Quantitative Analysis of Deep Learning Models Efficacy for Brain Tumor Detection

The escalation of abnormal cells in the brain causes the malady known as Brain Tumor. Meticulous as well as premature discernment of Brain Tumors is requisite for a successful diagnosis of the malady. Pilot detection doesn't only contribute to providing better medications but also encompasses the probability of saving an affected person's life. With the advancement in computerized diagnosis and biomedical informatics, Neuro-Oncologists have gained an upper hand in their domain. There has been the incorporation of various Machine and Deep Learning Algorithms in the particular prospect to resolve the human error in disease identification through automation. These algorithms have assisted in obtaining a greater accuracy than manual articulation. Although, with computer-aided architecture comes higher computational complexity and the lack of instances make it even worse. The proposed system resolves this complication by inculcating various lightweight Deep Learning Models. More precisely, the system possesses self-collated MRI Images classified as “Brain Tumor” and “No Brain Tumor” and projected over diverse Deep Learning architectures such as MobileNetV2, DenseNet121, ResNet50, and VGG16 which are selected based on varied parameters. The performance of these models excelled in every aspect giving us potent results with the manifestation of real-time profuse interpretation. The models have been juxtaposed over various metrics such as Area Under the Curve, Cohen Kappa Score, Precision, Recall, and F1-Score for validating the efficacy of our system.