|
 
 |
| WHITE PAPER |
|
| Year : 2021 | Volume
: 6
| Issue : 5 | Page : 8-18 |
|
Artificial intelligence and its contribution to overcome COVID-19
Arun Chockalingam1, Vibha Tyagi2, Rahul G Krishnan3, Shehroz S Khan4, Sarath Chandar5, Mirza Faisal Beg6, Vidur Mahajan7, Parasvil Patel8, Sri Teja Mullapudi9, Nikita Thakkar10, Arrti A Bhasin11, Atul Tyagi12, Bing Ye4, Alex Mihailidis4
1 Department of Medicine, University of Toronto; Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, Ontario, Canada
2 Office of Research Services, Innovation and Entrepreneurship, Durham College of Applied Arts and Technology, Oshawa, Ontario, Canada
3 Department of Computer Science, University of Toronto, Toronto, Ontario, Canada
4 KITE Research Institute, Toronto Rehabilitation Institute, University Health Network; Department of Occupational Science and Occupational Therapy, University of Toronto, Toronto, Ontario, Canada
5 Mila, University of Montréal, Montreal, Quebec, Canada
6 School of Engineering Science, Simon Fraser University, Vancouver, Ontario, Canada
7 Mahajan Imaging Centre, New Delhi, India
8 Radical Ventures, Toronto, Ontario, Canada
9 Noetic Fund, Toronto, Ontario, Canada
10 Department of Laboratory Medicine and Pathobiology, University of Toronto, Toronto, Ontario, Canada
11 Department of Health Research Methods, Evidence and Impact, McMaster University, Hamilton, Ontario, Canada
12 Alpha Global IT, Toronto, Ontario, Canada
| Date of Submission | 20-Sep-2021 |
| Date of Decision | 27-Sep-2021 |
| Date of Acceptance | 01-Oct-2021 |
| Date of Web Publication | 19-Nov-2021 |
Correspondence Address:
Dr Arun Chockalingam
Department of Medicine and Global Health, University of Toronto, Toronto
Canada
 Source of Support: None, Conflict of Interest: None  | Check |
DOI: 10.4103/2468-8827.330646
Artificial intelligence (AI) has a great impact on our daily living and makes our lives more efficient and productive. Especially during the coronavirus disease (COVID-19) pandemic, AI has played a key role in response to the global health crisis. There has been a boom in AI innovation and its use since the pandemic. However, despite its widespread adoption and great potential, most people have little knowledge of AI concepts and realization of its potential. The objective of this white paper is to communicate the importance of AI and its benefits to society. The report covers AI applications in six different topics from medicine (AI deployment in clinical settings, imaging and diagnostics, and acceleration of drug discovery) to more social aspects (support older adults in long-term care homes, and AI in supporting small and medium enterprises. The report ends with nine steps to consider for moving forward with AI implementation during and post pandemic period. These include legal and ethical data collection and storage, greater data access, multidisciplinary collaboration, and policy reform.
Keywords: Applications of artificial intelligence, artificial intelligence, artificial intelligence in healthcare, COVID-19, electronic health records, imaging, investor's perspective, long-term care homes, machine learning, pandemic
How to cite this article:
Chockalingam A, Tyagi V, Krishnan RG, Khan SS, Chandar S, Beg MF, Mahajan V, Patel P, Mullapudi ST, Thakkar N, Bhasin AA, Tyagi A, Ye B, Mihailidis A. Artificial intelligence and its contribution to overcome COVID-19. Int J Non-Commun Dis 2021;6, Suppl S1:8-18 |
How to cite this URL:
Chockalingam A, Tyagi V, Krishnan RG, Khan SS, Chandar S, Beg MF, Mahajan V, Patel P, Mullapudi ST, Thakkar N, Bhasin AA, Tyagi A, Ye B, Mihailidis A. Artificial intelligence and its contribution to overcome COVID-19. Int J Non-Commun Dis [serial online] 2021 [cited 2023 Mar 31];6, Suppl S1:8-18. Available from: https://www.ijncd.org/text.asp?2021/6/5/8/330646 |
| Introduction | |  |
Artificial intelligence (AI) has already become an inseparable part of our daily lives. It assists our daily living and makes our lives easier and better. For example, Gmail uses AI to predict a person's next words according to the email content and subject line, which helps the person to compose his/her email faster.[1] Navigation apps provide users with different routes to reach the destination.[2] AI is playing an important role in our lives, which has become even more apparent during the coronavirus disease (COVID-19) pandemic. The WHO has provided six guiding principles for the design and use of AI to increase public awareness in all countries.[3] Many AI applications have been developed to fight the pandemic such as a tracking system to spot virus outbreaks,[4],[5] and vaccine development.[6],[7] However, even with the strong presence of AI in our society, it has not reached a level to achieve global impact.[5]
The goal of this white paper is to describe the fundamentals of AI, its application in health care through five different examples, use of AI in business operations and an investor's perspective on AI. They are addressed in detail through different subheadings and concluded with recommendations to governments and policymakers.
