mental illness through absenteeism, unemployment, loss of productivity, and medical expenses.3

Accessing psychological services can be challenging — issues such as stigma,2 limited infrastructure,4 and lack of access often prevent people from seeking help.4,5 Mental health chatbots may help bridge these issues. Online interventions also provide patients with immediate access to resources and information.6

It has been reported that individuals interact with chatbots as they would a human therapist, and are more likely to share information on sensitive topics (such as mental health) than with human counterparts.2,7-9 Online interventions also provide anonymity for users.6 It has been reported that anonymity and rapport are important in interviews, as they lead to greater disclosure from patients.7

Availability in Canada

A number of chatbots are available in Canada for use. Some chatbots for mental health are available through a downloadable mobile app, or through Facebook Messenger. Some chatbots or ECAs are available through specific websites and do not require an account or sign-up to access.5

Internet-delivered interventions, AI or otherwise, are unique when compared to regular therapy in that they are continuously available to individuals who have access to the Internet or to a mobile phone.10 According to the Canadian Internet Registration Authority and the International Telecommunication Union, in 2016, 90% of Canadians had Internet access, although some individuals in remote regions did not, and those with access may not have had sufficient speed and quality, or access in a private home.11

Health Canada is currently looking into the regulation process for digital health, including AI and other mobile medical applications.12 Health Canada cannot make a blanket statement regarding the regulatory classification of AI in the health care field because it is largely dependent on the characteristics of the specific application; however, so far Health Canada has not licensed any chatbots (Medical Devices Bureau, Therapeutic Products Directorate, Health Products and Food Branch, Health Canada, Ottawa, ON: personal communication, 2018 Jul 4).

What Does It Cost?

Many chatbots appear not to have a paywall.2,5,13 Some chatbots have a monthly paywall, such as Tess, which has a platform fee of US$50 per month and a patient fee of US$1 per month.14 Wysa currently has a premium product, in addition to its free AI chat, called “Wysa Coach” (US$30 per month) in which a user can talk to a trained mental health professional and share content gathered from conversations with the AI chatbot to build mental resilience skills (Ramakant Vempati, Co-founder, Wysa Ltd., Bangalore, India: personal communication, 2018 Jun 4).

In comparison, a 50-minute individual therapy session is suggested to cost between $120 in British Columbia (2016)15 and $200 in Alberta (2018).16

Current Practice

Currently, Canadian guidelines recommend CBT, interpersonal therapy, and behavioural action as non-pharmacological treatments for mood disorders such as depression.17 Second-line therapies include telephone-, Internet-, and computer-assisted therapy (none of the interventions studied appearing to include an AI chatbot component).17 For anxiety disorders, CBT is often recommended as a nonpharmacological treatment.18

Published Studies

Three randomized controlled trials were identified in the literature that assessed AI chatbots and mental health-related outcomes.2,19,20 The studies compared Woebot to information-only control,2 iPhone-based Shim to wait list control,19 and MYLO to ELIZA.20 Not all participants had a diagnosed mental health condition, and no studies were performed in a Canadian population.


One potential safety issue is the unintended side effects of intelligent agents. For example, if the programmer inputs provided to an AI are flawed or biased, the output the patient receives will be inherently flawed or biased.21 This possible violation of the therapeutic process can be attenuated through the inclusion of trained clinicians to help develop and design the chatbots.2 Additionally, AI chatbots could potentially recommend medical procedures or solutions to a patient that have more side effects than necessary, and may not be able to adapt when a patient provides unusual responses.21

Risk to patients is managed in three ways by Wysa: practical means (apologizing when there’s a misunderstanding, allowing patients to direct the conversation), ethical means (taking no personal data, natural language processing [NLP] to recognize potential ideation on suicide or self-harm), and legal means (allowing for patients to remove data, security measures on the application) (Ramakant Vempati: personal communication, 2018 Jun 4). Woebot also has safety net procedures, including unskippable informed consent, and NLP to recognize suicidal ideation and to provide resources (Dr. Athena Robinson: personal communication, 2018 Jun 10). No information was identified on the impact of these measures on risks associated with the use of the technology.

Issues to Consider

Potential issues that may arise with Internet-based AI technologies are those of access to the Internet and computer illiteracy,22 and the lack of ability for the AI to learn from individual patients.23 There are also potential ethical issues associated with the use of this technology related to storage, and access to and confidentiality of personal data.22 Data protection and privacy laws may vary by jurisdiction. This issue is especially present when the chatbot is used through third-party websites, such as Facebook.

Related Developments

ECAs have been used to deliver mindfulness and lifestyle recommendations such as healthier eating and reduction of stress.24 ECAs have also been used to create gamebased simulations for primary health care providers to help screen patients for mental health issues.25

Outside of mental health applications, chatbots can also be used to address loneliness and act as a “friend” to chat to, with no therapy function.26 Chatbots are also being developed as a resource for individuals looking to answer simple health questions that do not require a physician’s visit.27

Looking Ahead

One of the pitfalls of text-based chatbot technology is the lack of recognition of nonverbal emotion by AI technologies.28 There is interest in future research to address these non-verbal nuances in order to maintain believability and smoothness in the conversation, and to build trustworthiness with the patient.29 Currently, there is research being done with a platform, SimSensei (or “Ellie,” a continuation of SimCoach), that combines real-time audio and visual cues from the patient, and has facial expression and body language recognition for diagnostic purposes.5,28

There are reported gaps in the literature regarding whether chatbots are an appropriate intervention for treating mental health conditions or whether they should primarily be a screening tool.30 It has been noted that there is a lack of evidence regarding the implementation of the technology, and that more research is required with larger samples, proper control groups, and clinical populations.31

