Improving optical diagnosis of colorectal polyps: the impact of artificial intelligence

Colorectal cancer develops from colorectal polyps. Colorectal polyps smaller than 5 mm represent a significant proportion of the polyps detected during a colonoscopy, even though they have a negligible risk of containing colorectal cancer. This thesis investigated how optical diagnosis, the assessment of the histology of colorectal polyps during a colonoscopy, could be improved. Specifically, the role of artificial intelligence (AI) was investigated.

To improve the accuracy of optical diagnosis, a computer-aided diagnosis (CADx) system based on artificial intelligence was developed. The accuracy of this CADx system was highest when multiple imaging techniques based on different colors of light were combined. Extending the CADx system’s diagnosis with a confidence value, also led to higher an accuracy.

Patients and gastroenterologists expected the quality of care to be increased with the use of artificial intelligence, but were concerned about losing personal contact. CADx-supported optical diagnosis was cost-effective compared to optical diagnosis without artificial intelligence in a model analysis.