- A brief overview of AI and its role during the pandemic
- Opportunities for machine learning (ML) during COVID-19
- Reinforcement of learning in drug discovery
- The benefits of AI in long-term care (LTC) homes
- AI for estimation of body composition from (CT) images
- Validation and deployment of clinical grade AI application through four examples
- AI in supporting small and medium enterprises (SMEs) in the time of COVID-19, and
- AI in health care in times of COVID-19 – An investor's perspective
- Critical analysis of existing system.
| Brief Overview of Artificial Intelligence and Its Role during the Pandemic | | |
AI is an umbrella term – it is an area of science and practice that includes a multitude of different areas such as autonomous systems, neural networks, pattern recognition, and language processing.
AI is playing an important role in fighting COVID-19 and in supporting the pandemic and postpandemic world. Specifically, one of the areas that is going to be seen as a common theme in our white paper is ML. One application of ML as we know is in collecting data from hospitals where patient data are obtained and linked to clinical outcomes. This step allows us to design and train ML models that can predict patient outcomes and take actionable steps to affect outcomes.
In terms of building the models, the key aspect is the patient data and the training model that is being developed. In addition, validating the model itself is critical because once the model has been validated, we can then deploy the ML models to predict a variety of useful outcomes such as whether a patient needs to be admitted or re-admitted to the hospital. Several groups are developing these decision-making tools using ML that help and support clinical staff when making decisions about their patients and the type of care to be provided.[8]
Population-level modeling has become truly critical not only in tracking the COVID-19 pandemic and its different variants but also in applying it to make decisions for patients.
In relationship to COVID-19, the AI community has been looking at a variety of these techniques involving ML and has utilized them very quickly when the pandemic struck.
COVID-19 has really accelerated the adoption of foundational digital infrastructure which will, in turn, accelerate the adoption of advanced technology such as AI. This helps to address many of the challenges emerging from the pandemic. For instance, we are seeing startup companies such as Toronto-based BlueDot[9] that has helped to monitor and track the spread of the pandemic. AI-based computed tomography CT scans and X-ray analyses have also helped us to understand the progression of the disease. Startups have helped to improve access to health care.
Finally, AI has been instrumental in the development of many messenger ribonucleic acid (mRNA) vaccines as well as in identifying therapeutics that can be used for treating COVID-19.[10] All these developments have presented many opportunities and challenges within the AI field around continuing development of AI solutions that are being implemented in health-care settings, specifically to help with the pandemic but also to address broader health care issues as well. If there are any positives to the pandemic, it is the fact that the AI community is embracing the difficult challenges and opportunities and working in cohesion to solve them. This will have far-reaching benefits beyond the pandemic.
| Opportunities for Machine Learning during COVID-19 | | |
Background
While the last year has been a traumatic experience for the world, science has kept pace with the spread of the vaccine. The research, development, and manufacture of vaccines for COVID-19 progressed at breakneck speed, and as countries ramped up their vaccination programs, the mitigation and eventual management of the pandemic lies in sight.
A year's worth of experience with COVID-19 around the world can be tabulated and compiled as data; and this brings unique opportunities for the application of ML models to glean useful and actionable insights to aid in disease management. We highlight two research opportunities where ML can help clinicians, policymakers, and epidemiologists in managing the pandemic using observational health record data.
Making predictions from observational data
The first opportunity we discuss comes in the form of leveraging data that characterize individuals and clinical outcomes of interest. This would require us to collect and tabulate large datasets of phenotypic characteristics of patients (such as their age, gender, demographic information, occupation, and their genetic data) alongside their clinical outcomes (such as the degree and severity of an infection, whether the patient was admitted to the intensive care unit and whether the patient needed intubation). Such data may be found in biobanks[11] linked to patient health record data making it a useful store of information for building ML models.
ML has seen numerous successes in predicting clinical outcomes from patient data.[12],[13],[14] In the context of COVID-19, one example is using chest X-ray data made available to researchers to build and create predictive models for infection severity[15] and to develop an ML model to predict the probability of a positive COVID-19 test using data from patient symptoms. Beyond prediction, a variety of techniques have been developed in ML that allows us to inspect models to understand how it makes predictions. In simple models, these can take the form of inspecting the model's parameters. For example,[16] one would develop regression models for predicting the probability of being infected with COVID-19; by inspecting the regression coefficients, they would study the relative risk of occupational groups for the likelihood of infection. If, after rigorous testing, the model proves to be robust and reliable, the knowledge from such a model can be deployed to prioritize vaccinations when supply is scarce.
Machine learning for population-level modeling
The next opportunity comes in the form of combining modern advances in ML with epidemiological compartmental models.[17] Compartmental models quantify variation in infection count using dynamical systems that model the proportions of susceptible, infected, and recovered patients in a population. The parameters of the model yield reproduction numbers which quantify the severity of infection.
There have been several interesting research directions pursuing this line of research on how compartmental models may be used as simulators in model predictive control to identify good strategies which, if followed, can help cities and provinces flatten their epidemic curve.[18] A Gaussian process-based compartmental model shows how it can be used to create a policy-impact predictor. Researchers use the model to visualize the effect of different interventional policies and their downstream impact on infection counts. However, little is known about what model represents the population-level dynamics of COVID-19. Knowing the answer to this question would require pooled data not just of (accurate) infection counts but also careful tabulation of the various interventions undertaken in different cities and communities.