Author: Charlotte Wells


  1. Radziwill NB, M. Evaluating Quality of Chatbots and Intelligent Conversational Agents. Software Quality Professional. 2017;19(3):25-36.
  2. Fitzpatrick KK, Darcy A, Vierhile M. Delivering Cognitive Behavior Therapy to Young Adults With Symptoms of Depression and Anxiety Using a Fully Automated Conversational Agent (Woebot): A Randomized Controlled Trial. JMIR Ment Health. 2017;4(2):e19.
  3. Pearson CJ, T; Ali, J. Health at a glance - Mental and substance use disorders in Canada 2015; Accessed 2018 Aug 16.
  4. Tielman ML, Neerincx MA, van Meggelen M, Franken I. How should a virtual agent present psychoeducation?: Influence of verbal and textual presentation on adherence. Technol Health Care. 2017 Oct 17;25(6). Accessed 2018 Aug 16.
  5. Meeker D, Cerully JL, Johnson M, Iyer N, Kurz J, Scharf DM. SimCoach Evaluation: A Virtual Human Intervention to Encourage Service-Member Help-Seeking for Posttraumatic Stress Disorder and Depression. Rand health q. 2016;5(3):13.
  6. Elmasri D, Maeder A. A conversational agent for an online mental health intervention. Brain Informatics and Health: International Conference, BIH 2016, 2016 Proceedings, 243-251; 2016 October 13-16; Omaha, NE.
  7. Lucas GMG, J.; King, A.; Morency, L.P. It’s only a computer: Virtual humans increase willingness to disclose. Computers in Human Behaviour. 2014;37:94-100.
  8. D'Alfonso S, Santesteban-Echarri O, Rice S, et al. Artificial intelligence-assisted online social therapy for youth mental health. Front Psychol. 2017;8(ArtID 796).
  9. Breso A, Martinez-Miranda J, Botella C, Banos R, Garcia-Gomez J. Usability and acceptability assessment of an empathic virtual agent to prevent major depression. Expert Systems: International Journal of Knowledge Engineering and Neural Networks. 2016;33(4):297-312.
  10. Schroeder J. Pocket Skills: A Conversational Mobile Web App to Support Dialectical Behavioral Therapy. 2018; Accessed Combinatorial Chemistry & High Throughput Screening2018 Aug 16.
  11. Canadian Internet Registration Authority. The state of Canada's internet. Ottawa (ON): Canadian Internet Registration Authority (CIRA); 2017: Accessed 2018 Aug 16.
  12. Government of Canada. Notice: Health Canada’s Approach to Digital Health Technologies. 2018; Accessed 2018 Aug 16.
  13. Wysa. London (GB): Wysa Ltd.; 2018: Accessed 2018 Aug 16.
  14. Tess. San Francisco: X2AI Inc.; 2018: Accessed 2018 Aug 16.
  15. BC Association of Clinical Counsellors. Fee schedule. 2016; Accessed 2018 Aug 16.
  16. Psychologists’ Association of Alberta. Recommended Fee Schedule - Fee Schedule as of January 1, 2018. 2018;, 2018 Aug 16.
  17. Parikh SV, Quilty LC, Ravitz P, et al. Canadian Network for Mood and Anxiety Treatments (CANMAT) 2016 Clinical Guidelines for the Management of Adults with Major Depressive Disorder: Section 2. Psychological Treatments. Can J Psychiatry. 2016;61(9):524-539.
  18. Katzman MA, Bleau P, Blier P, et al. Canadian clinical practice guidelines for the management of anxiety, posttraumatic stress and obsessive-compulsive disorders. BMC Psychiatry. 2014;14 Suppl 1:S1.
  19. Ly K, Ly A-M, Andersson G. A fully automated conversational agent for promoting mental well-being: a pilot RCT using mixed methods. Internet interventions. 2017;10:39-46.
  20. Bird T, Mansell W, Wright J, Gaffney H, Tai S. Manage Your Life Online: A Web-Based Randomized Controlled Trial Evaluating the Effectiveness of a Problem-Solving Intervention in a Student Sample. Behav Cogn Psychother. 2018:1-13.
  21. Harwich EL, K. Thinking on its own: AI in the NHS. 2018;
  22. Ferreri F, Bourla A, Mouchabac S, Karila L. e-Addictology: An Overview of New Technologies for Assessing and Intervening in Addictive Behaviors. Front Psychiatr. 2018;9:51.
  23. Bohannon J. The synthetic therapist: Some people prefer to bare their souls to computers rather than to fellow humans. Science. 2015;349(6245):250-251.
  24. Gardiner P, McCue K, Negash L, et al. Engaging women with an embodied conversational agent to deliver mindfulness and lifestyle recommendations: a feasibility randomized control trial. Patient Educ Couns. 2017 Dec 6.
  25. Albright G, Adam C, Goldman R, Serri D. A game-based simulation utilizing virtual humans to train physicians to screen and manage the care of patients with mental health disorders. Games for Health. 2013;2(5):269-273.
  26. Replika. Luka Inc.; 2018: Accessed 2018 Aug 16.
  27. Mesko B. How Chatbots Can Ease Primary Care Burden. 2018;
  28. Rizzo A, Scherer S, Devault D, et al. Detection and computational analysis of psychological signals using a virtual human interviewing agent. J Pain Manag. 2016;9(3):311-322.
  29. Scholten M. Self-Guided Web-Based Interventions: Scoping Review on User Needs and the Potential of Embodied Conversational Agents to Address Them. J Med Internet Res. 2017.
  30. Hoermann S, McCabe KL, Milne DN, Calvo RA. Application of synchronous text-based dialogue systems in mental health interventions: Systematic review. J Med Internet Res. 2017;19(8):91-100.
  31. Provoost S, Lau HM, Ruwaard J, Riper H. Embodied Conversational Agents in Clinical Psychology: A Scoping Review. J Med Internet Res. 2017;19(5):e151.


Using Artificial Intelligence for Stroke in the Emergency Setting

e-ASPECTS is an AI-enhanced decision support tool that measures the extent of ischemic damage in patients with suspected stroke.1 It is not intended to replace the expert assessment of an image but rather to assist physicians in treatment decisions by providing an unbiased and standardized approach to image interpretation.2

How It Works

e-ASPECTS (Brainomix Limited) uses machine learning — a form of AI — to facilitate the automated extraction and classification of imaging features taken with non-contrast computed tomography (CT) in the emergency setting. The machine learning algorithm uses the CT imaging data to quantify the volume of ischemia (inadequate blood supply) and apply the Alberta Stroke Program Early CT Score (ASPECTS).3

ASPECTS is a quantitative, 10-point, validated scoring tool that measures the extent of early ischemic changes — changes in blood flow to the brain — and provides an accurate prediction of functional outcomes after thrombolytic treatment — the breakdown of blood clots formed in blood vessels.4 It is used with non-contrast CT as part of the assessment to determine patient eligibility for mechanical thrombectomy.5,6 ASPECTS was developed to help interpret CT images taken within the first hours of the onset of suspected stroke. The interpretation of CT images is generally challenging,7 requires considerable expertise, and can be subject to inter-rater variability.8 While ASPECTS scores can also be affected by differences in the reader’s experience and clinical background,6 overall it is regarded as a useful tool for the standardized evaluation of the extent of ischemic damage.5

e-ASPECTS was developed to further standardize these variables and interpret data objectively.9 The e-ASPECTS application generates a heat map to aid clinicians in interpreting its output. This heat map provides information on the mechanism of arriving at the final score (Olivier Joly, Brainomix, Oxford, UK: personal communication, 2018 Aug 16). e-ASPECTS results can be accessed via picture archiving and communication systems (PACS), through a Web browser user interface, or sent via email to a smartphone.10

The software is intended to assist clinical experts in decision-making by providing a second opinion and confirming expert assessment. In addition to reviewing the AI assessment, a physician is required to assess each CT image to rule out hemorrhage and other pathologies.2

It has been shown that the integration of e-ASPECTS into mobile stroke units can help with triage decisions related to selecting the appropriate hospital to send stroke patients to, such as those with a comprehensive stroke unit or a primary stroke unit.3 e-ASPECTS may also play a role in aiding decision-making for patient selection conducted via telemedicine and in selecting patients for transfer to stroke centres that perform mechanical thrombectomy.11

Who Might Benefit?

Approximately 62,000 Canadians experience stroke each year, and approximately 13,000 die after having a stroke.12 It is estimated that 405,000 Canadians are living with the effects of stroke12 and this number is expected to increase to between 654,000 and 726,000 over the next 20 years.13

People over the age of 70 are the most likely to experience a stroke.14 However, according to a 2014 report, over the preceding decade, the number of strokes in people in their 50s and 60s increased by 24% and 13%, respectively.14 As well, stroke rates in younger people (between the ages of 24 and 64) are expected to double by 2030.14 Early treatment with thrombolytic drugs reduces the mortality and the morbidity of stroke, and more patients who receive early treatment are discharged home, rather than to a rehabilitation centre.15,16

Availability in Canada

e-ASPECTS is not currently approved in Canada. It was granted the Conformité Européenne (CE) certification as a Class IIa medical device in 2015 by the European Union and is used in Europe and Brazil.17

Beyond Europe and Brazil, e-ASPECTS is installed for research purposes in Canada (Olivier Joly: personal communication, 2018 Jul 2).

What Does It Cost?

Brainomix sells e-ASPECTS as an annual licence subscription that allows the hospitals to process the scans of all stroke patients who are admitted (Olivier Joly: personal communication, 2018 Jul 2).

