Automatically detecting Crohn's disease and Ulcerative Colitis from endoscopic imaging

Area of Science:

• Medical Artificial Intelligence
• Gastroenterology
• Computational Pathology

Background:

• The SI-CURA project focuses on developing AI for patient management and therapeutic follow-up in Ulcerative Colitis.
• The initiative aims to differentiate various pathologies, including Inflammatory Bowel Disease (IBD), using endoscopic imaging.

Purpose of the Study:

• To develop a deep learning (DL) prototype for disease pattern identification in endoscopic images.
• To perform three binary classification tasks: pathological vs. healthy, Ulcerative Colitis vs. Crohn's disease, and Ulcerative Colitis vs. healthy.

Main Methods:

• A deep learning model was trained and tested on endoscopic images.
• The model was evaluated on its ability to perform three distinct classification tasks.

Main Results:

• The DL model achieved a Matthews correlation coefficient (MCC) > 0.9 for distinguishing pathological from healthy samples (P vs N) and Ulcerative Colitis from healthy samples (UC vs N).
• The model achieved an MCC > 0.6 for differentiating Ulcerative Colitis from Crohn's disease samples (UC vs CD).

Conclusions:

• The DL model demonstrates high efficacy in discriminating between pathological and healthy samples.
• The model also effectively differentiates between subtypes of IBD, showing potential as an endoscopy-based diagnostic decision support tool.