It is important to be aware of and to highlight the technical challenges surrounding the research directions. First, the methodological developments must address the myriad realities of observational data collected during a pandemic such as missingness, bias, and noise. Second, it is important to encourage transparency and accuracy in the processes that underlie the collection and tabulation of observational health data since models built on untrustworthy data have limited utility. Third, to answer some of the questions, there will invariably be a need to build mechanisms to house and link diverse sources of data and bring interdisciplinary teams of individuals together – this should be encouraged. By building shared research infrastructure and encouraging openness and transparency, we can enable researchers to better aid in the management and care of COVID-19, besides creating mechanisms to be better prepared for the next global pandemic. Finally, regardless of the technical capabilities of a ML model, it remains vital to use them in conjunction with domain experts rather than as a standalone method to automate decision making.
| Reinforcement of Learning in Drug Discovery | | |
Background
A molecule is usually represented as strings or simplified molecular-input line-entry system (SMILES),[20] or as a graph where the node of the graph is the atom, and the edge is the bond itself. Several advances in natural language processing and in graph neural networks in the context of supervised, unsupervised, and self-supervised reinforcement learning have been leveraged to push for the frontiers in drug discovery. Again, this is an example of how a different area of AI such as natural language processing has been applied in this situation as well.
Role of machine learning
Initially, genetic algorithms were used to design molecules. Recently deep generative models like variational autoencoders and generative adversarial networks[21] were used on string or graph representations of molecules, and they performed at par. It can also be formulated as a reinforcement learning problem where each atom or bond can be added or deleted every step; it was followed by retrosynthesis pipelines. Here, neural networks trained to predict reaction outputs were used to plan the synthesis pathway to bridge these two disjointed pipelines. The reinforcement learning algorithms were proposed directly on the chemical reaction space and are currently considered to be state of the art for molecular synthesis.
While representations like molecular fingerprints[22] are good enough for training simple neural networks to predict properties such as QED or CLogP, SMILES, or graph representations were used for predicting molecular activities against the chosen target. The suitable ML and related different approaches that can be applied in drug discovery can be seen in this case.
| The Benefits of Artificial Intelligence in Long-term Care Homes | | |
Background
The proportion of older adults in Canada has been steadily growing over the past few decades and is expected to continue further.[23] The prevalence of many chronic diseases, such as heart disease, stroke, and dementia, increase with age[24] and consequently, the demand for LTC homes will continue to rise.[25] Staff shortage has always been a challenge in LTC homes,[26],[27] which was further aggravated during the COVID-19 pandemic with an increased workload for staff who were in fear of contracting the virus, leading to additional staff shortage as colleagues had to stay home with any COVID-19 related symptoms.[28],[29] As such, it is important and urgent to find new approaches to address these issues.
Role of artificial intelligence in long-term care homes
AI technology has the potential to significantly improve staff productivity, so the same number of staff can provide efficient and quality care to more residents. It, therefore, has the potential to reduce the staff workload and, at the same time, enhance the quality of care provided and residents' quality of life (QoL).[26],[27] AI technology is evolving and becoming more prevalent in our daily lives. Most older adults embrace new technology and have a positive attitude towards it.[30],[31],[32] In addition, the LTC homes that adopt AI technology could make them more competitive among other facilities.[3] In summary, with the intensive demand of staff and increased dependency on technology, it is urgent to expand the technology use in LTC homes.
Potential areas where artificial intelligence can be used in long-term care homes
The followings are some examples of technology that could be used in LTC homes. The examples are chosen as they address important needs in LTC homes.
Social isolation
A communication platform[33] that allows residents who have limited computer skills to connect with their families. This technology improves residents' QoL by providing socializing opportunities and improves staff productivity by not monitoring residents all the time.
Pain management
An ambient pain monitoring system[34] monitors facial expressions of pain in residents with dementia who have communication difficulties. Staff shortage makes frequent pain assessment impossible, and a system that automatically flags residents who might need attention will significantly improve staff productivity in identifying and treating those in pain.
Fall detection
A fall detection system[35],[36] that allows detecting and predicting falls and sends alerts to staff. The system will allow staff to provide timely care to residents and reduce emergency room visits.
Fall prevention
An ambient monitoring system[37],[38],[39] monitors residents' gait. It alerts staff when it detects that a resident's gait and balance are impaired, and they are at a high risk of falling. This system will allow staff to prioritize their time and attention to residents at high risk of falling and will result in fewer falls and related injuries.
Behaviors detection
A multimodal wearable and/or computer vision system[40] that monitors the behaviors of residents and sends alerts whenever an agitated or anomalous behavior is detected. This system will reduce injuries and accidents in the care homes and reduce the burden on caregiving staff.
Despite availability of many AI solutions, the uptake of technology remains low.[25],[27],[41],[42],[43],[44],[45] Many technologies lack the customization needed to be user friendly.[25],[46]
Barriers in technology implementation
Some examples of technology that are currently used in LTC homes include telehealth,[41],[44] real-time location systems,[45] and social robotics,[47] and adoption of these technologies is slow and time consuming.[41],[42],[44] Key barriers identified include, but not limited to, awareness, user acceptance and adoption, and government support.[25],[48]
Furthermore, if a system cannot perform as it is expected, has technical failures during use, or generates inaccurate information, it would further frustrate the users and result in technology abandonment.[45],[49] Therefore, technology needs to be reliable and have an intuitive design that requires minimum learning.