The cost of the device is not known and may vary depending on the setting in which it is used. In facilities with rapid network connections that do not require the installation of hardware, one installation can be accessed by several hospitals. In hospitals with less modern technological infrastructure, remote installation may not be feasible and the cost of the device may subsequently be higher.17

Current Practice

The 2015 Canadian Stroke Best Practice Recommendations note that brain imaging with non-contrast CT should be completed without delay for any patient with suspected stroke. To determine the eligibility for endovascular therapy, it is suggested that the initial brain CT should be assessed using ASPECTS to identify patients with a score of six points or higher. Patients may be eligible for endovascular therapy within six to 12 hours of onset of symptoms and should ideally begin treatment within 60 minutes of CT imaging.18

The American Heart Association guidelines also recognizes ASPECTS as a key tool for the management of acute stroke and suggests mechanical thrombectomy for patients with a baseline ASPECTS score of 6 or more.6

Published Studies

Four clinical utility studies on e-ASPECTS have been published, including one randomized controlled trial,1 one prospective cohort study,11 and two retrospective cohort studies.2,19 Three of the studies compare e-ASPECTS with either expertderived clinical scores using ASPECTS2,19 or expert opinion alone11 to predict functional outcomes. Two of these studies examine the correlation of e-ASPECTS scores with clinical outcomes after thrombectomy11 or thrombolysis.1 As well, a feasibility study was published that examines the clinical integration and utility of e-ASPECTS into a mobile stroke unit3 (an ambulance equipped with portable imaging equipment).

Issues to Consider

A concern for any AI application used in health care is that the data informing algorithms is applicable to the population that the AI tool will be used in. This underscores the importance of using AI to augment, rather than replace, a physician’s perspective. No information was found on how e-ASPECTS’ predictions are made or on the demographic used to train the algorithm.

It is suggested that, to be consistent with evidence-based practices, issues concerning transparency should be addressed prior to the integration of machine learning tools into clinical practice. Transparency regarding how predictions are made is lacking because the technical logic and mechanisms can be difficult to understand. This is known as the black-box paradox.20 Consideration should be given to the appropriate use of AI in reading and interpreting medical images. This may include establishing standards for AI interoperability, testing algorithms, and addressing regulatory, legal, and ethical issues.21

According to the manufacturer, the e-ASPECTS program was developed and tested using a largely Caucasian, adult population. While ethnic differences in brain anatomy and CT-based ischemic changes are reported to be minor,22 the generalizability of the e-ASPECTS platform to other ethnic groups could be further validated (Olivier Joly: personal communication, 2018 Jul 2).

e-ASPECTS can only be used in facilities that have access to CT. Canada has approximately 561 CT units23 but 89% of rural hospitals do not have access,24 indicating potential inequity in access to stroke care in urban and rural settings.

Related Developments

In addition to e-ASPECTS, Brainomix has developed a platform called e-CTA25 to analyze CT angiography of stroke patients and has partnered with Olea Medical to introduce the e-STROKE SUITE26 in Europe and Brazil. The e-STROKE SUITE utilizes information from other acute imaging modalities for stroke, including perfusion imaging. (Olivier Joly: personal communication, 2018 Jul 2).

Numerous AI-enhanced support systems have been developed to detect the presence of stroke from brain images. In 2018, the FDA approved a similar decision support software, developed by, that incorporates AI and analyzes CT images of the brain.27 The software sends a text notification to a patient’s specialist to alert them if a suspected large vessel blockage has been identified.27 There are a number of other AI-enhanced tools for stroke that use machine learning to automate segmentation to measure the volume of ischemic damage in brain CT to predict the outcome of stroke.28,29

Looking Ahead

Since “time is brain”30 and e-ASPECTS is anticipated to speed up the diagnosis of acute ischemic stroke — allowing earlier thrombolytic treatment — it may change how quickly stroke is diagnosed and managed. This software may be particularly useful to medical staff with limited experience in stroke imaging, such as family physicians and paramedics.20

The window of time in which it is optimal to treat stroke patients with mechanical thrombectomy has recently expanded from six hours up to 24 hours in select patients.31 e-ASPECTS may play a role in helping to quickly identify patients for mechanical thrombectomy within the expanded time frame.

Author: Andra Morrison


  1. Nagel S, Wang X, Carcel C, et al. Clinical Utility of Electronic Alberta Stroke Program Early Computed Tomography Score Software in the ENCHANTED Trial Database. Stroke. 2018;49(6):1407-1411.
  2. Herweh C, Ringleb PA, Rauch G, et al. Performance of e-ASPECTS software in comparison to that of stroke physicians on assessing CT scans of acute ischemic stroke patients. International Journal of Stroke. 2016;11(4):438-445.
  3. Grunwald IQ, Ragoschke-Schumm A, Kettner M, et al. First Automated Stroke Imaging Evaluation via Electronic Alberta Stroke Program Early CT Score in a Mobile Stroke Unit. Cerebrovasc Dis. 2016;42(5-6):332-338.
  4. Barber PA, Demchuk AM, Zhang J, Buchan AM. Validity and reliability of a quantitative computed tomography score in predicting outcome of hyperacute stroke before thrombolytic therapy. ASPECTS Study Group. Alberta Stroke Programme Early CT Score. Lancet. 2000;355(9216):1670-1674.
  5. Schroder J, Thomalla G. A Critical Review of Alberta Stroke Program Early CT Score for Evaluation of Acute Stroke Imaging. Front Neurol. 2016;7:245.
  6. Mokin M, Primiani CT, Siddiqui AH, Turk AS. ASPECTS (Alberta Stroke Program Early CT Score) measurement using Hounsfield unit values when selecting patients for stroke thrombectomy. Stroke. 2017 Jun;48(6):1574-1579.
  7. Patel SC, Levine SR, Tilley BC, et al. Lack of clinical significance of early ischemic changes on computed tomography in acute stroke. JAMA. 2001;286(22):2830-2838.
  8. Grotta JC, Chiu D, Lu M, et al. Agreement and variability in the interpretation of early CT changes in stroke patients qualifying for intravenous rtPA therapy. Stroke. 1999;30(8):1528-1533.
  9. Hampton-Till J, Harrison M, Kühn AL, et al. Automated quantification of stroke damage on brain computed tomography scans: E-Aspects. EMJ Neurol. 2015;3(1):69-74. Accessed 2018 Aug 22.
  10. e-ASPECTS. Oxford (GB): Brainomix; 2018: Accessed 2018 Aug 22.
  11. Pfaff J, Herweh C, Schieber S, et al. e-ASPECTS Correlates with and Is Predictive of Outcome after Mechanical Thrombectomy. AJNR Am J Neuroradiol. 2017;38(8):1594-1599.
  12. Stroke report 2016 just released. Ottawa: Heart & Stroke Foundation; 2016: Accessed 2018 Aug 22.
  13. What is the prevalence of stroke in Canada? 2015 Jul 23; Accessed 2018 Aug 22.
  14. 2014 stroke report. Together against a rising tide: advancing stroke systems of care. Ottawa: Heart & Stroke Foundation; 2014: Accessed 2018 Aug 22.
  15. Saver JL, Fonarow GC, Smith EE, et al. Time to treatment with intravenous tissue plasminogen activator and outcome from acute ischemic stroke. JAMA. 2013;309(23):2480-2488.
  16. Mobile stroke units for prehospital care of ischemic stroke. Ottawa: CADTH; 2017 Jun: Accessed 2018 Aug 22.
  17. e-ASPECTS: accelerating the stroke pathway. 2017; Accessed 2018 Aug 22.
  18. Casaubon LK, Boulanger JM, Blacquiere D, et al. Canadian Stroke Best Practice Recommendations: Hyperacute Stroke Care Guidelines, Update 2015. Int J Stroke. 2015;10(6):924-940.
  19. Nagel S, Sinha D, Day D, et al. e-ASPECTS software is non-inferior to neuroradiologists in applying the ASPECTS score to CT scans of acute ischemic stroke patients [abstract]. International Journal of Computer Assisted Radiology and Surgery - 30th International Congress and Exhibition, CARS 2016. 2016;11(1 Suppl 1):S122-S123.
  20. Lee EJ, Kim YH, Kim N, Kang DW. Deep into the brain: artificial intelligence in stroke imaging. J Stroke. 2017 Sep;19(3):277-285. Accessed 2018 Aug 22.
  21. The Royal College of Radiologists. Standards for interpretation and reporting of imaging investigations. 2018 Mar; Accessed 2018 Aug 22.
  22. Zilles K, Kawashima R, Dabringhaus A, Fukuda H, Schormann T. Hemispheric shape of European and Japanese brains: 3-D MRI analysis of intersubject variability, ethnical, and gender differences. Neuroimage. 2001;13(2):262-271.
  23. CADTH. Canadian Medical Imaging Inventory, 2017. Ottawa: CADTH; 2018: Accessed 2018 Aug 22.
  24. Fleet R, Bussières S, Tounkara FK, et al. Rural versus urban academic hospital mortality following stroke in Canada. PLOS ONE. 2018;13(1):e0191151.
  25. e-CTA. Oxford (GB): Brainomix; 2018: Accessed 2018 Aug 22.
  26. e-stroke suite. Oxford (GB): Brainomix; 2018: Accessed 2018 Aug 22.
  27. FDA permits marketing of clinical decision support software for alerting providers of a potential stroke in patients. 2018 Feb 13; Accessed 2018 Aug 22.
  28. Colak C, Karaman E, Turtay MG. Application of knowledge discovery process on the prediction of stroke. Comput Methods Programs Biomed. 2015;119(3):181-185.
  29. Beecy AN, Chang Q, Anchouche K, et al. A Novel Deep Learning Approach for Automated Diagnosis of Acute Ischemic Infarction on Computed Tomography. JACC Cardiovasc Imaging. 2018 May 11.
  30. Saver JL. Time is brain--quantified. Stroke. 2006;37(1):263-266.
  31. American Heart Association. More stroke patients may receive crucial treatments under new guideline. 2018 Jan 24; Accessed 2018 Aug 22.


