Ulcerative Colitis, LAIR1 and TOX2 Expression, and Colorectal Cancer Deep Learning Image Classification Using...

Area of Science:

• Oncology
• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Ulcerative colitis (UC) is a chronic inflammatory bowel disease increasing colorectal cancer (CRC) risk.
• Accurate histological classification of UC and CRC is crucial for patient management.

Purpose of the Study:

• To classify histological images of UC, normal colon, and CRC using deep learning (CNNs).
• To evaluate the utility of AI in differentiating disease subtypes and identifying immune markers.

Main Methods:

• A ResNet-18-based CNN was trained on H&E stained images from UC, normal colon, and CRC cases.
• Image classification performance was assessed using training, validation, and test datasets.
• Explainable AI (Grad-CAM) and immunohistochemistry (LAIR1, TOX2) were employed to analyze image features and immune microenvironment.

Main Results:

• The CNN achieved high classification accuracy: 99.1% for UC, 99.8% for CRC, and 99.1% for normal colon.
• AI models differentiated between steroid-requiring and mesalazine-responsive UC based on H&E, LAIR1, and TOX2 staining.
• LAIR1 expression was higher in steroid-requiring UC, while TOX2 expression was lower in isolated lymphoid follicles.

Conclusions:

• CNNs demonstrate high efficacy for classifying UC and CRC from histological images.
• LAIR1 and TOX2 are significant immuno-oncology markers in ulcerative colitis, aiding in disease characterization.