ARTIFICIAL INTELLIGENCE AND CAPSULE ENDOSCOPY: A BINARY CONVOLUTIONAL NEURAL NETWORK MODEL APPROACH FOR THE AUTOMATIC DETECTION OF ULCERS AND EROSIONS

Aims Ulcers and erosions are frequent findings in capsule endoscopy (CE) exams. CE is a key element in the follow up of patients with Crohn’s Disease (CD). Nevertheless, reading capsule endoscopy exams is time-consuming and prone to errors. Convolutional neural networks (CNN) are artificial intelligence tools with high performance levels in image analysis. This study aims to develop a CNN-based model for automatic identification of ulcers and erosions in CE images.

Methods The development of CNN was based on a database of CE images. This database included normal small intestinal mucosa images or non-erosive findings and images of enteric ulcers and erosions. For CNN development, 19340 images (16175 normal mucosa, 3165 ulcers, or erosions) were ultimately extracted. Two image datasets were created for CNN training and testing.

Results The network was 96% sensitive and 98% specific for detection of ulcers and erosions in the small bowel, providing accurate predictions in 98%. The CNN had a frame reading rate of 149 frames per second.

Conclusions The developed algorithm accurately detects ulcers and erosions in CE frames. The development of these automatic systems may allow to improve the diagnostic yield of CE for these lesions and increase the efficiency of the reading process of CE exams.

Publication History

Article published online:
14 April 2022