Detection of Cognitive Impairment and Dementia With Artificial Intelligence

Current diagnostic criteria for dementia have limited predictive ability for individuals.1,2 It is proposed that artificial intelligence (AI) technologies may improve screening for cognitive impairment and dementia.3,4 Machine-learning algorithms are being used to detect patterns of cognitive impairment by quantitatively analyzing medical data including imaging; demographic and genetic data; and speech, language, and cognitive neuropsychological test results.3,5,6

How It Works

BrainFx is an assessment tool that measures neurofunctional performance in individuals with mild to moderate brain dysfunction, and uses AI technology to support early detection and identification of risk factors for cognitive impairment and dementia.12

Dementia is a broad clinical term that includes vascular dementia, Parkinson disease dementia, and, most commonly, Alzheimer disease.7,8 Patients with dementia have damaged nerve cells in the brain, which impair memory, speech and language, and cognitive functions, and can eventually affect daily activities, physical mobility, and safety.9,10 Early detection during the pre-dementia stage known as mild cognitive impairment supports the timely and effective use of treatments to slow dementia progression.6,11

BrainFx SCREEN and 360 Assessment Tools

BrainFx (BrainFx, Pickering, Ontario) assesses complex cognitive skills and other neurofunctions using predictive analytics through interactive, tabletbased activities administered by trained clinicians.12 BrainFx does not provide a diagnosis or treatment recommendations — the information produced must be assessed by qualified clinicians.12

Individuals fill out a self-report prior to completing the BrainFx Assessments.12 The initial BrainFx SCREEN tool is a 10-minute screening assessment of seven cognitive skills and the BrainFx 360 Assessment tool is a 90- to 120-minute assessment that measures 30 cognitive skills including functional impacts.12

BrainFx uses a machine-learning algorithm to compare an individual patient’s performance data to anonymized patientconsented data from the BrainFx Living Brain Bank, and clinical neuroscience and demographic data to detect early cognitive impairment and dementia, and improve assessments and comparative data12- 14 (Tracy Milner, BrainFx, Pickering, ON: personal communication, 2018 Jul 13). The BrainFx SCREEN and 360 assessment performance results are produced as a report and integrated with a patient’s medical history and quality of life information to produce a secondary report.12,15 The patient’s health care team receives the reports, including the self-report, for use as a supplementary tool in the diagnostic and treatment-planning process.12,14,15

Using BrainFx to Support Proactive Disease Detection

A separate proactive detection tool is under development by BrainFx in partnership with Saint Elizabeth Healthcare, ThoughtWire Corp, (Wilfred) Laurier University, four family health teams in the Waterloo Wellington Local Health Integration Network, and the Southlake Regional Health Centre.13,14,16 It is intended as a brain health assessment and risk management tool for the early detection of cognitive impairment and dementia.13,14,16 BrainFx Assessments are being used to screen 10,000 patients to detect cognitive decline.16,17 This data will be combined with Canadian Institute for Health Information data to train machine-learning algorithms to identify early risk factors.16,17 The intention is for health care professionals to use the early identification algorithm under study to proactively scan electronic medical records to identify at-risk individuals for assessment.13,14,18,19

Who Might Benefit?

In Canada, more than 564,000 individuals live with dementia, and this is expected to increase to 937,000 by 2031.20 An estimated 30% of patients with dementia in Canada may never receive diagnostic assessment, and the identification approaches used may fail to detect 10% of the patients who are assessed.21

Risk factors for dementia include age, family history, unhealthy diet, alcohol, smoking, high blood pressure and cholesterol, diabetes, physical inactivity, severe brain injury, depression, and socioeconomic factors.21-24

Availability in Canada

BrainFx Assessments received a Class I Medical Device Establishment License from Health Canada in 201312,25 (Tracy Milner: personal communication, 2018 Jun 28). The assessments are used by more than 700 clinicians and approximately 120 health care organizations throughout Canada (Tracy Milner: personal communication, 2018 Jun 28). No information was found on regulatory approval of the BrainFx AI integrated application of the Living Brain Bank or for its use in proactive disease detection.

What Does It Cost?

The Canadian cost of BrainFx SCREEN and 360 Assessments is about $125 per month per individual health care provider, for use with an unlimited number of patients (Tracy Milner: personal communication, 2018 Jun 28). Discounted pricing is available for provider groups, including hospitals and clinics, and for individual assessments (Tracy Milner: personal communication, 2018 Jun 28). No information on the cost of the BrainFX AI integrated application using Living Brain Bank data or its use in proactive disease detection was obtained. As well, no studies on cost implications or considerations related to BrainFx technology were identified.

Current Practice

Dementia is typically detected by clinicians through manual neuropsychological assessments, the patient’s medical history, and observations by family and health care workers.11,26 An extensive series of cognitive screening tools including the Montreal Cognitive Assessment and the Mini-Mental State Examination are often used for assessment.9 Imaging methods including magnetic resonance imaging (MRI) and positron emission tomography (PET) are also used to measure and detect early changes in brain features.11 Currently, post-mortem analysis of brain tissue is considered to be the only method for obtaining a definite diagnosis because accurate diagnosis during a patient’s life is challenging, as the underlying pathologies of different neurodegenerative diseases can result in overlapping clinical symptoms.11,27

Summary of Evidence

Few studies have used machine-learning algorithms with behavioural, functional, and cognitive data to detect dementia.4 The BrainFx website lists several completed studies but no published study results in individuals with dementia were identified.19


Dementia testing can potentially impact patient care because of test inaccuracies and variations in the interpretation of results by clinicians, leading to imprecise detection.28,29 No information was found on the number of incorrect test results with the BrainFx Assessments.

