@Article{cmes.2026.084222,
AUTHOR = {Areeba Gul, Muhammad Ramzan, Romana Aziz, Qaiser Abbas, Ala Saleh Alluhaidan, Summair Raza, Mahwish Ilyas},
TITLE = {GastroNetV4: An Explainable AI-Based Hybrid Framework for Gastrointestinal Diseases Detection Using Endoscopic Images},
JOURNAL = {Computer Modeling in Engineering \& Sciences},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/CMES/online/detail/28122},
ISSN = {1526-1506},
ABSTRACT = {Gastrointestinal diseases (GI) are serious diseases that affect people of all ages. Early and accurate diagnosis helps reduce complications and the subsequent impact on patients. Precise diagnoses by endoscopy are important for reducing complications and mortality rates. Manual interpretation of endoscopic images is time-consuming, highly dependent on the specialist’s clinical judgment, and subject to variability in multi-class classification tasks. To overcome these limitations, this study proposes GastroNetV4, an explainable hybrid deep learning model for multi-class classification of gastrointestinal diseases and a urinary tract-related class included in the endoscopic dataset used in this study.GastroNetV4 employs an EfficientNet-B4 backbone with a tailored Convolutional Neural Network (CNN) head to extract high-quality discriminative features. Additionally, the proposed approach also integrates Explainable AI (XAI) methods to make AI decisions understandable and transparent. The proposed approach was compared against other publicly available pre-trained models for multi-class classification of GI diseases and a urinary-tract condition. t-Distributed Stochastic Neighbor Embedding (t-SNE) and predictive entropy analyses were used to visualize internal latent features and evaluate the model’s uncertainty calibration. Experiments have shown that the proposed approach outperformed other models across various evaluation metrics. The proposed approach achieved the highest accuracy of 98.25% among all evaluated models, showing the effectiveness of the proposed enhancements over the baseline models. The study analyzed uncertainty by showing that correct predictions are associated with low entropy, whereas incorrect predictions are associated with high entropy. By achieving high accuracy, interpretability, and well-calibrated predictions, GastroNetV4 will help develop clinical decision-support systems for the automatic diagnosis of gastrointestinal diseases and other abnormalities.},
DOI = {10.32604/cmes.2026.084222}
}