Oncology is seen as an early adopter of machine learning in healthcare, and as a doctor who has been working in medical oncology for over a decade, I have seen firsthand how technology has advanced the field. Now, I work at a computer vision company that is using artificial intelligence (AI) in gastroenterology and although these areas of medicine differ, I am struck by the similarities in opportunities that technology has to help advance care and improve outcomes for patients afflicted by diseases in both fields.
A deeper understanding of the molecular underpinning of malignancy and the rapid development of novel therapies required advancement in diagnostic precision which has been, in part, addressed by AI. AI is now used to enhance cancer detection on imaging, facilitate more standardized pathologic review, generate more nuanced molecular signatures, process clinical data from electronic medical records, predict prognosis and response, identify potential drug targets and more. There are many lessons to be learned from the burgeoning applications of AI in oncology that are applicable beyond cancer. First, however, we have to start by understanding the current challenges faced by providers in a given specialty and collaborating with technology companies to understand which of these problems are best suited for an AI-supported solution (and which are not). Then, we can address implementation obstacles to drive and expand adoption. Systematically addressing both of these will be necessary to bring AI-enabled products from an idea to the bedside.
Challenges faced by gastroenterologists in the endoscopy suite
In gastroenterology (GI), an area for potential positive impact using AI is the endoscopy suite. Gastroenterologists utilize endoscopic imaging for direct visualization of gastrointestinal mucosa of the esophagus, stomach, and colon. Key elements of effective endoscopy are the ability to see abnormalities that are present for further evaluation and pattern recognition for accurate diagnosis and disease management. Computer vision, a field of AI that enables computers and systems to derive meaningful information from digital images, could enhance current capabilities in endoscopy in a meaningful way.
The adoption of technology has lagged in GI. As a result, many of the same problems relating to diagnosing, treating, and researching conditions in GI have persisted; challenges that could be mitigated through innovative solutions.
- Endoscopists today may miss up to 26% of adenomas during colonoscopy. Studies have shown that there are physician and care delivery factors that can contribute to adenoma detection rates. For example, a physician’s reading of an endoscopy can be greatly influenced by patient volume, the time of day and general burnout – all factors that contribute to the number of adenomas detected when conducting a routine screening. With physician burnout at an all-time high, the risk of missed adenomas could increase. Furthermore, with a growing number of patients who will need a colonoscopy due to the lowering of the recommended age to start colon cancer screening, coupled with a specialist shortage occurring across the country, there will be a higher demand for endoscopists to perform and read endoscopies faster.
- Gastroenterologists who are also clinical researchers are losing patients to screen failure at rates of up to 70% in clinical trials for Ulcerative Colitis and Crohn’s disease (together known as inflammatory bowel disease or IBD). Despite increased efforts, recruitment rates for clinical trials in IBD are falling.
Computer vision
Specialties that rely on imaging for diagnosing and monitoring disease, like gastroenterology, have the potential to use AI to standardize care and mitigate the interpretation discrepancies that can occur. By training algorithms for standard pattern recognition, which computer vision is optimized to do, AI can improve diagnostic accuracy and facilitate stronger shared care communication between providers now using a single system for the detection and documentation of disease, which benefits patients by eliminating potential care mismanagement and the need for repeat testing.
Oncology has been an early adopter of advanced computer-aided technologies because of its reliance on two-dimensional imaging – and the field has progressed to a point where computer vision is becoming more precise because of the extensive library of data that is available. Let’s take radiology, for example. Cross-sectional imaging generates a lot of clinical image data and because of this, radiologists are tasked with sorting images, writing analyses, and confirming a diagnosis. Computer vision can be inserted into a radiologist’s workflow to sort images, identify images of concern and act as a tool to enhance a radiologist’s interpretation.
What’s next for the GI field
Using computer vision for GI brings the field into the future because it will be applied to endoscopy videos rather than two-dimensional images or pathology slides. Colonoscopies use video footage collected from the scope, not still imagery. GI’s ability to utilize AI to analyze video footage and four-dimensional images will propel scientific advancements forward in the field and start a new wave of machine learning when it comes to disease analysis. Take inflammatory bowel disease (IBD), for example. In addition to detection, using computer vision can also support the analysis of the disease and the sequencing of therapy. There is so much data involved in disease severity scoring, that by implementing an AI-equipped tool, gastroenterologists could better track patient response to various therapies and continue to build their data library for patients down the line. Further, the more an AI tool is used, the better it becomes, as it analyzes data to have a stronger understanding of pattern recognition, allowing for more accurate and acute predictions down the line.
To address the growing demand for screening colonoscopies and concerns over missed adenomas, computer vision has been developed to assist in polyp detection during routine screening colonoscopy. There are multiple devices now available to assist gastroenterologists in detecting polyps. However, broader, general adoption will require device companies to address challenges such as altered workflow, cost of a device, technical challenges related to model performance, compatibility with different endoscopy systems and reimbursement for time spent. Technology companies will need to partner with providers to work through the implementation challenges in order to realize the ultimate benefit of reducing rates of advanced colon cancer with the broader adoption of AI.
Utilizing computer vision can better facilitate communication between physicians and enable quality measurement, the sharing of findings, and outcomes research from observational data – moving toward a more standardized and democratized version of evidence-based medicine at the point of care.
About Shrujal Baxi
Shrujal Baxi, MD, MPH, is the Chief Medical Officer of Iterative Health. She is a highly experienced research physician who has successfully transitioned her academic interests into the health technology space. Dr. Baxi has spent her career identifying opportunities for improved clinical outcomes through the development of novel technological solutions for patients, providers and researchers. She previously worked at Flatiron Health and built out the Clinical Science team dedicated to expanding the utility of real-world evidence throughout the drug life-cycle including clinical development, regulatory and health authority decision making. Prior to joining Iterative Health, Dr. Baxi was the Senior Vice President of Clinical and Scientific Solutions at Verana Health, where her voice as a clinician was vital in the development of high-quality real-world data from ingestion through technology-enabled curation and analysis.