Issues to Consider

Data Collection

AI technologies for dementia and cognitive impairment detection can be introduced into standard medical protocols, and data can be retrieved quickly and easily to complement other diagnostic tools used by clinicians.30 Additionally, clinicians do not need to know AI algorithm-programming details to use the technology.2

However, a large and diverse data set is needed to train machine-learning algorithms.31 Multi-centre health care data that is currently available may not be usable due to unclear labelling for diagnoses, inadequate disease specifications, and differences in data collection protocols, which prevents the merging of multiple data sets.31 Multi-modal data creation can also be challenging if necessary data types are missing because of clinical trial costs, high dropout rates, imaging equipment availability, or a lack of patient consent.3,6

There is also uncertainty regarding how algorithms trained for specific tasks perform their analysis, and what the values in the algorithms represent.32 This “black box” problem makes understanding the mechanism by which an AI system arrived at decisions difficult.32 Legal concerns that have been raised include medical malpractice concerns for clinicians who do not use AI systems to support diagnosis.33,34

Cost Considerations

Canadian out-of-pocket and public health system costs for the care of individuals with dementia in 2016 were estimated at $10.4 billion; this is expected to increase to $16.6 billion by 2031.20,30 In the UK, the cost for each dementia diagnosis is an estimated $6,000 at minimum, not including indirect costs to patients.33,35 Earlier detection of dementia could reduce costs by enabling earlier intervention.20 One UK, industry-sponsored, cost-utility study reported that higher up-front and treatment-related costs were offset by downstream savings in patient care — related to more timely treatment, resulting in less severe disease and time spent in institutions — for patients with suspected mild cognitive impairment.35

Related Developments

Automated analysis of speech samples can be used as a tool for the early detection of dementia.30 WinterLight Labs (Toronto, Ontario) has created a tablet-based assessment system, which can characterize and quantify language and speech patterns to detect and monitor dementia and other cognitive disorders.36,37

Several AI-based approaches incorporating imaging data to support the early detection and monitoring of Alzheimer disease progression are in development. Avalon Ai (London, UK) has developed software that uses machine learning to detect signs of brain degeneration and dementia disease progression based on MRI scans.37-39 In the Netherlands, researchers are combining machine-learning algorithms with an MRI technique called arterial spin labelling imaging.40 The program recognizes patterns on perfusion maps — images showing the amount of blood delivered to various brain areas — to distinguish patients with varying cognitive impairments, and to predict the stage of Alzheimer disease in newly diagnosed individuals.40 Scientists at McGill University have developed an algorithm capable of recognizing signs of dementia two years prior to onset using one PET scan of the brain of individuals at-risk for Alzheimer disease.41

Aequa Sciences (London, UK) is using machine learning and artificial neural networks to predict Alzheimer disease onset based on genetic data from healthy individuals and those with the disease.38

Looking Ahead

Research using machine-learning algorithms for the prognosis and detection of dementia is progressing quickly. There is interest in incorporating additional model inputs such as socioeconomic and lifestyle factors, behavioural data, and consideration of patient comorbidities in the algorithm assessment models, and in using deep-learning algorithms for the early detection of dementia.4,6,42

Author: Humaira Nakhuda


  1. Lins AJCC, Muniz MTC, Garcia ANM, Gomes AV, Cabral RM, Bastos-Filho CJA. Using artificial neural networks to select the parameters for the prognostic of mild cognitive impairment and dementia in elderly individuals. Comput Methods Programs Biomed. 2017;152:93-104.
  2. Bryan RN. Machine learning applied to Alzheimer disease. Radiology. 2016;281(3):665-668.
  3. Li R. Data mining and machine learning methods for dementia research. Methods Mol Biol. 2018;1750:363-370.
  4. Battista P, Salvatore C, Castiglioni I. Optimizing neuropsychological assessments for cognitive, behavioral, and functional impairment classification: A machine learning study. Behav Neurol. 2017(1850909).
  5. Liu F, Wee CY, Chen H, Shen D. Inter-modality relationship constrained multi-task feature selection for AD/MCI classification. Medical image computing and computer-assisted intervention : MICCAI 2013;International Conference on Medical Image Computing and Computer-Assisted Intervention. 16(Pt 1):308-315.
  6. Liu X, Chen K, Wu T, Weidman D, Lure F, Li J. Use of multimodality imaging and artificial intelligence for diagnosis and prognosis of early stages of Alzheimer's disease. Translational Research. 2018;194:56-67.
  7. Gates NJK, S.; Rutjes, A.W.; Ware, J.; March, E.; Vernooij, R. W. Computerised cognition-based interventions for preventing dementia in people with mild cognitive impairment [protocol]. Cochrane Database Syst Rev. 2016.
  8. About dementia: what is dementia? 2018; Accessed 2018 Jun 7.
  9. Orimaye SO, Wong JSM, Golden KJ, Wong CP, Soyiri IN. Predicting probable Alzheimer's disease using linguistic deficits and biomarkers. BMC bioinformatics. 2017;18(1):34.
  10. Al-Naami B, Abu Mallouh M, Kheshman AA. Automated intelligent diagnostic of alzheimer disease based on neuro-fuzzy system and discrete wavelet transform. Biomed Eng Appl Basis Commun. 2014;26(3):1450035.
  11. Nanni LZ, N.; Salvatore, C.; Castiglioni,I.; Alzheimer's Disease Neuroimaging Initiative. An ensemble of classifiers for the early diagnosis of Alzheimer's disease. Padua (IT): University of Padua, Department of Information Engineering; 2018: Accessed 2018 Aug 16.
  12. BrainFx. Pickering: BrainFx; 2017:
  13. Laurier researchers to evaluate the impact of using artificial intelligence to detect earliest stages of Alzheimer's. Waterloo (ON): Wilfrid Laurier University; 2018: Accessed 2018 Jan 1.
  14. Gallo M. Harnessing the power of machine learning for brain health. 2018;
  15. Searles CM. Test-retest reliability of the BrainFX 360 performance assessment [Master's thesis]. Murfreesboro (TN): Middle Tennessee State University; 2015 Aug: Accessed 2018 Aug 16.
  16. Abe E. Laurier professors produce early Alzheimer’s detector. 2018;
  17. Weidner J. Artificial intelligence being used to detect Alzheimer’s early. 2018;
  18. Ontario improving patient care through innovative health technologies. Toronto (ON): Ontario Ministry of Health and Long-Term Care; 2017: Accessed 2018 Aug 16.
  19. BrainFx. BrainFx in Research Summary. 2017;
  20. Alzheimer Society of Canada, Public Health Agency of Canada. Prevalence and Monetary Costs of Dementia in Canada. Toronto (ON): Alzheimer Society of Canada; 2016: Accessed 2018 May 15.
  21. Manuel DG, Garner R, Fines PB, C., Flanagan W, Tu K. Alzheimer’s and other dementias in Canada, 2011 to 2031: a microsimulation Population Health Modeling (POHEM) study of projected prevalence, health burden, health services, and caregiving use. Popul Health Metr. 2016;14(37).
  22. Dementia. Ottawa: Government of Canada; 2018: Accessed 2018 May 15.
  23. Alzheimer Society of Canada. Risk Factors. 2018; Accessed 2018 May 15.
  24. Lang L, Clifford A, Wei L, et al. Prevalence and determinants of undetected dementia in the community: a systematic literature review and a meta-analysis. BMJ Open. 2017 Feb;7(2):e011146.
  25. Medical devices establishment license listing. Ottawa: Health Canada; 2018: Accessed 2018 Jun 21.
  26. Mullin E. AI can spot signs of Alzheimer’s before your family does. 2018; Accessed 2018 Aug 16.
  27. Harper L, Fumagalli GG, Barkhof F, et al. MRI visual rating scales in the diagnosis of dementia: Evaluation in 184 post-mortem confirmed cases. Brain. 2016;139(4):1211-1225.
  28. Skinner TR, Scott IA, Martin JH. Diagnostic errors in older patients: a systematic review of incidence and potential causes in seven prevalent diseases. International journal of general medicine. 2016;9:137-146.
  29. So A, Hooshyar, D., Park, K.W., and Lim, H.K. Early Diagnosis of Dementia from Clinical Data by Machine Learning Techniques. Applied Sciences. 2017.
  30. Lopez-De-Ipina K, Martinez-De-Lizarduy U, Calvo PM, et al. Advances on automatic speech analysis for early detection of alzheimer disease: A non-linear multi-task approach. Curr Alzheimer Res. 2018;15(2):139-148.
  31. Mandal PK, Shukla D. Brain Metabolic, Structural, and Behavioral Pattern Learning for Early Predictive Diagnosis of Alzheimer's Disease. Journal of Alzheimer's disease : JAD. 2018.
  32. Szymkowiak T. The Artificial Intelligence Black Box Problem & Ethics. 2017; Accessed 2018 Jun 22.
  33. Rudzicz F. Toward Dementia Diagnosis via Artificial Intelligence. 2016; Accessed 2018 May 22.
  34. Morrison A, Mason J, Visintini S. An Overview of Clinical Applications of Artificial Intelligence [forthcoming]. Ottawa (ON): CADTH; 2018: Accessed 2018 Aug 16.
  35. Getsios D, Blume S, Ishak KJ, Maclaine G, Hernandez L. An economic evaluation of early assessment for Alzheimer's disease in the United Kingdom. Alzheimer's & dementia : the journal of the Alzheimer's Association. 2012;8(1):22-30.
  36. WinterLight Labs. About WinterLight. 2018; Accessed 2018 May 18.
  37. Sennaar K. Artificial Intelligence for Dementia Diagnosis – Genetic Analysis, Speech Analysis, and More. 2018; Accessed 2018 Aug 16.
  38. Artificial Intelligence For Early Alzheimer’s Detection. 2016 Dec 28; Accessed 2018 Aug 16.
  39. Eurostars. Artificial Intelligence helps early detection of dementia. 2018; Accessed 2018 Aug 16.
  40. RSNA. Artificial Intelligence May Aid in Alzheimer’s Diagnosis. 2016; Accessed 2018 May 22.
  41. Dupuis J. Artificial intelligence predicts dementia before onset of symptoms. 2017; Accessed 2018 May 22.
  42. Dallora AL, Eivazzadeh S, Mendes E, Berglund J, Anderberg P. Machine learning and microsimulation techniques on the prognosis of dementia: A systematic literature review. PLoS ONE. 2017;12(6):e0179804. Accessed 2018 Aug 16.


