<p>Segmentation of BTs is an important area in medical imaging, and the latest advancement of deep learning algorithms offers great promise to further increase the accuracy and effectiveness of tumor delineation and detection. This proposed model for deep learning in the segmentation of medical pictures will combine pre-processing methods with a hybrid optimization strategy as well as a modified U-Net + + architecture with both spatial and temporal attention mechanisms. In this work, pre-processing includes applying a Gaussian filter to reduce noise and histogram equalization to improve contrast and guarantee better feature extraction later on. A modified U-Net + + _spatial temporal Attention (U-Net + + _SpaTempAtt) is used in the segmentation model to efficiently capture temporal and spatial relationships in the image features. The architecture has the encoder-decoder structure which includes skip connections that help feature maps move from the encoder to the decoder. The spatial attention layer is applied at the bottleneck of the encoder to focus on the most salient characteristics. Further to fine-tune the feature maps, the temporal attention layer is utilized after the decoder. In optimizing the parameters of the model, crayfish-sailfish hybrid optimizer has been utilized, as this optimizer offers the benefit of local exploitation by CFO as well as SFO ability to make worldwide exploration. This hybrid approach enhances the precision and efficiency of the segmentation results.</p>