In recent years there have been exciting advances in gastroenterology due to the novel application of artificial intelligence (AI) and other machine learning (ML) based technologies towards solving certain problems that were herein considered to be under exclusive domains of physicians. The various applications of AI and ML in GI are towards detection, diagnosis and outcome prediction. Lesion/object detection algorithms are usually implementations of convolutional neural networks (CNNs) which attempt to detect objects of interest during an endoscopic procedure. Lesion/object diagnostic AI is similar to the previous group, but the goal is to provide a definitive diagnosis for a lesion of interest. The last group of algorithms attempt to predict the incidence of various clinical events of interest by analyzing large amounts of data to identify patterns. In the near future, we anticipate development of ensemble systems that will integrate multiple algorithms within a single free-standing or networked tower or device, which can perform both lesion detection and diagnosis. Well validated and standardized AI algorithms will be supplied to hospitals including GI divisions across the country with the ability to implement their own local flavor using their retrospective data as training sets in order to develop predictions that are tailored to the patients in their catchment area. AI thus, will eventually form an integral part of GI clinical practice, and will reduce the burden of several routine tasks on practitioners.
However, the need for deeper physician involvement has never been higher. AI algorithms may be extremely fast and do not fatigue like humans, but they come with their own set of problems. Attempts to dismiss these issues as something that should concern only the original developers and institutions can result in harm to one’s patients. The quality of all AI and ML programs are directly dependent on the quality of their training. Though they possess the ability to learn and modify their predictions according to new data perpetually, strong biases and prejudices inadvertently or unconsciously built into the algorithms may take too long to overcome if left to its natural course. Physicians thus have a responsibility to understand the mechanisms and implications of AI algorithms in their practice. They should not hesitate to explore the mechanisms and evaluate results presented with a critical, scientific mind and to reject counter-intuitive suggestions/predictions if necessary. The need for vigilance cannot be overstated since our medical practice is more complex than finance and marketing.
Well validated and standardized AI algorithms will be supplied to hospitals including GI divisions across the country with the ability to implement their own local flavor using their retrospective data as training sets in order to develop predictions that are tailored to the patients in their catchment area.
Along with ground level changes in clinical practice, AI brings with it the need for institutional and regulatory involvement and oversight. There also arise legal and ethical questions from the implementation of AI in practice. There has been no legal precedent set for assigning liability in case of an AI generating erroneous outputs that resulted in harm to the patient. Physicians will need to collaborate with lawmakers, and data scientists in order to bring about the responsible and accountable use of AI technology. The medical community must strive to settle these questions before any such events occur. Prevention is always better than cure.
Like the invention of the microscope, the germ theory of disease and the discovery of DNA before, AI brings to us its significant promise and progress. Its innovations will reduce the time and effort spent by physicians in routine and repetitive tasks and should lead them to take a more hands-on role in the development, implementation and monitoring of health care algorithms. In addition, AI will bring back enhanced physician-patient interactions that we all miss now by reducing physicians’ time on mundane tasks. Physicians will be able to serve a larger number of population and focus on the complex cases.
AI is merely the next step in the unending journey of our scientific progress. Since time immemorial, in every community there have been individuals who sought to distinguish themselves in the service of the well-being of their fellow men and women. The technology and science available at their disposal did not change their utility and purpose. From the shaman-healer of prehistory to the bloodletters of the 18th century to the gastroenterologists of today, the distinguishing character is not the instrument used, but the purpose in mind. Physicians must take the effort to educate themselves about the underlying mechanisms and capabilities of AI in order to maximally utilize the advantages provided. If technology helps to reduce costs, make more accurate diagnosis, predict and prevent complications, and in general improve the health of the community in any way whatsoever, it must be embraced wholeheartedly. Remember, seasons don’t fear the reaper.
Drs. Bijun and Guha have no conflicts to disclose.
Dr. Guha is a member of the AGA Center for GI Innovation & Technology Committee and CPT Advisory Committee.
- We anticipate creation of ensemble systems that combine multiple detection and diagnostic AI algorithms into a single device.
- AI is greatly labor saving and has tremendous potential for disruption of current clinical practices leading to more efficient workflows.
- Multiple serious issues with AI such as training bias, explainability, accountability and legal liability persist.
- It is paramount for all gastroenterologists to take a more hands on role in understanding, and even implementation and development of AI algorithms to minimize the potential for patient harm.