Focus On: Artificial Intelligence in Population and Public Health

In January 2018, CBC News reported that the Public Health Agency of Canada had partnered with an Ottawa-based market research firm to use artificial intelligence (AI) to analyze Canadians’ public Facebook posts to predict increases in suicide risk. This was in an effort to improve prevention and intervention.1 Since then, other media outlets have reported on how AI could use new and existing sources of data to open the door to novel approaches to population and public health.2-4

Why Population and Public Health?

AI systems rely on large amounts of data to learn and function.5,6 For that reason, health care specialties — such as population and public health (particularly disease surveillance and disease forecasting), which already use consistent and reliable sources of data — are areas of active research in the AI community. Existing data collection systems often rely on clinician reporting, or data that can be weeks or months old, which may impair the ability to identify outbreaks, for example.7,8 The ability of AI systems to consider additional and potentially more timely or complex sources of data such as electronic medical records (EMRs) and information from social media sites is of interest to address current limitations, and may help improve the ability to predict future disease from past data.2,3 A 2018 workshop jointly hosted by the Canadian Institutes of Health Research’s Institute of Population and Public Health and CIFAR identified potential opportunities for incorporating AI into public health practices such as supporting and promoting healthy behaviour and modelling policy decisions to understand their impact on public health.9

Types of AI Used

The variety of challenges encountered in population and public health lend themselves to a range of AI solutions. Scenarios where determining or predicting the presence or absence of an outcome is required (such as whether a virus can be transmitted from animals to humans) can be evaluated by support vector machine (SVM) programs.6,10-13 More complex problems, such as predicting disease outbreaks from multiple unlinked data sources with unclear relationships, may be better suited to using artificial neural networks,6,8,14-18 or deep learning.5

Natural language processing (NLP) and text mining — AI approaches that analyze the content of text-based data sources for patterns and relationships — are used by researchers to explore problems where data are unstructured or vary by source, as in EMRs and social media posts.7,19-22

Areas of Active Research

Population and public health researchers have explored using AI for a number of clinical areas including infectious diseases,5-8,10,11,15,17,18,21-27 food-borne illnesses,13,14,16 the opioid crisis,19,20 and transmission of viruses and bacteria from animals to humans.12,17

Applications identified within the clinical areas include using AI for disease forecasting and modelling,11,15,22,23,25 and using data from emerging sources such as social media posts8,10,19-21 or geographical information systems10,14,28,29 to improve existing surveillance and forecasting techniques.

Infectious Diseases

Influenza is a common respiratory illness affecting thousands of Canadians each year.30 Research into how AI could change the approach to influenza management includes:

  • using clinical notes in EMRs from four US health systems to develop, train, and test the ability of two NLP tools to detect influenza cases across organizations7
  • improving the monitoring of influenza outbreaks by combining location data and an SVM approach to better distinguish social media posts that describe a case of influenza from those that do not10
  • testing the ability of three artificial neural networks to predict regional influenza-like illness from social media posts compared to CDC-Centers for Disease Control and prevention reports over a 55-week period in the US8
  • using an artificial neural network to determine what avian influenza viruses could infect humans.17

Dengue fever is a mosquito-borne illness that affects tens of millions of people each year.31 The role of AI in predicting and monitoring dengue has been investigated.11,15,22,24 AI research has included: the prediction of cases of dengue in Thailand24 and China11 by combining population data, climate data, dengue case data, and (in the Thai study) mosquito infection data and performing SVM analysis; a comparison of four dengue outbreak prediction models, including one artificial neural network model, in Brazil;15 and the creation of a text mining tool to monitor media reports for dengue cases in India.22

The severity of the 2014-2015 Ebola outbreak in West Africa led researchers to explore how AI could help improve our understanding of the virus.6,21,23 Research includes efforts to develop a model to predict the prognosis for people infected with Ebola using a publicly available clinical data set and artificial neural network and SVM approaches.6 Researchers also explored public perception and concern in response to an Ebola case in Dallas using the text mining of social media posts.21 Preparing for future Ebola outbreaks is important, but it is challenging to predict epidemic situations, particularly in regions with no prior outbreaks. To address this challenge, one study used machine learning to predict the spread of, and response to, a simulated Ebola outbreak in Beijing.23

