Abstract
INTRODUCTION: Medical image segmentation is essential for analyzing medical data, improving diagnostics, treatment planning, and research. However, current methods struggle with different imaging types, poor generalization, and rare structure detection. METHODS: To address these issues, we propose MedFusion-TransNet, a novel multi-modal fusion approach utilizing transformer-based architectures. By integrating multi-scale feature encoding, attention mechanisms, and dynamic optimization, our method significantly enhances segmentation precision. Our method uses the Context-Aware Segmentation Network (CASNet) and Dynamic Region-Guided Optimization (DRGO) to enhance segmentation by focusing on key anatomical areas. RESULTS: These innovations tackle challenges like imbalanced datasets, boundary delineation, and multi-modal complexity. Validation on benchmark datasets demonstrates substantial improvements in accuracy, robustness, and boundary precision, marking a significant step forward in segmentation technologies. DISCUSSION: MedFusion-TransNet offers a transformative tool for advancing the quality and reliability of medical image analysis across diverse clinical applications.