Loading...
Automated Detection of Intestinal Pathology in a Mouse Model of Food Allergy via Deep Learning
Ruppert, Maximilian Matthias
Ruppert, Maximilian Matthias
Type
Tesis de maestría
Citations
Altmetric:
Abstract
"Food allergy triggers an immune-mediated inflammatory response in the gastrointestinal tract, and murine models enable its controlled study. However, histological assessment of intestinal tissue remains manual, and differences between naïve and sensitized tissue are difficult to identify consistently. No prior application of deep learning to food allergy intestinal histopathology was identified in the literature. This work develops a weakly supervised deep learning pipeline to classify food-allergy-induced pathology in murine swiss-roll whole-slide images without patch-level annotations. The dataset comprises 171 H&E-stained whole-slide images from 112 mice. After tissue segmentation and patch extraction, features from four foundation models pretrained on human histopathology are combined with three multiple instance learning architectures and evaluated using nested and group-stratified cross-validation. In nested cross-validation, three of the four encoders achieve AUC above 0.94, with the highest mean AUC of 0.974. In group-stratified cross-validation, pooled AUC values remain above 0.95 for the three top-performing encoders. However, one of three experimental batches shows below-chance accuracy across all encoders, indicating a batch effect or a possible labeling issue. Contribution heatmaps provide spatial interpretability but have not been pathologically validated. Overall, the pipeline can classify food-allergy-induced murine intestinal pathology in this dataset, though the batch effect limits generalizability".
Description
Date
2026-06-04
Journal Title
Journal ISSN
Volume Title
Publisher
Instituto Tecnológico de Buenos Aires (ITBA)
Collections
Files
Research Projects
Organizational Units
Journal Issue
Keywords
FOOD ALLERGY, MOUSE MODEL, DEEP LEARNING, WEAKLY SUPERVISED LEARNING, MULTIPLE INSTANCE LEARNING, FOUNDATION MODELS, HISTOPATHOLOGY, WHOLE-SLIDE IMAGE