Zika virus has also drawn the attention of AI researchers.25,26 One initiative studied an approach that combined climate data (such as annual rainfall), environmental conditions (such as relative humidity), and socioeconomic factors (such as travel time to the nearest city), with global data on Zika virus infection to map the risk of virus transmission using three machine-learning models.25 To investigate a new way of screening for Zika virus infection, another study used a machine-learning approach to analyze mass spectrometry data from blood samples to identify biomarkers and patterns that can be used to predict whether a patient is infected with Zika virus.26

Other uses of AI in infectious diseases include using a machine-learning approach to determine how clinical and magnetic resonance imaging-related signs of hand-foot-and-mouth disease interact to help predict the risk of developing severe disease,27 developing intelligent building systems (for example, hot water systems) that could help prevent legionnaires’ disease,18 and combining mobile technology with deep learning analysis of chest X-rays to support the timely diagnosis of tuberculosis in remote and resource-poor communities.5

Food-Borne Illnesses

Food-borne illnesses caused by bacteria, viruses, and parasites cause an estimated 4 million Canadians to become sick each year, resulting in more than 11,000 hospitalizations and more than 200 deaths.32 Detecting and preventing outbreaks of food-borne illnesses such as salmonella, E. coli, norovirus, and Listeria, using AI, have been studied.13,14,16 Researchers have used an artificial neural network to explore relationships between salmonella and E. coli infection data, and socioeconomic data, from five US states;14 predicted oyster-associated norovirus outbreaks along the US Gulf Coast by combining historical outbreak data and environmental predictors of outbreaks such as ocean temperatures, rainfall, and offshore winds, and analyzing with an artificial neural network;16 and explored how SVM compares to expert opinion when determining the persistence of L. monocytogenes in a deli environment.13

Opioid Crisis

In 2017, there were more than 3,900 apparent opioid-related deaths in Canada.33 AI may help public health officials combat this crisis: researchers have explored text-mining social media posts to supplement formal surveys of non-medical use of prescription drugs to improve the understanding of their use20 and to help public health officials evaluate the public’s perception and reaction to the opioid crisis.19

Animal to Human Disease Transmission

Understanding which viruses and bacteria have the potential to move from animal hosts to human hosts is important for disease surveillance and preparing for outbreaks. By combining genomic information about viruses such as influenza,17 and bacteria such as E. coli O157,12 with historical outbreak data, researchers have investigated the ability of artificial neural networks and SVMs to distinguish between variants that have the potential to jump between species and those that do not.

Other Uses

In an effort to redefine existing climate regions and understand how climate affects the hospitalization of elderly people, researchers in the US used machine-learning approaches to analyze combined data from satellite images, meteorological data, spatial data, and Medicare subscriber data.28 At the University of Waterloo, a proof-of-concept study used AI to detect blue-green algae levels in drinking water.34

Issues to Consider

While not unique to population and public health applications of AI, the quality, quantity, availability, and interoperability, of data needed for AI systems to perform accurately should be considered.5,6 Creating generalizable machine-learning models requires data from as many patients in as many settings as possible.6 Missing data in existing data sets complicate the ability to create predictive models.6 Some diseases, like tuberculosis, may lack large, publicly available data sets.5 Developing such a database could require many steps and resources or be challenging because of the complex presentation of the disease.5

Data sets from emerging sources, such as EMRs, are often created to be tailored to a specific setting, making it challenging to compare unstructured information across organizations.7 NLP programs may also be challenged by regional or organizational variations in terminology.7

Incorporating new data sources (such as social media posts) may also raise questions about privacy and ethics.35 For example, genome sequences or geospatial data could inadvertently identify individuals, and additional care and thought may be needed to explain how their data will be used.35

Some AI programs, such as SVMs, may be prone to overfitting of data — where the information used to make distinctions between groups is random, or not biologically relevant, making generalizability beyond a set of training data difficult.12 Some machine-learning algorithms perform better than others for different problems, so it may be necessary to compare a number of approaches against each other to determine which one provides the best answer, given the data at hand.25

Additional challenges identified by participants at the joint Canadian Institutes of Health Research’s Institute of Population and Public Health and CIFAR workshop included a need to build crossdisciplinary relationships between AI and public health researchers, and educating and training public health researchers and practitioners in AI skills.9

Looking Ahead

AI applications in population and public health are still in their early stages, with many possible applications yet to be studied and potential yet to be realized.2,3 AI could also provide researchers with new ways of understanding how largely incompatible data, such as the social determinants of health, are connected and interrelated, broadening our understanding of how they affect the health of Canadians.3 It could also reduce the need for human intervention by automatically monitoring our environment for potential disease outbreaks.18 As the influence and uses of AI expands, a deeper knowledge of health informatics may prove an important skill for health professionals’ understanding of increasingly data-reliant health systems.36

