GAA-Enhanced UNet++ Model for Brain Tumor Segmentation

Advanced techniques in medical image segmentation often leverage deep learning methodologies. UNet and its variants plays a major role in image segmentation of medical images. UNet based networks have so many limitations that there needs an improvement in the segmentation performance. The proposed UNet++ architecture integrates dense skip connections and nested convolutions to enhance feature representation and improve accuracy of segmentation. Its superior performance and versatility make it a precious tool for assisting clinicians in detection of diseases and planning for treatment, ultimately improving patient outcomes in neuro-oncology. In this paper, the new method called GAAUNET++ (Gated Axial Attention UNet++) have been proposed with enhancements like 1. The inclusion of the Axial Attention gate aims to enhance performance while also capturing global features of the image. 2. Different metrics like precision, dice coefficient, recall and IoU have been used. The results shows that the proposed approach achieves finer performance in segmentation and accuracy also has been improved.