To design a useful and usable technology, it is critical to understand the users' needs and apply a user-centered approach when designing technology. The user-centered approach is an iterative design that underlines user's needs and includes users in the entire process of technology design.[50] Several studies have demonstrated that actively engaging all LTC stakeholders (i.e. staff, residents, and family caregivers) could increase the acceptability of LTC technology and help with technology implementation.[45],[51],[52],[53],[54] Involving LTC stakeholders will, in turn, increase their awareness of the available technology and improve overall technology use in LTC homes. It is challenging to have a technology to fulfill all LTC residents' needs.[25] Therefore, it is essential that technology has the capability to be customized to meet individual needs.[30],[46]
The cost could be another factor for technology acceptance.[46],[48],[55] Currently, most new technologies developed for LTC use have little or no reimbursement from a third party (i.e. insurance company) and/or government.[48] Therefore, such technology would be only available to those who can afford it and who are willing to pay for it,[48] which could create a power differential among the residents. The management of the LTC homes will only consider a technology if the perceived benefit justifies the required cost and time investment.
| Validation and Deployment of Clinical Grade Artificial Intelligence Application through Four Examples | | |
Background
There is no doubt that AI is the future of medicine. Industry experts estimate that the global market for AI-based solutions in health care will be $118.2 Billion by 2027[56] and that will reduce systemic waste to the tune of $200B[57] in the health-care ecosystem. That said, the adoption of AI in clinical settings, for clinical purposes, is insufficient. This is due to three main reasons:
- There are too many AI solutions available – it is estimated that there are more than 200 startups across the world building niche medical imaging AI solutions
- There is no way for doctors to evaluate and thereby trust AI
- There is no universal, single way for AI to be integrated into clinical workflows.
Role of artificial intelligence and solution
This challenge can only be solved by an “operating platform” that connects three disparate components of the AI ecosystem and workflow:
- Healthcare providers, or users of the AI tools
- Analytics providers, or developers of the AI tools
- Health tech providers, or the companies that build software and hardware using which hospitals run – EMR, picture archiving and communication system, and radiology information system, etc.
CARPL[58] – developed by CARING – the Centre for Advanced Research in Imaging, Neurosciences and Genomics – is such a platform. It is a decentralized technology platform that connects the fragmented health-care ecosystem and enables the transition of AI from bench to clinic.[3] It broadly comprises of the following features:
- Dataset management and search – Any AI algorithm, before to deployment in clinical workflows, needs to be validated extensively from a clinical perspective
- Annotation platform – all AI needs to be tested against strong ground truth
- Validation platform – the heart and soul of CARPL, the validation (or predeployment testing) platform allows clinical users to evaluate, and thereby build trust, in AI algorithms. The platform automatically calculates relevant statistics, shows relevant false-positive and false-negative cases, and allows users to even check for explainability and bias within AI algorithms
- Deployment platform – the final piece of the puzzle is the integration of live AI algorithms into clinical workflows. CARPL allows users to integrate AI algorithms of their choice using digital imaging and communications in medicine and HL7 protocols, which are standard communication protocols used across the world.
In summary, we use CARPL as an example of what a “universal” platform for AI should be able to do for the successful clinical deployment of AI algorithms. Clinical deployment of AI will always follow validation and integration of the algorithms into clinical workflows. Each AI algorithm is different, hence needs to be tested differently, depending on the input/output of the algorithm, and even the clinical context of the deployment of the algorithm. Lastly, predeployment testing is key – think of it as an “interview” for the AI algorithm.
| Artificial Intelligence for Estimation of Body Composition from Computed Tomography Images | | |
Background
Body composition, which is made up of skeleton muscle (SM) mass and fat, is an important metric to predict individual's clinical outcomes. Variation in the amount of SM can explain the variability in clinical outcomes, especially in cancer treatment. CT images provide a direct visualization of internal anatomy and can be used to measure internal SM precisely. However, in the past, only 2D single slice images have been used because of ease of measurement. Current research is done to see if we can draw additional information from these CT images to predict more accurate clinical outcomes or if we can extract it with more convenience to enable large scale biomarker discovery as well as incorporate them in clinical workflows.
Role of artificial intelligence
Literature cites several methods to extract measurements of muscle and fat through single CT slices, but it showed variations by 15%–30%.[59],[60] This study was done to develop AI algorithms (a) to improve the measurement of muscle and fat by shifting from manual measurements to automated measurement and (b) to convert 2D images to 3D images. The study established that AI-empowered process could automatically segment the CT images in a cost-effective manner through data analysis facilitation suite and provides annotation of various vertebral levels allowing them to be compared across different individuals.
The future diagnostics will use the AI framework and align it with an individual's sociodemographic factors, genetic factors, lifestyle, metabolic factors, and physiological factors to predict their clinical outcomes more precisely.
The main goal of the software developed in this study is to segment all the organs inside an individual. A prediction model is developed that predicts what the outcomes of a particular kind of measurements that go into the treatment. Once a new patient comes in, same measurements can be performed, and previously trained clinical model can be used to make the prediction of that individual for the given treatment. This is precision medicine for individuals using AI. These prediction models can be used in different clinics and different workflows around the world.
| Artificial Intelligence in Supporting Small and Medium Enterprises in the Time of COVID-19 | | |
Background
In addition to its impact on public health, COVID-19 has triggered the deepest economic recession in nearly a century, threatening health, disrupting economic activity, and harming personal well-being and job security. Gita Gopinath, the IMF's chief economist, said the crisis could knock $9 trillion (≤7.2 trillion) off global GDP over the next 2 years.[61] The IMF also estimates that the global economy shrunk by 4.4% in 2020, a decline that is estimated to be the worst since the great depression in the 1930s.
