Open Access

ARTICLE

Optimizing Capsule Endoscopy via (SHAP) Perturbations and Optimal Data Distribution for Pancreatic Disease Classification

Anurag Sinha^1,*, Pranto Halder², Aditya Pandey³, Avi Mohan Kumar Shukla⁴, Shravan Kumar⁵, Ashutosh Rastogi⁶, Asima Akter Chowdhury⁶, Suryansh Rai⁷, Sagar Singh⁸

1 School of Computing and Information Science, IGNOU, New Delhi, India
2 Department of Computer Science and Engineering, Faculty of Science and Engineering, Teesta University, Rangpur, Bangladesh
3 Department of Electronics and Computer Engineering, National Institute of Advanced Manufacturing Technology, Ranchi, India
4 LKCRMS, Ranchi, India
5 Program Coordinator, Vanderbilt University, Nashville, TN, USA
6 Stagity Solutions LLC, Farmers Branch, TX, USA
7 Department of Science, Dev Indrawati Mahavidyalaya (Dr. Ram Manohar Lohia Awadh University), Tanda, Ambedkar Nagar, Uttar Pradesh, India
8 CSE (AI&ML), ABES Engineering College, Ghaziabad, India

* Corresponding Author: Anurag Sinha. Email:

Journal of Intelligent Medicine and Healthcare 2026, 4, 125-153. https://doi.org/10.32604/jimh.2026.075373

Received 30 October 2025; Accepted 04 May 2026; Issue published 14 September 2026

Abstract

Accurate classification of pancreatic endocrinogenesis-related abnormalities in capsule endoscopy images remains challenging because of class imbalance, high intra-class variability, and limited annotated data. This study proposes an optimal data distribution framework based on Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) perturbations to enhance automated classification performance. The workflow combines perturbation-guided data augmentation, feature-importance analysis, and deep-learning classifiers, including convolutional neural networks (CNNs), U-Net, and You Only Look Once (YOLO). The approach is evaluated on the Kvasir-CapsuleSeg dataset using accuracy, macro F1-score, balanced area under the receiver operating characteristic curve (AUC), and confusion-matrix analysis. Results indicate that perturbation-guided augmentation improves robustness and class balance, with consistent gains across models and particularly strong performance for underrepresented classes. These findings suggest that explainability-driven data optimization can improve automated capsule endoscopy analysis and support more interpretable pancreatic disease classification.

---

Keywords

---