Automated Models for the Classification of Magnetic Resonance Brain Tumour Images

Brain tumours are the second largest cause of cancer death in children under 15 and young adults until age 34. Also, among people over 65, these tumours are the second-fastest-growing cause of cancer death. Computer-assisted tumour diagnosis is challenging, and efforts to increase the accuracy of tumour classification and generalisation are continually being made despite the plethora of studies conducted. This study of automated multi-class brain tumour classification utilising Magnetic Resonance Images aims to design and develop three automatic brain tumour classification approaches to categorise the brain tumours as glioma, meningioma, and pituitary tumours, which assist clinicians in making brain tumour diagnoses and developing further treatment plans to save patient's life. This research proposes a transfer learning approach using ResNet 50, hand-crafted features with machine learning classifiers, and a hybrid firefly-optimised multi-class classifier for tumour classification. The hybrid methodology yields the highest classification accuracy of 99% using the Figshare dataset. Furthermore, using the Figshare dataset, the hybrid technique yields the highest sensitivity (recall) of 99% for meningioma and pituitary tumours, the highest precision of 100% for pituitary tumours, and the highest F1-measure of 99% for pituitary tumours.