Trustseg-Net: Uncertainty-Calibrated Medical Image Segmentation with Boundary-Aware Confidence Estimation

Medical image segmentation plays an important role in computer-aided diagnosis by identifying anatomical structures and pathological regions from medical images. Deep learning models have achieved significant improvements in segmentation accuracy, but they may produce uncertain predictions and inaccurate boundaries, particularly in low-contrast and irregular regions. This study proposes TRUSTSEG-NET, an uncertainty-calibrated medical image segmentation framework with boundary-aware confidence estimation. The proposed framework combines deep learning-based segmentation, boundary-aware learning, uncertainty estimation, and confidence calibration to improve segmentation accuracy and prediction reliability. Medical images are preprocessed using resizing, normalization, and augmentation techniques before model training. The framework generates pixel-level segmentation masks while estimating uncertainty in difficult regions. Performance is evaluated using Dice Similarity Coefficient, Intersection over Union, Precision, Recall, and Boundary F1-Score. The proposed approach provides both segmentation results and confidence information, supporting reliable medical image analysis and computer-aided diagnostic applications.