AUTOMATIC DETECTION AND CLASSIFICATION OF PLEOMORPHIC SMALL BOWEL LESIONS WITH DIFFERENT BLEEDING POTENTIAL USING A CONVOLUTIONAL NEURAL NETWORK: A MULTICENTRIC STUDY

Aims Capsule endoscopy enables the detection of enteric pleomorphic lesions with different bleeding potentials. However, reading CE exams is a time-consuming and monotonous task prone to errors. Convolutional neural networks (CNNs) are highly efficient artificial intelligence tools for image analysis. Our group developed a CNN-based model for detecting and differentiating pleomorphic small bowel lesions with distinct hemorrhagic potential using CE images.

Methods Our group developed, trained, and validated a denary CNN based on CE images. Each frame was labeled according to the type of lesion (lymphangiectasia, xanthomas, ulcers, erosions, vascular lesions, protruding lesions, and blood) by three experts in CE. Saurin’s classification assessed the hemorrhagic potential: P0 – lesions without bleeding potential; P1 – lesions with uncertain bleeding potential; P2 – lesions with high bleeding potential; P3 – luminal blood. 55380 frames of the enteric mucosa were obtained from 2565 CE exams from two different centers (1483 from São João University Hospital and 1082 from ManopH Gastroenterology Clinic). 90% of the frames were used to create the training dataset, and 10% used to test the network. The patients included in the training dataset were excluded from the testing dataset.

Results The model had an overall accuracy of 98.3%, a sensitivity of 89.6%, and a specificity of 98.9%

Fig. 1

Conclusions The authors developed a CNN for the automatic identification and classification of pleomorphic lesions in CE videos and tested it in AI naïve patients. This represents an evolution in the technology readiness level into a real-life clinical setting that may significantly improve the diagnostic yield of CE exams.

Publication History

Article published online:
14 April 2022