AUTOMATIC DETECTION OF MULTIPLE LESIONS IN DEVICE-ASSISTED ENTEROSCOPY USING A CONVOLUTIONAL NEURAL NETWORK – A PILOT STUDY

Introduction Device-assisted enteroscopy, particularly double-balloon enteroscopy (DBE), facilitates an invasive exploration of the small bowel, with the advantage of enabling tissue sampling and endoscopic therapy. There has been a growing interest in the application of artificial intelligence mechanisms to different endoscopic techniques; however, there are no studies on the application of these algorithms for the detection of lesions in DBE. With the present work, we aimed to develop and test an algorithm based on a convolutional neural network (CNN) for the automatic detection of multiple lesions in DBE exams.

Materials and Methods Using frames from 250 DBE exams, we developed a CNN. We included a total of 12,870 images – 6,139 frames of normal mucosa, 2,668 frames of protruding lesions, 1,547 frames of blood and hematic residues, 1,450 frames of vascular lesions, 633 images of ulcers and erosions, and 433 frames of other findings. We used 80% of those images to create a training dataset (n = 10,296), and the remaining 20%, to validate the network. The output provided by the network was compared to a consensus classification provided by two gastroenterologists with experience in DBE. We calculated the sensitivity, specificity, accuracy, positive and negative predictive values, and area under the curve (AUC).

Results After optimizing the architecture of the network, our model automatically detected small bowel protruding lesions with an accuracy of 95.6%. Our CNN had a sensitivity, specificity, and positive and negative predictive values of 96.2%, 95.0%, 95.6%, and 95.7%, respectively.

Conclusion The authors developed a pioneer AI algorithm for the automatic detection of multiple lesions (ulcers, erosions, blood, vascular lesions, protruding lesions and others) during DBE exams. The good computational performance demonstrated enables the application of these tools in real time, which can make their use significantly easier and user-friendly. The development of these tools may enhance the diagnostic yield of deep enteroscopy exams.

No conflict of interest has been declared by the author(s).

Publication History

Article published online:
16 March 2023