Deep Image Segmentation Using Explainable Attention Mechanisms: Applications in Biomedical Imaging

Correct and discernible segmentation of an image is an important part of biomedical imaging, especially when anatomical structures and pathological regions are identified. Although deep learning architectures, like U-Net and its variants, have performed well, their lack of interpretability, transparency and contextual reasoning has limited their clinical adoption. This research helps to overcome the most important issue of high-performance medical segmentation by introducing a new explainable framework that combines convolutional encoders with transformer-based decoders and dual attention mechanisms. There are three aspects of this work, based on the following objectives. (1) To boost the performance of segmentation by means of hybrid local-global feature modelling; (2) To bring about clarity through visual explanation tools; and (3) To conduct clinical viability checks through expert assessments. The architecture proposed includes CBAM for fine spatial and channel attention and combines Grad-CAM++ and SHAP for local and global explainability. S