Several essential industries such as food, medical, travel, and transportation are grappling with the immediate impact of COVID-19 due to the drop in retail footfall, the disruption of manufacturing and supply chain operations, and workplace employee health concerns. According to an Accenture study, the COVID-19 pandemic has hit SMEs the hardest. SMEs represent almost 90% of total businesses, and over 50% of employment worldwide.[62] The vulnerability of SMEs has become particularly pronounced with the COVID-19 crisis as most of them lack the capital reserves to weather months-long interruptions. The three major factors that determine the intensity of this pandemic's impact on SMEs are (a) type of business, (b) financial fragility of small businesses, and (c) duration of COVID-related disruptions.
Measures and role of artificial intelligence
To adapt to the complexity and uncertainty of the pandemic, many small businesses pivoted and adopted technological measures such as using contactless deliveries to make their services available, asking employees to learn new skills to support changes to the business models, integrating new technology processes into their business, etc.
Emerging technologies have the capability to resolve problems and issues and support businesses in various ways such as automating work or predicting sales and detecting fraud. Small businesses and startups are starting to utilize the techniques of AI and ML to survive the pandemic. The major areas are:
- Conversational AI platforms – These platforms are proving to be more cost-effective and efficient and replacing traditional customer support methods
- AI in logistics operations – The use of AI in logistics is enabling companies to manage volatile demands, resource efficiency, and upsurge in costs due to the COVID-19 pandemic
- AI in business analytics - AI empowered business analytics are helping small businesses to make data-driven decisions and transforming them into an automated and efficient business by reducing operational costs
- AI-based digital marketing - Recommender systems play a crucial role in effectively targeting the desired audience to gain more benefits in marketing in any organization regardless of the size
- AI to hire talents – AI tools help to automate recruiting workflows, especially to auto-screen candidates, and conduct sentiment analysis on job descriptions to identify potentially biased language, among others.
Overall, postpandemic, small businesses have been seen shifting towards the integration of AI services into their operations. Some examples are:
- Industries such as mining, energy, manufacturing, and food quickly embraced AI tools such as computer vision and robotics for quality assurance
- Companies have invested in powerful cloud-computing technologies and centralized more corporate data in the cloud
- Online shopping has allowed retailers and AI companies to adopt virtual customer-service agents and improve their algorithms to provide personalized recommendations and generate more sales
- As employees are coming back to work, they are adopting more AI-based process automation software.
In conclusion, the survival of small businesses across the economy during and postpandemic will require new business models and in turn will accelerate the adoption of new technologies. New technologies such as AI and NLP are now available and can uncover potential cost management opportunities from within structured and unstructured enterprise data. The proliferation of AI and machine-learning innovations, along with the expansion of open-source software and free data-analytics tools, means that firms do not have to create technology solutions in-house through R and D, freeing their resources for other value-generating efforts. AI tools and technologies can be employed principally to support efforts of policymakers, the medical community, and society at large to manage every stage of the COVID-19 crisis and its aftermath: detection, prevention, response, recovery, and to accelerate research.
| Artificial Intelligence in Health Care in Times of COVID-19 – An Investor's Perspective | | |
Background
COVID-19 has significantly accelerated the adoption of AI technologies due to systemic, behavioral, and technology changes. At the same time, AI technologies have helped us address issues arising from COVID-19 across the health spectrum including tracking the pandemic, improving access to health care, and accelerating therapeutic solutions to the pandemic. As the adoption of AI increases, we will see increased access to care at a lower cost post the pandemic. However, we need to address privacy and equitable access challenges that arise from the use of AI.
Acceleration of artificial intelligence technologies
We have expected technology adoption in health-care settings to come in two waves.[63]
Wave 1: Broad-based adoption of digital technologies
Deployment of advanced technologies such as AI needs a strong foundation of basic digital infrastructure. For example, strong electronic health record (EHR) systems and the ability to capture, store, and retrieve medical data across all formats such as text, image, and video with ease; seamless access to health-care professionals through multiple mediums such as email, phone, and video.
Wave 2: Adoption of deep tech solutions such as artificial intelligence
Once the foundational infrastructure mentioned above is deployed widely, it becomes easier to access data required for training AI models, enhance collaboration between clinical and technical experts, and to deploy these models in production.
The pandemic has accelerated the deployment of digital infrastructure by several years due to the following changes:
- Systemic: Accelerated deployment of digital infrastructure with regulators and administrators on board
- Behavioral: Increased comfort with health-care professionals collaborating with each other virtually, and patients interacting with their doctors virtually
- Technology: Rapid innovation and acceptance of AI for the design of therapeutics, clinical trials, diagnostics, etc.
Artificial intelligence's contribution to managing the pandemic
New technology tools, including AI, have helped us to manage the pandemic across multiple fronts:
Tracking the pandemic
Detection of how the pandemic is spreading and what can be done to stop it:
- E.g. BlueDot,[9] a Toronto-based startup was the first to identify “unusual pneumonia” cases in Wuhan in late December 2019. They have been used by bodies at federal, state, and city levels to track the spread of the pandemic and make data-driven decisions.
