Scalable quantum non-local neural network optimised with the tyrannosaurus algorithm for brain tumour detection using MRI images

Early and reliable detection of brain tumours using magnetic resonance imaging (MRI) is essential for timely diagnosis and effective treatment planning. However, automated tumour detection remains challenging due to tumour heterogeneity, complex brain anatomy, image noise, and the limited ability of many deep learning models to capture both global and fine-grained features. Existing approaches also often suffer from high computational complexity and limited generalisability across datasets. This study aims to develop an efficient and robust automated framework for accurate brain tumour detection from MRI images while addressing limitations related to feature representation, computational efficiency, and model generalisability. A novel framework integrating advanced preprocessing techniques, accurate tumour segmentation, hierarchical feature extraction, and optimised classification is proposed. The model was evaluated using two publicly available benchmark datasets, BraTS 2018 and Figshare, comprising multi-class brain tumour MRI images with different tumour types and grades. The proposed framework achieved superior performance, with classification accuracy reaching up to 99.8%. Comparative analysis demonstrated improvements in precision, recall, and F1-score over existing state-of-the-art methods. The model also exhibited enhanced feature representation capabilities and reduced computational errors when processing complex tumour structures. The proposed approach provides a reliable and efficient computer-aided diagnostic tool for brain tumour detection. Its high accuracy and robustness across datasets make it a promising tool for assisting clinicians in early diagnosis and informed decision-making, thereby potentially improving clinical outcomes.