Automatic brain tumor segmentation from magnetic resonance images using superpixel-based approach

• Published in: Multimedia tools and applications 2022-11, Vol.81 (27), p.38409-38427
• Main Authors: Iqbal, Muhammad Javaid, Bajwa, Usama Ijaz, Gilanie, Ghulam, Iftikhar, Muhammad Aksam, Anwar, Muhammad Waqas
• Format: Article
• Language: English
• Subjects: Algorithms; Brain; Brain cancer; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Decision trees; Feature extraction; Glioma; Image segmentation; Magnetic resonance imaging; Medical imaging; Multimedia Information Systems; Sensitivity; Special Purpose and Application-Based Systems; Tumors; Variability
• ISSN: 1380-7501; 1573-7721
• DOI: 10.1007/s11042-022-13166-7

Cancer is the second leading cause of deaths worldwide, reported by World Health Organization (WHO). The abnormal growth of cells, which should die at the time but they remained in body organ which makes tumor and brain tumor is one of them. During its treatment planning, brain tumor segmentation plays its vital role, Magnetic Resonance Imaging (MRI) is most widely used medical imaging modalities to scan brain tissues, and segmentation of brain tumor from MRI scans is still a challenging task, due to the variability in spatial, structure and appearance of the brain tumor. The existing brain tumor segmentation techniques are still suffering from an inadequate performance, dependent on initial assumptions, and required manual interference. The main challenge is to segment out the accurate tumor from MRI images, and to give the solution for its variability in size due to spatial change in image slices. The proposed model in an automated manners segment out abnormal tissues from MRI images. The proposed model has some aspects like we apply some pre-processing techniques, and apply superpixel-segmentation with their improved tuned parameter values. We have extracted different features for the superpixels in the images such that statistical features, fractal features, texton features, curvature feature and SIFT features. Due to unbalanced feature vector, we have proposed class balancing algorithm, and then apply SVM, KNN, Decision Tree and Ensemble classifiers, to classify the normal and abnormal superpixels. To evaluate the proposed model, we used MICCAI BRATS-2017 MRI training dataset. The Dice Coefficient (DSC), precision, sensitivity, and balanced error rate (BER) against the ground truths for FLAIR sequence in LGG volumes have been obtained as 0.8593, 87%, 93%, and 0.08 respectively. The DSC, precision, sensitivity, and BER against the ground truths for FLAIR sequence in HGG volumes have been obtained as 0.8528, 87%, 97%, and 0.08 respectively. It is evident from the quantitative and visual results that the proposed model provides a close match to the expert delineation for the FLAIR sequence.