<![CDATA[Brain tumor segmentation from MRI is crucial for diagnosis, treatment planning, and patient monitoring, yet many automated segmentation models struggle to capture global contextual information while preserving fine spatial details, leading to inaccurate delineation of complex tumor boundaries and reduced clinical reliability. Manual segmentation by radiologists is also constrained by observer variability, prolonged analysis time, and reliance on individual expertise. To address these limitations, this study proposes Hybrid DeepLabV3+, an automated segmentation model that combines DeepLabV3+ multi-scale context modeling through Atrous Spatial Pyramid Pooling (ASPP) with spatial-detail preservation via U-Net skip connections. The objective is to develop and rigorously evaluate a more accurate and generalizable brain tumor segmentation model compared with U-Net, DeepLabV3+, and EfficientNet baselines. The model is trained on the BraTS dataset using T2-weighted MRI and evaluated using Dice score, intersection over union (IoU), precision, recall, and accuracy. Experimental results show that the proposed model achieves a Dice score of 0.785, IoU of 0.740, precision of 0.796, recall of 0.985, and accuracy of 0.991, surpassing the comparison models. These results indicate that Hybrid DeepLabV3+ can identify and map tumor regions with high fidelity, supporting its potential integration into clinical decision-support systems.]]>
Copyrights © 2026
