Deep Transfer Learning Framework for Brain Tumor Detection in MRI Slices: A Study

The tumour in brain is a medical emergency and early detection and treatment is crucial. Magnetic Resonance Imaging (MRI) based brain tumor screening is one of the common clinical procedure and the outcome of this process helps to plan and execute the necessary process. This research aims to develop a deep-learning (DL) based approach in detecting the brain-tumour (BT) with better accuracy. The various phases of this work included the following; (i) image collection and resizing, (ii) feature extraction with conventional DL-model, (iii) best model selection and generating fused-feature-vector (FFV) based on identified features after 50% dropout, and (iv) classification and performance verification using the binary classifiers. This work proposed a framework to examine 1200 images of healthy/tumor class MRI and achieved a detection accuracy of >97% with Random-forest classifier. This verifies that this approach is good on the MRI database and in future, it can be considered to verify the tumour in real clinical MRI slices.