A Fine Tuned Pre-trained Model for Classification of Brain Tumor using Magnetic Resonance Imaging

Any growth inside of the brain's restricted areas can harm humans. It is a challenging task for radiologists to classify tumor types because cancers are heterogeneous. The manifold collection of cells fosters prominent molecular squiggle with discriminative levels of sensitivity to treatment. Medical diagnosis has changed and diversified due to deep learning trends, particularly in magnetic resonance imaging (MRI). This study suggests a binary class classification and detection method for the early diagnosis of brain tumors as a solution to this issue. The pre-trained ResNet50, VGG19 and InceptionV3 deep learning models of the transfer learning approach make this possible. The different scenarios were assessed with the help of two classes of publicly available MR brain tumor datasets and with data gathered from hospitals. The accuracy rates for the pre-trained method InceptionV3, ResNet50, and VGG19 produced 99.72%, 98.84% and 94.65%, respectively. The three pre-trained models are ensembled. The highest performance accuracy in the detection of brain tumors was achieved using Inceptionv3, and VGG19 on testing samples. The results of the proposed method evaluation outperform those of existing methods