TriVerseNet: Multi-Branch ResNet Framework for Enhanced Endoscopic Lesion Classification via Edge and Texture Integration

The accurate classification of diseases using computer vision is critical for early diagnosis and effective treatment planning. In this research, a novel deep learning (DL) framework for accurate classification of gastrointestinal (GI) lesions in endoscopic images is presented. The proposed model, TriVerseNet, is designed to improve lesion classification by integrating complementary visual features through a multibranch architecture. It utilizes three parallel branches where each branch is based on a customized ResNet34 backbone truncated at mid-depth. One branch analyzes the standard RGB input, another focuses on edge information extracted via Canny Edge and the third captures textural cues using a Laplacian Operator. The resulting low-dimensional feature embeddings are concatenated and processed by fully connected layers for final multi-class classification. The proposed model is evaluated on the Kvasir v2 dataset, covering eight lesion types by achieving an accuracy of 92.63 %, an AUC-ROC score of 0.9944, and a high MCC of 91.57 %, proving its capability. The proposed model results demonstrate that combining edge, texture, and appearance information significantly improves model robustness and classification accuracy in challenging medical image scenarios. The proposed approach offers a promising direction for computer-aided diagnosis (CAD) systems in GI endoscopy.