Spirochetosis detection in colon histopathology images via fine-tuning and boosting techniques using foundation models

Agata Polejowska, Fazael Ayatollahi, Ayse Selcen Oguz Erdogan, Francesco Ciompi, Annemarie Boleij

Keywords: Human intestinal spirochetosis, bacteria, histopathology, foundation models, transfer learning, feature extraction, XGBoost

Abstract:

Spirochetes are bacteria that can be found on the boundaries of colon epithelial tissue, causing several diseases ranging from spirochetosis, inflammatory bowel disease, to cancer. Despite their relevance, spirochetes often remain undetected in histological analysis.
We propose the first computational pathology approach to characterize spirochetes, leveraging prior spatial knowledge to detect spirochetes in whole-slide images of colon polyps and biopsies, and differentiate these bacteria as belonging to normal or abnormal tissue.
We focus on transfer learning by fine-tuning state-of-the-art computational pathology foundation models and by training an additional XGBoost classifier on downstream tasks.

Submission Number: 99