Using deep learning for predicting cleansing quality of colon capsule endoscopy images

TL;DR

This work investigates predicting cleansing quality in colon capsule endoscopy images using a ResNet-18 classifier trained with stratified $K$-fold cross-validation and enhanced by iterative structured pruning to achieve up to $88\%$ cross-validation accuracy with $79\%$ sparsity. The authors assess explainability with ROAD-based metrics across CAM variants and calibrate predictions on an external dataset using a variant of adaptive temperature scaling (HnLTS). Results show pruning can substantially reduce parameters while preserving accuracy, though explainability metrics vary across cleansing categories and ROAD has limitations. Calibration with HnLTS improves external-test reliability, especially for higher-capacity pruned models, supporting deployment in resource-constrained clinical settings.

Abstract

In this study, we explore the application of deep learning techniques for predicting cleansing quality in colon capsule endoscopy (CCE) images. Using a dataset of 500 images labeled by 14 clinicians on the Leighton-Rex scale (Poor, Fair, Good, and Excellent), a ResNet-18 model was trained for classification, leveraging stratified K-fold cross-validation to ensure robust performance. To optimize the model, structured pruning techniques were applied iteratively, achieving significant sparsity while maintaining high accuracy. Explainability of the pruned model was evaluated using Grad-CAM, Grad-CAM++, Eigen-CAM, Ablation-CAM, and Random-CAM, with the ROAD method employed for consistent evaluation. Our results indicate that for a pruned model, we can achieve a cross-validation accuracy of 88% with 79% sparsity, demonstrating the effectiveness of pruning in improving efficiency from 84% without compromising performance. We also highlight the challenges of evaluating cleansing quality of CCE images, emphasize the importance of explainability in clinical applications, and discuss the challenges associated with using the ROAD method for our task. Finally, we employ a variant of adaptive temperature scaling to calibrate the pruned models for an external dataset.

Figures (16)

• Figure 1: Example images of the different Leighton–Rex scale classes
• Figure 2: The classic three-step framework—train, prune, and fine-tune from the study by Han2015.
• Figure 3: Scores obtained for two images (titled Original) belonging to the "Poor" category of Leighton-Rex, and visualizations and scores of the different metrics.
• Figure 4: Mean cross-validation accuracy and $\pm$ one standard deviation across 10 folds versus pruning steps.
• Figure 5: Overall sparsity and layer sparsity for pruning steps.