Access to health care
Ease of access of health-care information:
- For example, PocketHealth, a Toronto-based startup has made it easy for patients to access their medical imaging records without visiting the hospital or clinic to get a Digital Video Disc (DVD) which was a norm of the past.
Improving access to mental health by smarter triaging of patients using automated chatbots and smarter surveys.
Accelerating therapeutic solutions
Using AI for drug and vaccine development:
- E.g. Moderna's use of AI for the development of its mRNA vaccine has been well documented and discussed.[10]
Opportunities and challenges to implement artificial intelligence in healthcare
AI is expected to affect all parts of the health-care ecosystem including therapeutic development, clinical trials, clinical diagnosis, treatment selection, and driving process efficiencies in health-care settings. However, it is important to address some fundamental challenges that arise from the use of AI:
Equitable access for all
AI can significantly help the populations in emerging economies like India, where access to health care is still limited. However, most of the AI research is currently undertaken in the western world (with the exception of China). It is important that the solutions developed by companies and research organizations in the West are done with the global population in mind (e.g. representative datasets and focus on diseases affecting emerging economies).
Patient data privacy
AI solutions require a significant amount of data to train new models. This will require getting access to detailed and varied patient data. It is important that patient privacy is respected in this process. New technologies such as federated learning and differential privacy can be helpful in maintaining privacy while providing data to train these models.
| Critical Analysis of Existing System | | |
For ML and AI to be more effective, we will require a large pool of data developed over time, strategically utilized, and analyzed to enable learning. It is possible when the data is captured at various points of contacts and is either shared between the relevant stakeholders or is centrally pooled. EHR/EMR adoption among clinicians is quite low[64] even in developed countries and that needs to increase globally if we need real time, accurate, and large database. Furthermore, there are many EMR/EHR vendors, and it is critical that interoperability and data sharing happens seamlessly among these health information platforms.
We also need to explore how we can integrate the Internet of Things (which consists of a network of smart and connected products) with data science, big data, ML, and AI to achieve the following:
- To develop and finetune optimal diagnosis and intervention policy when the systems are operating abnormally due to uncertainty in the sequential diagnosis of the pandemic impact and ways to handle the excessive and adverse load on the system
- To integrate and utilize smart products to gather data and create methodologies to learn and analyze that data to finetune the health care system.
The sharing of data and trans visibility of health-care data between health care providers, public health, and policymakers will support decision-making.
| Conclusion and Recommendations | | |
Moving forward, this pandemic has shown how the community of researchers can mobilize very quickly and how the AI community can pivot and respond in different application areas and domains. This is going to be important as we continue to pivot to apply our learnings and our algorithms that have been developed in the AI world to other domains as well. AI has shown to be a powerful tool to help access a wide range of health applications whether there is a pandemic or not and this is not going to change. In Canada, telehealth is becoming an important decision-making support tool in health and is expected to grow. This application has implications in the real world.
Recommendations
- The legal and ethical perspective of data collection and storage. As we collect multiple different digital health data, we should be conscientious as to the use of the data, and how it is going to be shared across the world including future planning for pandemic preparedness
- It is critical to provide access to health-care data to researchers in academia and perhaps selected industry partners for the computation of AI and ML
- A systematic integration of data collected by clinicians, nurses, and frontline workers with proper analysis and interpretation by domain expertise to drive future research questions critical to deal with current and future pandemics
- Data collected during the pandemic using telemedicine needs to be harnessed to prepare us for any future pandemics
- To increase the uptake of the technology in LTC homes, government involvement and financial support are needed for the research, development, and implementation of AI technology
- For an effective Decision Support System and to provide inputs for MLs and AI, we need to have a system where each health information system is integrated and has seamless data sharing
- There have been over ten significant epidemics in this world in the last century. Due to data recording and analysis techniques during these epidemics, an increasing amount of social contact data with time stamps has been collected. We can use this data for machines to learn and create models for detailed analysis. We need to study the similar patterns that have emerged and also the differences. The learning will help us:
- Develop models to accurately predict, prevent, and provide personalized, efficient, effective, and economical health care and achieve better outcomes for the patient, society, and the economy
- Understand the progression of those epidemics
- Understand the triggers of the pandemics to understand them and arrest them before they happen in future
- Some of these epidemics have been generated through human research. We need to explore methods to start collecting data right when the research is initiated to measure, monitor, and manage any such fallout as that we have seen during SARS 1 and COVID-19
- Explore ways to use AI and help us muster/reallocate the resources to address the pandemic globally rather than regionally.
- It is important to understand the impact of the pandemic on economic, social, and mental health and develop policies and action plan for these areas
- We need to understand which industries were adversely affected by COVID-19 and develop policies and actions required to minimize the adverse effects on those industries, thus helping economic recovery.
Ethical approval
As this is a white paper and a compilation of research evidence, this work was exempted from any necessary ethical approval.
Financial support and sponsorship
Nil.
