Post written by Hiroaki Saito, MD, from the Department of Gastroenterology, Sendai Kousei Hospital, Miyagi, Japan.
We constructed artificial intelligence to detect protruding lesions of the small bowel in wireless capsule endoscopy images and validated its ability. The results show that the system is able to analyze large amounts of images quickly and accurately.
Small bowel capsule endoscopy, which requires the analysis of more than 10,000 images, is one of the most burdensome areas for physicians and is one of the areas where AI support is expected.
The morphology of protruding lesions in the small intestine varies, and diagnosis is often difficult. There are few intensive reports of AI diagnosis of small intestinal protruding lesions; therefore, revealing the capabilities and challenges of AI-based diagnosis is important.
The constructed AI was able to satisfactorily detect protruding lesions in the small intestine on an image-by-image basis (90.7%) or on an individual basis (98.6%). It is interesting to note the differences in sensitivity by CEST category. We extracted and analyzed the features of the images that resulted in false positives and false negatives; the partialness and the similar color of the protruding lesions to the surrounding healthy mucosa were the major reasons for false-negative results, and these might cause the relatively low sensitivity for polyps and SMT; meanwhile, the normal mucosa was sometimes tagged as polyps or SMT. The establishment of methods for training and building AI to improve accuracy is an issue for the future. Also, it is needed to evaluate the impact on physicians’ diagnosis or clinical outcomes when the AI-based diagnosis is introduced.
Figure 1. Flow chart of the study design. CNN, Convolutional neural network
Read the full article online.
The information presented in Endoscopedia reflects the opinions of the authors and does not represent the position of the American Society for Gastrointestinal Endoscopy (ASGE). ASGE expressly disclaims any warranties or guarantees, expressed or implied, and is not liable for damages of any kind in connection with the material, information, or procedures set forth.