*EMBARGOED All research presented at the 2021 ACG Annual Scientific Meeting and Postgraduate Course is strictly embargoed until Sunday, October 24, 2021, at 3:30 pm EDT.
Oral 10 Machine Learning for Classification of Indeterminate Biliary Strictures During Cholangioscopy
Author Insight from Bachir Ghandour, MD, Johns Hopkins University Hospital
What’s new here and important for clinicians?
The aim of our project is to develop a machine learning software tool that classifies indeterminate biliary strictures as benign or malignant using both cholangioscopy images and clinical data. This would enable accurate prediction of the nature of indeterminate biliary strictures, independently from the operator’s experience in interpreting cholangioscopy based visual findings. While most of the current machine learning softwares in Gastroenterology are based solely on procedural images and videos, we believe that by incorporating clinical information we can improve the diagnostic accuracy and predictive power of our final model. For us to accomplish this, a total of 1,371,605 cholangioscopy images were obtained from 528 patients at 25 worldwide centers. So far, we managed to develop a convolutional neural network that can identify key cholangioscopy image features suggestive of malignancy – papillary projection/mass, dilated tortuous vessels and ulcerated mucosa – and differentiate them from images of normal biliary mucosa. This shows promise for the final software tool that we are aiming to develop, that combines these visual findings with the patients’ clinical data to predict malignancy in indeterminate biliary strictures.
What do patients need to know?
Indeterminate bile duct strictures remain a diagnostic challenge despite the various endoscopic, radiologic and laboratory testing that a patient may undergo. Up to 25% of patients presumed to have malignant strictures are identified to have a benign pathology after undergoing a major surgical intervention. Cholangioscopy, is an endoscopic procedure that allows endoscopists to directly examine the bile ducts using a digital endoscope. Although this greatly improves the diagnostic accuracy for predicting malignant bile duct strictures, yet, the interpretation of the visual findings remains challenging and dependent on the experience of the endoscopist. Machine learning is a method of data analysis that automatically detects data patterns and then use them to make future predictions that can support decision making with minimal human intervention and thus minimal human error. Our current project aims at developing a machine learning software that can combine patient clinical information and their cholangioscopy images or videos to predict whether an indeterminate bile duct stricture is benign or malignant. This can improve current diagnostic accuracy and reduce its dependency on the endoscopist’s experience, thus improving patient clinical outcomes through accurate early diagnosis.
Author Contact
Bachir Ghandour, MD, Johns Hopkins University Hospital
bghando1 [at] jhmi [dot] edu
Media Interview Requests
To arrange an interview with any ACG experts or abstract authors, please contact Becky Abel of ACG via email at mediaonly [at] gi [dot] org.