Conflicts of interest
There are no conflicts of interest.
| References | | |
| 1. | |
| 2. | |
| 3. | |
| 4. | Cho A. AI systems aim to sniff out coronavirus outbreaks. Science 2020;368:810-1.  |
| 5. | |
| 6. | Crooke SN, Ovsyannikova IG, Kennedy RB, Poland GA. Immunoinformatic identification of B cell and T cell epitopes in the SARS-CoV-2 proteome. Sci Rep 2020;10:14179. |
| 7. | Keshavarzi Arshadi A, Webb J, Salem M, Cruz E, Calad-Thomson S, Ghadirian N, et al. Artificial intelligence for COVID-19 drug discovery and vaccine development. Front Artif Intell 2020;3:65. |
| 8. | Peiffer-Smadja N, Rawson TM, Ahmad R, Buchard A, Georgiou P, Lescure FX, et al. Machine learning for clinical decision support in infectious diseases: A narrative review of current applications. Clin Microbiol Infect 2020;26:584-95. |
| 9. | BlueDot.global. Toronto: BlueDot Outbreak Risk Software; c2020. Available from: https://bluedot.global/. [Last accessed on 2021 Sep 04]. |
| 10. | Ho D. Addressing COVID-19 drug development with artificial intelligence. Adv Intell Syst 2020. p. 2000070. |
| 11. | Sudlow C, Gallacher J, Allen N, Beral V, Burton P, Danesh J, et al. UK biobank: An open access resource for identifying the causes of a wide range of complex diseases of middle and old age. PLoS Med 2015;12:e1001779. |
| 12. | Ghassemi M, Naumann T, Doshi-Velez F, Brimmer N, Joshi R, Rumshisky A, et al. Unfolding physiological state: Mortality modelling in Intensive Care Units. KDD 2014;2014:75-84. |
| 13. | Henry KE, Hager DN, Pronovost PJ, Saria S. A targeted real-time early warning score (TREWScore) for septic shock. Sci Transl Med 2015;7:299ra122. |
| 14. | Krishnan RG, Razavian N, Choi Y, Nigam S, Blecker S, Schmidt AM, et al., editors. Early Detection of Diabetes from Health Claims 2013: NIPS Workshop in Machine Learning for Clinical Data Analysis and Healthcare. NIPS, 2013. |
| 15. | Cohen JP, Morrison P, Dao L, Roth K, Duong TQ, Ghassemi M. COVID-19 image data collection: Prospective predictions are the future. ArXiv 2020. https://arxiv.org/abs/2006.11988. |
| 16. | Mutambudzi M, Niedwiedz C, Macdonald EB, Leyland A, Mair F, Anderson J, et al. Occupation and risk of severe COVID-19: Prospective cohort study of 120 075 UK Biobank participants. Occup Environ Med 2020;78:307-14. |
| 17. | Tolles J, Luong T. Modeling epidemics with compartmental models. JAMA 2020;323:2515-6. |
| 18. | Wood F, Warrington A, Naderiparizi S, Weilbach C, Masrani V, Harvey W, et al. Planning as inference in epidemiological models. ArXiv 2021. https://arxiv.org/abs/2003.13221. |
| 19. | Qian Z, Alaa AM, van der Schaar M. When to lift the lockdown? Global COVID-19 scenario planning and policy effects using compartmental gaussian processes. ArXiv 2020. https://arxiv.org/abs/2005.08837. |
| 20. | |
| 21. | |
| 22. | |
| 23. | Statistics Canada. Canada's Population Estimates: Age and Sex; 01 July, 2020, 29 September, 2020. |
| 24. | Government of Canada. Aging and Chronic Diseases: A Profile of Canadian Seniors; December, 2020. |
| 25. | Tak SH, Benefield LE, Mahoney DF. Technology for long-term care. Res Gerontol Nurs 2010;3:61-72. |
| 26. | Ontario Long Term Care Association. This is Long-Term Care 2019. Toronto: Ontario Long Term Care Association; 2019. |
| 27. | Strategic Innovation Council. Accelerating Our Innovation Potential: Actions to Advance Innovation in Ontario's Long-Term Care Ecosystem. Toronto: The Ontario Long Term Care Association; 2018. |
| 28. | Centers for Disease Control and Prevention. Strategies to Mitigate Healthcare Personnel Staffing Shortages. CDC; 2021. |
| 29. | Centers for Disease Control and Prevention. Preparing for COVID-19 in Nursing Homes. CDC; 2020. |
| 30. | Bian C, Ye B, Hoonakker A, Mihailidis A. Attitudes and perspectives of older adults on technologies for assessing frailty in home settings: A focus group study. BMC Geriatr 2021;21:298. |
| 31. | |
| 32. | Anderson M, Perrin A. Tech Adoption Climbs among Older Adults. Pew Research Center; 17 May, 2017. |
| 33. | FamliNet: Are You a Caregiver and Feeling Exhausted? Toronto: FamliNet; c2021. Available from: https://www.famlinet.com/. [Last accessed on 2021 Sep 04]. |
| 34. | Rezaei S, Moturu A, Zhao S, Prkachin KM, Hadjistavropoulos T, Taati B. Unobtrusive pain monitoring in older adults with dementia using pairwise and contrastive training. IEEE J Biomed Health Inform 2021;25:1450-62. |
| 35. | Nogas J, Khan SS, Mihailidis A. DeepFall: Non-invasive fall detection with deep spatio-temporal convolutional autoencoders. J Healthc Inform Res 2020;4:50-70. |
