Brain Tumor Detection and Classification by Using UU-Net with Dual Attention Mechanism

The study investigates the recognition, classification, and prediction of brain tumors via the use of a deep learning framework that integrates sophisticated segmentation, pre-processing, and hybrid modeling approaches. Input quality is maximized by pre-processing procedures such picture loading, scaling, and Gaussian smoothing for noise reduction. The suggested UU-Net with a Dual Attention Mechanism improves the network's emphasis on important areas and characteristics for segmentation by combining spatial and channel attention, extending the U-Net design. The model improves significantly in the tumor segmentation accuracy when tested on brain MRI datasets, showing a high F1-score of 95.00%, precision of 94.70%, recall of 95.30%, and AUC-ROC of 0.96, outperforming the baseline U-Net and Attention U-Net models. By providing accurate identification of the tumor in the middle of complicated anatomical components, the model's resilience and clinical significance are proven through the examination of the training progression and segmentation results.