The Special Session Title ‘Intelligent Computing in Healthcare’ Dual Classification Framework for the Detection of Brain Tumor

Abstract

Cancer is a very common disease caused by mutation in DNA of cells and arises everywhere in the world. According to American Cancer Society, vast number of new cancer cases are predicted which are increasing every year. So, the early detection of brain tumor may decline the death percentage, and it would be helpful for the cancer specialist to cure the patients in primary stage. Image processing concepts are applied for the detection of normal and malignant brain scans using computer-aided tools. A new framework has been proposed to detect the brain tumor using magnetic resonance imaging (MRI) modality. The steps of the implementation of proposed framework are (a) image collection, (b) preprocessing, (c) segmentation, (d) statistical feature extraction, and (e) dual classification. Fuzzy K-mean segmentation is executed followed by statistical feature extraction in terms of texture. After that, dual classification is performed consequently one after another, i.e., autoencoder deep neural network followed by binary SVM. The primarily training has been completed by autoencoder, and the resultant of this is again trained by SVM. Finally, testing has been accomplished by SVM only. The performance of the proposed framework has been assessed with Multimodal Brain Tumor Segmentation Challenge dataset (BraTS 2013) in terms of sensitivity, specificity, AUC-ROC, and accuracy. The performance of the proposed framework is remarkable, and the accuracy of autoencoder + SVM achieved by this framework is 96.4% as compared to SVM alone.