Classification Of Brain Images For Identification Of Tumors

Abstract

Early detection of brain tumors is very crucial as they grow extremely fast. To extend patients' life expectancy, correct treatment planning and precise diagnoses are critical. Manual diagnosis can be prone to errors and is a time-consuming and complex task for radiologists because of how minute variations in the tumor could lead to a completely different diagnosis. The proposed method is focused on creating an automated way of classifying brain MRI images by using SOTA models like VGG-16 and InceptionV3 and building on them. The brain MRI images are classified into four classes by extracting significant features and experimented with and without pre-processing. The experimental results have shown that the VGG-16 model used, although without any image augmentation, has given a high validation accuracy of 74%. The inceptionV3 model without image augmentation techniques reported a worse validation accuracy of 69%, defining VGG-16 to be the better classifier.