MRI Brain Tumour Classification Using Machine Learning and Deep Learning

A brain tumour is a critical, though rare disease that disrupts blood flow and exerts an exceedingly high pressure on vital parts of the brain. Around 300,000 cases of tumours are diagnosed every year, with 81.7% cases being accounted for by adults. While it is challenging to determine a singular reason for the origin of a tumour, it is well-established that genetic factors, age, radiation exposure, and medical history play a significant role. To predict the grade and region of the tumour, existing approaches employ non-invasive techniques, Machine Learning (ML) and Deep Learning (DL). The machine learning techniques used for classification are employed in traditional Random Forest algorithms, Support Vector Machines (SVM), and K-Nearest Neighbour (KNN) algorithms. On the other hand, Deep learning architectures of Custom Mask RCNN, variants of conventional CNN, Residual Net (Res-Net), and VGG networks are also adept at estimating classes. This survey reviews various ML and DL techniques introduced in recent years, with the aim of summarizing research findings for reproducibility. The ML and DL based architectures exhibit an accuracy of over 99.5% and yield similarly successful results in terms of precision and F1 scores, thus proving to be indicative of the tremendous potential of ML and DL in tumour detection.