Combining TernausNet and Attention Aware Faster RCNN For Brain Tumor Segmentation And Classification In MRI Images

Main Article Content

Priyanka Kaushik , Dr. Rajeev Ratan

Abstract

currently, brain tumor segmentation and classification in MRI images based on conventional preprocessing, feature extraction and classification methods are not sufficient. For instance, previous methods have the issues of huge loss of information, not suited for large volume of data and also contain white Gaussian noise. This paper proposes Brain tumor Segmentation and Classification (BORSTAL)-MRI for accurate classification of the brain tumor classes. The proposed brain tumor segmentation and classification model has four steps. In the first step, the proposed model uses Multi-Stage Preprocessing, in which artifacts are removed by performing intensity redistribution, noise removal, and contrast enhancement, respectively. This step produces the final outcome with respect to clear blurriness and smoothen image. In the second step, TernausNet based brain tumor segmentation is performed which is better as compared to the existing deep learning algorithms (CNN, U-Net, etc.). In third step, three types of features are extracted as Color, Texture, and Shape. These features are extracted using Faster R-CNN algorithm and finally this algorithm is used for classification into three classes as: High Tumor, Low Tumor, and No Tumor. The proposed model is implemented using MatlabR-2020a in which various performance metrics are computed as follows: PSNR, SSIM, MSE, Error Rate, Foreground Precision, Background Precision, Dice Similarity, Accuracy, Specificity, Sensitivity and AUC. The comparison results show that the performance of the proposed BROSTAL –MRI model receives higher performance than the previous methods.

Article Details

Section
Articles