“We’re now in an era that some consider an AI renaissance, with enormous amounts of computing power — unimaginable only a few decades ago — now available to institutions and even individual researchers. Machine learning algorithms and AI are performing feats once considered to be exclusive domains for humans.” –Sushovan Guha, MD, PhD
In recent years, the prevalence of artificial intelligence (AI), machine learning (ML), and other breakthrough computer technologies has seen significant growth in medicine. Utilized by radiologists, neurologists, pathologists, and many other specialists, artificial intelligence has greatly expanded human capabilities and proved incredibly useful in parsing and aggregating enormous amounts of medical data. The ability to use this data to learn by itself and improve on its capabilities via machine learning has made AI a constantly self-evolving resource. Deep learning (DL) – a subset of ML – has shown exceptional performance in image analysis through its use of the convolutional neural network (CNN), adding to the growing list of AI applications in healthcare and gastroenterology specifically.
Artificial Intelligence Applications in Gastroenterology
The use of AI in medicine has now expanded to include gastroenterology, in which computer algorithms can help identify bleeders, colorectal polyps, and angioectasias with very high accuracy rates. In this field, clinicians are presented with vast datasets and utilize a host of imaging devices – ranging from colonoscopies to capsule endoscopies – all of which can be made more effective with the help of artificial intelligence. Its applications in the field of gastroenterology have been examined in several studies, revealing that AI can assist physicians in making diagnoses and predicting prognoses via statistical analysis, as well as analyzing images for improved patient care.
As research has revealed, AI can help colonoscopists identify up to 20% more polyps than they would normally, significantly improving adenoma detection rates, which vary from 7% to 53% depending on the physician. The 2018 study revealed the CNN’s ability to identify polyps with a cross-validation accuracy of 96.4%, indicating a potential breakthrough in decreasing interval colorectal cancer rates.
AI carries significant potential to improve the accuracy of diagnostic conclusions and efficiency of care thus, improving patient outcomes, reducing health care costs, physician workload, burnout levels, staff shortages, and possible complications. Due to its rapid progression, the rise of AI in gastroenterology requires physicians to learn the utility, strengths, and limitations of the software while preparing for the forthcoming changes and effects of the technology on real clinical practice.
Challenges for the Future of AI
According to a recently published paper in the World Journal of Gastroenterology, the application of AI in the field is not without its faults. Physicians using artificial intelligence algorithms need to be aware of the potential inherent pitfalls of selection bias, overfitting, and spectrum bias, which have the possibility of overestimating the accuracy of a diagnosis and generalizing results. As a result, external validation is mandatory and many AI systems still require further investigation to ensure both the safety and efficacy of their application.
Experts further underscore the potential challenges facing the future of AI in gastroenterology. During Digestive Disease Week 2019 in San Diego, Medscape spoke to Dr. Sushovan Guha, a gastroenterologist and physician executive director at Banner Digestive Disease Institute in Phoenix, Arizona about the barriers affecting the field. Although the technologies are promising, there are several concerns to address before artificial intelligence can be enter the clinical arena, including the legality of the practice. Despite high accuracy and sensitivity rates of these machines, errors and miss rates are unavoidable and the liability for those mistakes must be determined.
Another issue mentioned by Dr. Guha is the process of validation of the thousands of computer algorithms being developed. Currently, there is little standardization for the validation and approval of these technologies, however, the Center for GI Innovation and Technology and FDA are working on establishing procedures and guidelines. Not only does there need to be specific and comprehensive regulations for the use of AI, but clinicians need to be well-versed and prepared to adopt a wide variety of novel technological tools. Both providers and patients will need to understand the role of AI in the traditional health care setting, its utility, and the strengths and limitations of each algorithm.