Algorithms are utilized behind the scenes for providing customized services to individuals’ unique choices, in fields such as music streaming platforms to social media feeds, and search engines. Though the use of algorithms in health care has been investigated since the beginnings of artificial intelligence, advances in deep learning methods over the last decade have enabled clinicians to go after a previously inaccessible enormous amount of data, transforming the way doctors and clinical researchers detect, diagnose, and treat diseases
In addition to increased data-computing capacity and advanced algorithms, clinicians can now enter data using written and spoken words rather than just quantitative lab and imaging results. As they converse with patients regarding subjective feelings and pain levels, detailed interpretations can be coded for supplementing “poking and prodding” data collected via sensors, providing machine-learning algorithms with a more complete picture. With enough input, algorithms will be able to generate a series of patterns that physicians can utilize in their clinical practice for improving disease diagnosis and understanding.
What took place in the previous eight or so years is that new results based on these deep learning methods are permitting us to go after hard data issues that could not be accessed before, said Don Brown, senior associate dean for research at the University of Virginia, Quantitative Foundation esteemed Professor in Data Science, and professor in the Engineering Systems and Environment Department.
Brown was approached five years ago by pediatric gastroenterologist Dr. Sana Syed, who was then a gastroenterology fellow in her first year on the UVA faculty.
One of the methods she was trained to perform as a GI fellow was video capsule endoscopy, which is a procedure utilizing a tiny wireless camera for capturing pictures of your digestive tract. A capsule endoscopy camera is housed within a vitamin-sized capsule that we swallow. The camera captures thousands of pictures as the capsule travels through the digestive tract, which is transmitted to a recorder worn on a belt around our waist. Capsule endoscopy allows doctors to view inside our small intestine, which is difficult to reach with more-traditional endoscopy methods. Syed’s role as a fellow was principally that of an algorithm once these videos were downloaded from the recorder.
Syed was given the task of tagging any abnormal images, like polyps, bleeding, and ulcers, by screening through thousands of images captured by a video capsule. The time-consuming task compelled Syed to investigate the use of artificial intelligence (AI) to detect disease patterns in not only videos from capsule endoscopy, but also in other image-based diagnostic tools like radiology and pathology.
That’s when Syed approached Don Brown about combining big data and medicine. With such an accomplishment, physicians could begin to more quickly and more accurately predict important inflammatory bowel disease outcomes – not only aiding physicians but also giving patients more peace of mind.
What happens is that when a patient arrives, their data will be plugged into these larger algorithms that have been trained to learn off patterns that may represent us. Then we will be able to predict the outcome of a specific patient, Syed explained. The idea is that we will be able to get more customized data and provide specific risk percentages.
Syed is investigating the benefits of Artificial intelligence in other inflammatory bowel diseases, like Crohn’s and celiac disease, in addition to recognition of patterns in environmental enteropathy. Algorithms with the ability to forecast scar tissue development in Crohn’s patients, or individualized risk percentages for thyroid disease and celiac patients affected with diabetes, according to Syed, would be a turning point in deciding more precise diagnosis and disease likelihood, permitting clinicians in developing targeted medications and treatments. The intention is not just trying to stop the disease but in preventing and curing it.
Brown stated that for AI in medicine to be effective, data scientists and clinical researchers must be conscious of fundamental biases existing in smaller data sets that are frequently missed when that data is generalized.
Brown explained that algorithms work in a manner that the highest reward is obtained by it and if it can obtain that reward via a biased set of data, it will utilize it. We must ensure that we are providing [the algorithm] a fair cross-section of data so that it can provide us with results that truly answer the question that we are seeking.
One of the most serious issues, according to Syed, is the scarcity of publicly available data sets. Syed and her colleagues combed through 2,000 to 3,000 electronic health records for collecting data from the target population when determining patterns in celiac patients. Syed hopes to collect medical information via large industry partners like Takeda, an R&D-driven global biopharmaceutical company, rather than renewing the wheel by labeling hundreds of thousands of biopsies.
Syed explained that a lot of such cross-thinking occurs when data can be accessed openly, however that data has to be polished, organized, and thoughtfully put out there.
Once the data has been collected and confirmed to be free of bias, the final step is to ensure interpretability. When many clinicians lack a background in data science, the only way for crossing the disciplinary boundary is to ensure that the data is effectively communicated – a responsibility that falls back on the data science association.
Brown explained that the major important factor is to make it decipherable to clinicians so that they understand the matter conveyed to them by the system and the way it can be utilized, and the worth of it. They can then examine individual patients and determine what makes sense.
Machine learning algorithms are being used in a variety of medical disciplines around the world. Inside UVA Grounds, algorithms are being utilized for forecasting the effects of cardiac disease treatments, understanding health conditions via smartwatches in UVA’s Link Lab, analyzing real-time diabetes-monitoring data, etc. As Syed, Brown, and other researchers continue to improve their AI capabilities, clinicians will see a shift in how they can accurately predict, diagnose, and treat disease.
Syed stated that the more she learned regarding [big data], the more she realized its importance in medicine across all specialties.
