Enhanced Brain Tumor Segmentation in MRI Scans Using ResNet-150

Accurate diagnosis and treatment planning depend on the vital work of brain tumor segmentation from MRI data in medical imaging to increase life expectancy. Trained on a whole brain tumor dataset comprising four different classes-no tumor, glioma, meningioma, and pituitary-this work proposes an advanced segmentation method utilizing a ResNet-150 model. With a training accuracy of 99.7%, validation accuracy of 98.6%, and test accuracy of 99.1%, the model attained amazing performance metrics, thereby stressing the generalizability and resilience over several data subsets. Using Dice scores and Intersection over Union (IoU) measures, performance was assessed still more. High accuracy in accurately recognizing and segmenting tumor areas was shown by the training and validation Dice scores of 99.8 and 99, respectively. Furthermore, underlining the capacity to precisely overlap the expected segmentation with the actual tumor areas, the training and validation IoU were noted at 99.3 and 97.9 respectively. Known for its deep layers and residual connections, the ResNet-150 architecture was essential in capturing complex data from the MRI scans, hence supporting the great accuracy and precision of the segmentation work. This work highlights how well deep learning models might improve the accuracy of brain tumor segmentation, therefore opening the path for more dependable and automated diagnosis instruments in clinical environments and decrease death rate. Expanding the dataset and investigating transfer learning approaches will be the main priorities of next work in order to raise the performance and usability of the model in practical medical situations.