Attention-UNet with Residual Connections for Enhanced Kidney Tumor Segmentation in Multi-Modal Imaging

Renal tumor segmentation with multi-modal medical imaging is still very difficult because to the intricacy of organ architecture, variations in tumor forms, and noise in different modalities. To overcome this. an improved segmentation scheme based on Attention-UNet with Residual Connections is suggested. While residual connections enable improved feature replication and gradient flow, the inclusion of attention gates enables the model to focus on pertinent regions, improving tumor localization. Using both CT and MRI modalities, this mixed-modality architecture is trained and verified on publically accessible datasets like KiTS and BraTS The suggested model outperforms U-Net (DSC 85.12%), Residual U-Net (DSC 87.26%), and Attention U-Net (DSC 88.43%) by achieving a Dice Similarity Coefficient (DSC) of 90.78%, IoU of 84.91%, Precision of 91.65%, Recall of90.12%, and F1-Score of90.88%. The accuracy and reliability of automated kidney tumor segmentation are significantly increased by this technique, which has promise for early identification and treatment planning.

Hali tarjima qilinmagan