Author: Jeff Mason


  1. Ruskin B. Feds to search social media using AI to find patterns of suicide-related behaviour. 2018; Accessed 2018 Aug 22.
  2. Berstein. Fighting the flu: We need a new kind of intelligence. 2018 Jan 22; Accessed 2018 Aug 22.
  3. Hoffman. The great promise of Artificial Intelligence for public health. 2018 Apr 5; Accessed 2018 AUg 22.
  4. Kennedy. How AI is helping to predict and prevent suicides. 2018; Accessed 2018 Aug 22.
  5. Alcantara MF, Cao Y, Liu C, et al. Improving tuberculosis diagnostics using deep learning and mobile health technologies among resource-poor communities in Peru. Smart Health. 2017;1-2:66-76.
  6. Colubri A, Silver T, Fradet T, Retzepi K, Fry B, Sabeti P. Transforming Clinical Data into Actionable Prognosis Models: Machine-Learning Framework and Field-Deployable App to Predict Outcome of Ebola Patients. PLoS Negl Trop Dis. 2016;10(3):e0004549.
  7. Ferraro JP, Ye Y, Gesteland PH, et al. The effects of natural language processing on cross-institutional portability of influenza case detection for disease surveillance. Appl Clin Inform. 2017;8(2):560-580.
  8. Hu H, Wang H, Wang F, Langley D, Avram A, Liu M. Prediction of influenza-like illness based on the improved artificial tree algorithm and artificial neural network. Sci Rep. 2018;8(1):4895.
  9. Canadian Institutes of Health Research. Application of Artificial Intelligence Approaches to Tackle Public Health Challenges – Workshop Report. 2018; Accessed 2018 Aug 22.
  10. Allen C, Tsou MH, Aslam A, Nagel A, Gawron JM. Applying GIS and Machine Learning Methods to Twitter Data for Multiscale Surveillance of Influenza. PLoS One. 2016;11(7):e0157734.
  11. Guo P, Liu T, Zhang Q, et al. Developing a dengue forecast model using machine learning: A case study in China. PLoS Negl Trop Dis. 2017;11(10):e0005973.
  12. Lupolova N, Dallman TJ, Matthews L, Bono JL, Gally DL. Support vector machine applied to predict the zoonotic potential of E. coli O157 cattle isolates. Proc Natl Acad Sci U S A. 2016;113(40):11312-11317.
  13. Vangay P, Steingrimsson J, Wiedmann M, Stasiewicz MJ. Classification of Listeria monocytogenes persistence in retail delicatessen environments using expert elicitation and machine learning. Risk Anal. 2014;34(10):1830-1845.
  14. Akil L, Ahmad HA. Salmonella infections modelling in Mississippi using neural network and geographical information system (GIS). BMJ Open. 2016;6(3):e009255.
  15. Baquero OS, Santana LMR, Chiaravalloti-Neto F. Dengue forecasting in Sao Paulo city with generalized additive models, artificial neural networks and seasonal autoregressive integrated moving average models. PLoS One. 2018;13(4):e0195065.
  16. Chenar SS, Deng Z. Development of artificial intelligence approach to forecasting oyster norovirus outbreaks along Gulf of Mexico coast. Environ Int. 2018;111:212-223.
  17. Chrysostomou C, Partaourides H, Seker H. Prediction of Influenza A virus infections in humans using an Artificial Neural Network learning approach. Conf Proc IEEE Eng Med Biol Soc. 2017;2017:1186-1189.
  18. Sincak P, Ondo J, Kaposztasova D, Vircikova M, Vranayova Z, Sabol J. Artificial intelligence in public health prevention of legionelosis in drinking water systems. Int J Environ Res Public Health. 2014;11(8):8597-8611.
  19. Glowacki EM, Glowacki JB, Wilcox GB. A text-mining analysis of the public's reactions to the opioid crisis. Subst Abus. 2017:1-5.
  20. Kalyanam J, Katsuki T, G RGL, Mackey TK. Exploring trends of nonmedical use of prescription drugs and polydrug abuse in the Twittersphere using unsupervised machine learning. Addict Behav. 2017;65:289-295.
  21. Lazard AJ, Scheinfeld E, Bernhardt JM, Wilcox GB, Suran M. Detecting themes of public concern: A text mining analysis of the Centers for Disease Control and Prevention's Ebola live Twitter chat. Am J Infect Control. 2015;43(10):1109-1111.
  22. Villanes A, Griffiths E, Rappa M, Healey CG. Dengue Fever Surveillance in India Using Text Mining in Public Media. Am J Trop Med Hyg. 2018;98(1):181-191.
  23. Zhang P, Chen B, Ma L, et al. The Large Scale Machine Learning in an Artificial Society: Prediction of the Ebola Outbreak in Beijing. Comput Intell Neurosci. 2015;2015:531650.
  24. Kesorn K, Ongruk P, Chompoosri J, et al. Morbidity Rate Prediction of Dengue Hemorrhagic Fever (DHF) Using the Support Vector Machine and the Aedes aegypti Infection Rate in Similar Climates and Geographical Areas. PLoS One. 2015;10(5):e0125049.
  25. Jiang D, Hao M, Ding F, Fu J, Li M. Mapping the transmission risk of Zika virus using machine learning models. Acta Trop. 2018;185:391-399.
  26. Melo CFOR, Navarro LC, de Oliveira DN, et al. A Machine Learning Application Based in Random Forest for Integrating Mass Spectrometry-Based Metabolomic Data: A Simple Screening Method for Patients With Zika Virus. Front Bioeng Biotechnol. 2018;6(31).
  27. Zhang B, Wan X, Ouyang FS, et al. Machine Learning Algorithms for Risk Prediction of Severe Hand-Foot-Mouth Disease in Children. Sci Rep. 2017;7(1):5368.
  28. Liss A, Koch M, Naumova EN. Redefining climate regions in the United States of America using satellite remote sensing and machine learning for public health applications. Geospat Health. 2014;8(3):S647-659.
  29. VoPham T, Hart JE, Laden F, Chiang YY. Emerging trends in geospatial artificial intelligence (geoAI): potential applications for environmental epidemiology. Environ Health. 2018;17(1):40.
  30. Infection Prevention and Control Canada. Seasonal Influenza, Avian Influenza and Pandemic Influenza: Current Seasonal Influenza Update. 2018; Accessed 2018 Aug 22.
  31. World Health Organization. Dengue and severe dengue. 2018; Accessed 2018 Aug 22.
  32. Government of Canada. Causes of food-borne illness in Canada. 2015;, 2018 Aug 22.
  33. Special Advisory Committee on the Epidemic of Opioid Overdoses. National report: Apparent opioid-related deaths in Canada (January 2016 to December 2017). Ottawa: Public Health Agency of Canada,; 2018: Accessed 2018 Aug 22.
  34. Jin C, Mesquita MMF, Deglint JL, Emelko MB, Wong A. Quantification of cyanobacterial cells via a novel imaging-driven technique with an integrated fluorescence signature. Sci Rep. 2018;8(1):9055.
  35. Mooney SJ, Pejaver V. Big Data in Public Health: Terminology, Machine Learning, and Privacy. Annu Rev Public Health. 2018;39:95-112.
  36. Fridsma DB. Health informatics: a required skill for 21st century clinicians. BMJ. 2018;362.


Mini-Roundup: Recent Reports From CADTH and Other Agencies


Health Technology Update newsletter

Issues in Emerging Health Technologies bulletins

Recent Reports on Artificial Intelligence Technologies From Other Agencies

Agencies Included in the Mini-Roundup below:

Cancer, Imaging, and Radiology

Home Care

Trends, Forecasts, and Strategic Initiatives

Canadian Initiatives on Artificial Intelligence


About This Document

Disclaimer: The information in this document is intended to help Canadian health care decision-makers, health care professionals, health systems leaders, and policymakers make well-informed decisions and thereby improve the quality of health care services. While patients and others may access this document, the document is made available for informational purposes only and no representations or warranties are made with respect to its fitness for any particular purpose. The information in this document should not be used as a substitute for professional medical advice or as a substitute for the application of clinical judgment in respect of the care of a particular patient or other professional judgment in any decision-making process. The Canadian Agency for Drugs and Technologies in Health (CADTH) does not endorse any information, drugs, therapies, treatments, products, processes, or services.

While care has been taken to ensure that the information prepared by CADTH in this document is accurate, complete, and up-to-date as at the applicable date the material was first published by CADTH, CADTH does not make any guarantees to that effect. CADTH does not guarantee and is not responsible for the quality, currency, propriety, accuracy, or reasonableness of any statements, information, or conclusions contained in any third-party materials used in preparing this document. The views and opinions of third parties published in this document do not necessarily state or reflect those of CADTH.

CADTH is not responsible for any errors, omissions, injury, loss, or damage arising from or relating to the use (or misuse) of any information, statements, or conclusions contained in or implied by the contents of this document or any of the source materials.

This document may contain links to third-party websites. CADTH does not have control over the content of such sites. Use of third-party sites is governed by the third-party website owners’ own terms and conditions set out for such sites. CADTH does not make any guarantee with respect to any information contained on such third-party sites and CADTH is not responsible for any injury, loss, or damage suffered as a result of using such third-party sites. CADTH has no responsibility for the collection, use, and disclosure of personal information by third-party sites.

Subject to the aforementioned limitations, the views expressed herein do not necessarily reflect the views of Health Canada, Canada’s provincial or territorial governments, other CADTH funders, or any third-party supplier of information.

This document is prepared and intended for use in the context of the Canadian health care system. The use of this document outside of Canada is done so at the user’s own risk.

This disclaimer and any questions or matters of any nature arising from or relating to the content or use (or misuse) of this document will be governed by and interpreted in accordance with the laws of the Province of Ontario and the laws of Canada applicable therein, and all proceedings shall be subject to the exclusive jurisdiction of the courts of the Province of Ontario, Canada.

The copyright and other intellectual property rights in this document are owned by CADTH and its licensors. These rights are protected by the Canadian Copyright Act and other national and international laws and agreements. Users are permitted to make copies of this document for non-commercial purposes only, provided it is not modified when reproduced and appropriate credit is given to CADTH and its licensors.

About CADTH: CADTH is an independent, not-for-profit organization responsible for providing Canada’s health care decision-makers with objective evidence to help make informed decisions about the optimal use of drugs, medical devices, diagnostics, and procedures in our health care system.

Funding: CADTH receives funding from Canada’s federal, provincial, and territorial governments, with the exception of Quebec.

Contact with inquiries about this notice or legal matters relating to CADTH services.

ISSN: 1715-555X


Related Content

  • Evidence on Digital Health