Improving Brain Tumor Detection and Segmentation with DeepLabV3 and ResNet-50

Accurate diagnosis and treatment planning rely on brain tumor segmentation from MRI data in medical imaging, critical for increasing life expectancy. This study proposes an advanced segmentation method utilizing the DeepLabV3 model with ResNet-50 architecture. Trained on a whole brain tumor dataset comprising four different classes— no tumor, glioma, meningioma, and pituitary—this model achieved outstanding performance metrics: a training accuracy of 99.7%, validation accuracy of 99.7%, and test accuracy of 99.1%. Performance was further evaluated using Dice scores and Intersection over Union (IoU) measures, with training and validation Dice scores of 0.936 and 0.893, respectively, and training and validation IoU of 0.736 and 0.713, respectively. Known for its efficiency and robustness, the DeepLabV3 with ResNet-50 architecture was essential in capturing complex data from MRI scans, supporting the high accuracy and precision of this segmentation work. This study highlights the potential of deep learning models to enhance brain tumor segmentation accuracy, paving the way for more reliable and automated diagnostic tools in clinical settings leading to a healthy life. Future work will focus on expanding the dataset and exploring transfer learning approaches to further improve performance and practical applicability.