Publication

Abstract : Classifying brain images is a challenging task, but one of the most practical and commonly used methods. Deep learning, a branch of artificial intelligence, is one of the state-of-the-art methods enabling new strategies to automate the interpretation of the medical images. Deep learning based hybrid segmentation and classification models for activity recognition using MRI brain data. The hybrid model is the first type, original for a fully convolutional network and consists of a residual network for segmentation and classification. The second type is a hybrid consisting of SegNet for the segmentation and MobileNet for the classification. In this paper, the annotated dataset of Meningioma, Pituitary tumor and Non tumor images is used. The proposed hybrid model is coded in python. The model obtained a high accuracy in the simulation on MRI brain dataset 93.9% and 91.3%. Compare and evaluate with other existing hybrid models using cube score, positive predictive value (PPV), false predictive value (FPV), precision, recall, specificity and F1 score. Thus, the proposed deep learning technique can help physicians and radiologists in the early diagnosis of brain tumors.