| 36. | Khan SS, Nogas J, Mihailidis A. Spatio-temporal adversarial learning for detecting unseen falls. Pattern Anal Appl 2020;24:381-91. |
| 37. | Mehdizadeh S, Dolatabadi E, Ng KD, Mansfield A, Flint A, Taati B, et al. Vision-based assessment of gait features associated with falls in people with dementia. J Gerontol A Biol Sci Med Sci 2020;75:1148-53. |
| 38. | Mehdizadeh S, Sabo A, Ng KD, Mansfield A, Flint AJ, Taati B, et al. Predicting short-term risk of falls in a high-risk group with dementia. J Am Med Dir Assoc 2021;22:689-95.e1. |
| 39. | Ng KD, Mehdizadeh S, Iaboni A, Mansfield A, Flint A, Taati B. Measuring gait variables using computer vision to assess mobility and fall risk in older adults with dementia. IEEE J Transl Eng Health Med 2020;8:2100609. |
| 40. | Spasojevic S, Nogas J, Iaboni A, Ye B, Mihailidis A, Wang A, et al. A pilot study to detect agitation in people living with dementia using multi-modal sensors. J Healthc Inform Res 2021;5:342-58. |
| 41. | Alexander GL, Powell KR, Deroche CB. An evaluation of telehealth expansion in U.S. nursing homes. J Am Med Inform Assoc 2021;28:342-8. |
| 42. | Bjarnadottir RI, Herzig CT, Travers JL, Castle NG, Stone PW. Implementation of electronic health records in US nursing homes. Comput Inform Nurs 2017;35:417-24. |
| 43. | Eghtesadi M. Breaking social isolation amidst COVID-19: A viewpoint on improving access to technology in long-term care facilities. J Am Geriatr Soc 2020;68:949-50. |
| 44. | Seifert A, Batsis JA, Smith AC. Telemedicine in long-term care facilities during and beyond COVID-19: Challenges caused by the digital divide. Front Public Health 2020;8:601595. |
| 45. | Weernink CE, Felix E, Verkuijlen PJ, Dierick-van Daele AT, Kazak JK, van Hoof J. Real-time location systems in nursing homes: State of the art and future applications. J Enabling Technol 2018;12:45-56. |
| 46. | Howard J, Fisher Z, Kemp AH, Lindsay S, Tasker LH, Tree JJ. Exploring the barriers to using assistive technology for individuals with chronic conditions: A meta-synthesis review. Disabil Rehabil Assist Technol 2020. p. 1-19. |
| 47. | |
| 48. | Chapman S, Miller J, Spetz J. The Impact of Emerging Technologies on Long-Term Care and the Health Workforce. San Francisco, CA: UCSF Health Workforce Research Center on Long-Term Care; 2019. |
| 49. | Okoniewska B, Graham A, Gavrilova M, Wah D, Gilgen J, Coke J, et al. Multidimensional evaluation of a radio frequency identification Wi-fi location tracking system in an acute-care hospital setting. J Am Med Inform Assoc 2012;19:674-9. |
| 50. | Abras C, Maloney-Krichmar D, Preece J. User-centered design. In: Bainbridge W, editor. Encyclopedia of Human-Computer Interaction. Thousand Oaks: Sage Publications; 2004. |
| 51. | Broyles LM, Tate JA, Happ MB. Videorecording in clinical research: Mapping the ethical terrain. Nurs Res 2008;57:59-63. |
| 52. | Masciadri A, Comai S, Salice F. Wellness assessment of Alzheimer's patients in an instrumented health-care facility. Sensors (Basel) 2019;19:3658. |
| 53. | Niemeijer A, Frederiks B, Depla M, Eefsting J, Hertogh C. The place of surveillance technology in residential care for people with intellectual disabilities: Is there an ideal model of application. J Intellect Disabil Res 2013;57:201-15. |
| 54. | Niemeijer AR, Frederiks BJ, Riphagen II, Legemaate J, Eefsting JA, Hertogh CM. Ethical and practical concerns of surveillance technologies in residential care for people with dementia or intellectual disabilities: An overview of the literature. Int Psychogeriatr 2010;22:1129-42. |
| 55. | Xiong C, Ye B, Mihailidis A, Cameron JI, Astell A, Nalder E, et al. Sex and gender differences in technology needs and preferences among informal caregivers of persons with dementia. BMC Geriatr 2020;20:176. |
| 56. | |
| 57. | |
| 58. | Carpl.ai. CARING Research: Taking AI from Bench to Clinic. Mahajan Imaging Research and Development Private Limited; c2020. Available from: https://www.carpl.ai/platform. [Last accessed on 2021 Sep 04]. |
| 59. | Chung YE, Kim MJ, Park YN, Choi JY, Pyo JY, Kim YC, et al. Varying appearances of cholangiocarcinoma: Radiologic-pathologic correlation. Radiographics 2009;29:683-700. |
| 60. | Hemke R, Herregods N, Jaremko JL, Åström G, Avenarius D, Becce F, et al. Imaging assessment of children presenting with suspected or known juvenile idiopathic arthritis: ESSR-ESPR points to consider. Eur Radiol 2020;30:5237-49. |
| 61. | |
| 62. | |
| 63. | |
| 64. | Dutta B, Hwang HG. The adoption of electronic medical record by physicians: A PRISMA-compliant systematic review. Medicine (Baltimore) 2020;99:e19290. |
|
![JOURNAL_FULL_NAME]](images/logo.gif){kind=logo}
Search Pubmed for