PhD defence Ayla Thijssen

Keywords: Artificial intelligence, Colorectal carcinoma, Colorectal polyps, Endoscopy

"Artificial intelligence for optical diagnosis of colorectal polyps and carcinoma: Exploring diagnostic performance, explainability, and trust"

Colorectal carcinoma (CRC) arises from colorectal polyps. Endoscopists can characterize different types of colorectal polyps during endoscopy through optical diagnosis. If an early stage CRC is recognized as such, this enables organ sparing treatment instead of surgical resection. However, in general, the quality of optical diagnosis is insufficient.

This thesis investigated several applications of artificial intelligence (AI) to improve optical diagnosis of colorectal polyps and carcinoma. Studies in this thesis described two forms of explainable AI (textual and visual) that could help endoscopists understand why AI provides a certain diagnosis. Additionally, the first European AI system with the ability to diagnose CRC on endoscopic images was introduced and compared to optical diagnosis by 78 international endoscopists. This showed promising diagnostic performance of AI. Evaluation of human-AI interaction showed that trust in AI needs to be improved. Results of this thesis contribute to future developments of AI in gastrointestinal endoscopy, aiming to improve patient care.

Click here for the live stream.