Interpretable deep learning for enhanced multi-class classification of gastrointestinal endoscopic images

Gastrointestinal (GI) endoscopy serves as a vital tool for assessing the GI tract and diagnosing related disorders. Recent progress in deep learning has shown significant improvements in identifying anomalies using sophisticated models and data augmentation strategies. This study introduces an enhanced approach to improve classification accuracy using 8,000 labeled endoscopic images from the Kvasir dataset, categorized into eight distinct classes. Leveraging EfficientNetB3 as the backbone, our proposed architecture eliminates the reliance on data augmentation while maintaining moderate model complexity. Our model achieves a test accuracy of 94.25%, alongside precision and recall of 94.29% and 94.24%, respectively. Furthermore, Local Interpretable Model-agnostic Explanation (LIME) saliency maps are employed to enhance interpretability by highlighting critical regions in the images that influence model predictions. To facilitate real-world usability, a user-friendly interface was developed using Gradio, enabling users to upload images, generate predictions, view confidence levels, and maintain a history of past results. This work underscores the importance of integrating high classification accuracy, interpretability, and accessibility in advancing medical imaging applications.