Related Experiment Video
Updated: Apr 25, 2026
09:21
A Novel Dual-Modal Deep Learning Approach for Real-Time Removal of Hepatic Fluorescence in Indocyanine Green-Guided Laparoscopic Cholecystectomy
Published on: April 17, 2026
29
An explainable hybrid deep-learning and machine learning framework for automatic coeliac disease detection from
Souaad Hamza-Cherif1, Adil Gaouar1, Zineb Aziza El Aouaber1
1Biomedical Engineering Laboratory, Abou Bekr Belkaid University, Tlemcen, Algeria.
Health Information Science and Systems
|April 24, 2026
Summary
An AI framework accurately detects coeliac disease (CD) from endoscopy images, improving diagnosis. This explainable AI offers a data-efficient solution for gastroenterology, aiding early detection and patient care.
Area of Science:
- Artificial Intelligence in Medicine
- Gastroenterology
- Medical Imaging Analysis
Background:
- Coeliac disease (CD) diagnosis is challenging due to reliance on invasive biopsies and subjective interpretation of endoscopic findings.
- Current diagnostic methods for CD are limited by invasiveness and the need for expert analysis of subtle endoscopic signs.
- There is a need for automated, data-efficient, and explainable tools to improve CD detection from duodenal endoscopy images.
Purpose of the Study:
- To develop an artificial intelligence (AI) framework for automatic coeliac disease (CD) detection using duodenal endoscopy images.
- The study aims to create a data-efficient and explainable AI model for robust CD identification.
- To enhance diagnostic accuracy and accessibility for coeliac disease through advanced imaging analysis.
Main Methods:
- A dataset of 164 expert-annotated duodenal endoscopic images (89 normal, 75 CD) was utilized.
- Two AI pipelines were evaluated: end-to-end fine-tuning and a hybrid approach with feature extraction, MixUp augmentation, and ML classifiers.
- Otsu-based segmentation, cross-validation, ablation studies, runtime benchmarking, and interpretability methods (LIME, Grad-CAM) were employed for model assessment.
Main Results:
- The hybrid ConvNeXt-Tiny + RBF SVM model achieved the highest performance (87.39% accuracy, 87.87% precision, 87.39% recall, 87.28% F1-score).
- This hybrid model significantly outperformed the best end-to-end model (ResNet-18, 85.19% accuracy; p=0.0084).
- Explainability techniques highlighted key mucosal patterns indicative of villous atrophy and scalloping, crucial for CD diagnosis.
Conclusions:
- The developed explainable hybrid AI framework offers robust and interpretable coeliac disease detection, particularly effective in low-data scenarios.
- Its modular design and computational efficiency make it suitable for integration into clinical gastroenterology workflows.
- This AI approach shows promise for computer-assisted diagnosis, potentially reducing reliance on invasive procedures and improving patient outcomes.
