ARCHIVES

Brain tumor is one of the highly critical neurological problems which needs early and precise diagnosis to achieve better survival rate and prognosis. This explains the importance of intelligent computer-aided diagnostic systems. MRI Manual interpretation is time-consumingas it depends on radiological expertise. This study presents novel DL based framework named Smart Neuro Vision. It is for automated brain tumor segmentation and classification using U-Net and ResNet-101 architecture. The proposed scheme is divided into four distinct main blocks namely image preprocessing, tumor segmentation, tumor classification and performance evaluation. Firstly the MRI images are preprocessed to get better quality images. Then a U-net model segments the tumor regions from MRI scans. The segmented tumor images are then classified to these four categories namely Glioma, Meningioma, Pituitary Tumor and No Tumor using ResNet-101. Experiments were carried out using an open source brain MRI dataset provided by Kaggle with four tumor classes. The segmentation model attained Dice Coefficient of 96.8%, IoU score of 94.5% and segmentation accuracy of 97.2%. Moreover, the classification model achieved 98.4%accuracy. This indicates good detection performance of tumor and reduced overfitting. The obtained results validate proposed framework of combining U-Net segmentation with ResNet-101 classification for automated brain tumor diagnosis, which offers a powerful, reliable, and efficient solution. The proposed system could aid radiologists in making clinical decisions and has great potential to be implemented in actual healthcare settings.