Transfer Learning for MRI image reconstruction: Enhancing model performance with pretrained networks

These Neurologists and radiologists have the important responsibility of finding brain tumors early on. Brain tumor detection and segmentation using Magnetic Resonance Imaging (MRI) data is complex and error-prone when done manually.  That is why an automated approach to detecting brain tumors is so important for early detection. This study introduces a new approach to brain tumor classification that makes use of DL and the MobileNet model. The Brain Tumor MRI Dataset follows preprocessing routines by resizing images along with converting them to grayscale before performing normalization. MobileNet implements depth-wise separable convolutions during training, which utilizes categorical cross-entropy loss for performance evaluation through accuracy, precision, recall and F1-score methods. EfficientNetV2-S reaches 99.23% accuracy while maintaining 99.42% precision, 99.34% recall, and 99.35% F1-score, which exceeds the performance of VGG19 (96%) and EfficientNetV2-S (96.19%). The model presents high precision (99.42%) and recall (99.34%) metrics, which support its ability to detect positive cases effectively. MobileNet demonstrates its value as both a trustworthy technology and efficient system for brain tumor diagnostic systems used in medical practice.