Enhanced Brain Tumor Detection Using a Modified Efficient Net Model with MRI Imaging

Abstract

Recognition of brain tumor is one of the main challenges for brain tumor detection.  The earlier we diagnose brain tumor, the more easily would be the treatment of tumor, and the less pain the patient suffer for the treatment.  The current paper introduces a novel system for brain tumor detection, which is based on transfer learning with deep convolutional neural network. Based on the MRI images,  we choose a fine pretrained deep convolutional neural network model ‘VGG19’ based image classification for brain tumor detection.  Various imaging preprocessing and augmentation techniques are applied to improve the quality of brain tumor images and also to boost the generalization ability of the classification system. The implementation of image preprocessing techniques include image normalization, image resizing, random shifting, zooming and horizontal flipping.  Once the images are preprocessed, the MRI images are split into train, validation and test set. The deep features are extracted once the deep learning architecture is applied and the deep features are used to reach the classification of MRI images. The existing work is used to classify the MRI images into glioma,  meningioma,  pituitary tumor,  and No tumor types. The experimental comparative analysis report shows high classification performance of precision, recall and ROC-AUC metrics for each class using this framework. The developed framework apparently establishes an effective computer-aided diagnosis system for fast and precise identification of brain tumors for the benefit of the patients.