Predicting brain tumours through transfer learning and optimized hyperparameters using the modified whale’s algorithm

Brain tumour classification from magnetic resonance imaging (MRI) is essential for timely diagnosis and treatment planning; however, deep learning models often require extensive training and careful hyperparameter tuning. This study proposes a domain-specific transfer learning framework that integrates the ResNet18 architecture with a Modified Whale Optimization Algorithm (MWOA) for automated hyperparameter optimization. Unlike conventional transfer learning approaches based on ImageNet pretraining, the proposed framework progressively constructs a domain-specific pretrained model using multiple publicly available brain tumour MRI datasets comprising more than 10,000 images. The MWOA optimizes the learning rate, batch size, number of epochs, weight decay, dropout, gradient clipping, and early stopping parameters before final model training. Performance was evaluated using independent test partitions, including accuracy, precision, recall, F1 Score, confusion matrix, and one-versus-rest ROC AUC. Experimental results demonstrated improved classification performance and substantially reduced retraining time compared with conventional ImageNet-based transfer learning. Although the framework achieved excellent performance on the benchmark datasets evaluated, further validation using independent multi-centre clinical datasets together with repeated experimental trials and statistical validation is required before clinical deployment.