Adaptive Edge-Guided Segmentation with Biftransnet for Gastrointestinal Tract Image Segmentation

Objectives: The primary goal of this research is to improve the segmentation accuracy of Gastrointestinal (GI) tract images, addressing the challenges posed by complex anatomical structures, noise artifacts, and varying imaging modalities. The study aims to overcome the limitations of traditional techniques like thresholding and region growing, as well as mitigate the drawbacks of current deep learning-based models, such as high computational costs and overfitting. Methods: UW-Madison GI Tract Images are collected from the Kaggle repository. the proposed method introduces an Adaptive Edge-Guided Segmentation (AEGS) framework that integrates BifTransNet with enhanced edge detection and region-growing techniques. The model uses the batch size, epochs, and learning rate as parameters. This approach incorporates attention mechanisms, advanced boundary delineation, and regularization techniques to reduce overfitting while improving segmentation accuracy. The model is trained with regularization to optimize its performance, making it more robust and reliable. Findings: The experimental results show significant improvements, with the proposed AEGS framework achieving a segmentation accuracy of 98.6% and a training loss of 0.12. This marks a substantial enhancement in both robustness and reliability compared to traditional methods and existing deep-learning models for GI tract image segmentation. Novelty: The key innovation lies in the hybrid approach of combining deep learning (BifTransNet) with improved edge detection and region-growing techniques. This hybridization not only enhances segmentation performance but also addresses common issues such as overfitting and computational costs, setting a new benchmark in GI tract image segmentation. Keywords: Gastrointestinal Tract, Image Segmentation, BifTransNet, Deep Learning, Overfitting