Automated Lewis Score Estimation for Small Bowel Crohn Disease Using Capsule Endoscopy: Retrospective Cohort Study.

What To Know

This JMIR Medical Informatics abstract reports development and preliminary internal validation of Eval-Model, a deep-learning system that automatically estimates the Lewis score from small-bowel capsule endoscopy (SBCE) images in patients with small-bowel Crohn disease.

The model combines lesion detection (ulcers, edema), stenosis classification, and a temporal regression module.

In internal testing the system produced reasonable agreement with expert consensus scores and very fast per-frame inference; the authors underscore that larger multicenter prospective validation with independent external annotation is required before clinical use.

The study used retrospective SBCE data from 814 patients at a single center; only 89 cases had expert-consensus Lewis scores used for model training and testing, and a small same-center temporal cohort (20 cases) was used for preliminary robustness checks.

Reported performance metrics include mean absolute error, root mean squared error, and lesion detection average precision. The authors caution that the temporal-validation numbers are from a small same-center sample and need cautious interpretation.

If you use capsule endoscopy images in care, this type of tool aims to standardize and speed Lewis score estimation but is not yet validated for clinical decision-making. Larger, multicenter prospective studies with independent annotation are needed before deployment.