Multi-class Brain Tumour Classification Using Optimized ResNet50

Artificial Intelligence has been proving its excellence in almost all the domains including agriculture, medicine, finance, education etc., with the growing fusion of Artificial Intelligence (AI) and medicine, various algorithms for the automatic diagnosis and prediction of diseases have been evolving in recent years. An algorithm for automating classification of different types of tumours is not just a necessity, but a moral imperative. Malignant tumors, as we know are life-threatening and brain cancer of all of them has the lowest survival rate. Early detection of tumours whether benign or malignant, is the best cure when it comes to preventing brain cancer, the tenth leading cause of death worldwide. With this machine learning model, medical technicians can shave off countless man-hours and garner an accurate diagnosis in a fraction of the time. Human eyes can also find it hard to discern between cancer cells and healthy cells; which is another issue solvable by this neural network’s implementation in hospitals. Hence, in this paper we are using an optimised transfer learning model ResNet50, for the automatic classification of three types of tumours in MRI brain scans. The proposed Resnet model classifies Meningioma, Glioma and Pituitary, across varied datasets like kaggle, figshare and BraTS 2015 with an average accuracy rate of 97.48% on validation data, 94.06% on test data and an F1-score of 0.95.