Deep learning-based brain tumor detection via transfer learning

Doctors rely on magnetic resonance imaging (MRI) to make critical decisions about brain tumors. Manual reviewing of MRI images might involve mistakes and calls for a high level of experience. For this reason, our model is a framework using deep learning and transfer learning to help identify brain tumors automatically. The main goal of the research is to develop a classification system that effectively separates pictures of brain tumors from non-tumor images by using pre-trained convolutional neural networks (CNNs). The system uses ResNet50 and ResNet18 models that were initially trained on ImageNet and then trains them again with a handpicked set of 3,929 labeled MRI images. The models are modified through transfer learning to do binary classification which cuts down on training time and also boosts their performance with limited data. MRI data is prepared and augmented before feeding it to the model, so that it fits the training better and is usable outside the dataset. Experimental findings show that ResNet50 has much higher accuracy (98.6%) than ResNet18 (96.4%) which confirms that more layers help the network learn better differences between images. Additional performance measures such as precision, recall and F1-score demonstrate that the approach is reliable. It proves that applying deep residual networks is possible in clinical work for brain tumor detection. Thanks to classified results from automation, there is less risk of variations in diagnosis and better opportunities for radiologists to spot early tumor signs. Now, the frame-work will be improved to help diagnose a wider range of tumors and include more transparent ways of working to facilitate clinical use.