Advancing Brain Tumour Detection: Transfer Learning-Based Approach Fused with Squeeze-and-Excitation (SE) Attention Mechanism in Computer Vision

Brain tumour (BT) is one of the most severe forms of disease for humans. Computer vision can be applied to diagnose it at an early stage to make optimal decisions for patients and pathologists. Early diagnosis can help with prior treatment and reduce the mortality rate. In this research, we proposed a transfer learning (TL) approach fused with a squeeze-and-excitation (SE) attention mechanism to accurately diagnose brain tumours on a brain tumour MRI dataset. It incorporates one-hot encoding in the image preprocessing step. Our method incorporates the EfficientNetV2M model integrated with SE attention. In terms of accuracy, precision, recall and AUC, the approach we used outperformed all current models. In addition, this approach has been benchmarked with the previous seven state-of-the-art (SOTA) models on the same dataset. Our proposed techniques obtained the best results for both the validation and testing datasets. On the validation data of the MRI brain tumour, we achieved the highest results, with an accuracy of 95.92%, precision of 95.89%, recall of 95.24% and AUC of 99.00%.