Health care is one sector for which the massive amount of data produced daily holds great promise. Data from electronic health records, medical devices and health-related apps (and other sources) has grown so large that they exceed human capabilities for analysis. This is where Artificial Intelligence (AI) comes into play. Through various modeling techniques, AI can help medical researchers analyze massive data sets and identify patterns, which can in turn improve medical diagnostic capabilities and health care services. AI is already being applied to improve and accelerate the accuracy of diagnosis and disease screening, develop new medications, and support public health interventions. Numerous research studies focused on AI have helped to identify diseases such as skin cancers, signs of stroke, diabetes complications or eye disease. There have also been initiatives to predict a patient’s response to a treatment, including heart transplant rejection, cancer treatments, and how well immunotherapy will work on patients with melanoma. Besides AI models for treatment prediction and medical diagnosis, there have also been initiatives to develop tools that can increase quality of life for those living with chronic diseases. For example, Project Euphonia  is an initiative that focuses on training speech recognition models to aid those suffering from atypical speech patterns caused by certain conditions. Through their work, they have been able to help many people who lost their speaking ability communicate with their family and friends; as in the case of American professional football player Tim Shaw, who progressively lost his speaking ability due to ALS. 

The Risks of AI

Despite its great promise, there are many risks that must be considered when using AI for healthcare. One of the biggest risks is replicating human bias through AI models. The Alan Turing Institute identified this situation as one of the biggest obstacles to why, in the United Kingdom, AI had made little impact in fighting the COVID-19 pandemic. MIT Assistant Professor Dr. Marzyeh Ghassem also identified this risk while conducting her dissertation in computer science, when she discovered that models performed differently for certain racial minority groups. Based on this experience she concluded that algorithms developed for care delivery need to be trained with data sets that represent diverse patient populations and not designed using “black box” models. In other words, to avoid replicating these biases, AI research needs to include large, representative, and balanced data sets which have undergone testing to verify their accuracy before being used on actual patients. On this note, Dr.Andrew Ng also highlights another risk, which is that some models use data based on specific training sets that depend on local data, thus their performance may diminish significantly when/if applied in a different context. 

How to Overcome These Obstacles?

Although these risks are not issues exclusive to AI’s application in healthcare, it is important to design mitigation and good practice strategies to help practitioners and researchers address them. The WHO, for instance, presented a recent global report on Artificial Intelligence (AI) in health that maps the sector’s progress, obstacles, and future, and outlines six guiding principles for AI’s design and use in health. In terms of research, the AIDE Project, an initiative that seeks to understand the human factors that will help or hinder the implementation of AI in healthcare, has also taken on the task of identifying how AI is expected to impact the healthcare sector and determine the main concerns that need to be addressed to promote its acceptance and adoption. There are also initiatives such as the nonprofit Nightingale Open Science, that are actively working to improve the quality and scale of data sets available to researchers by collecting patient’s data, anonymizing them, and granting access for nonprofit research.

Join us in this edition of Links We Like as we explore some of the ways AI is being used in the healthcare sector, from maternal healthcare to mental health. Resources are available in English, Spanish, and Portuguese.