Detection of glioma on brain MRIs using adaptive segmentation and modified graph neural network based classification

Abstract

Gliomas constitute the prevalently seen brain tumours in humans. The real-time utilization of Computer Aided Diagnosis system depends on brain Magnetic Resonance Imaging (MRIs) has the ability of helping radiologists and professionals to identify the presence of glioma tumours. It is very difficult to segment brain tumours because of the brain image and it has a complex structure. A fully automated, accurate, segmentation and classification model is developed using a modified Graph Neural Network (MGNN) for brain tumours. Proposed work steps are, image registration, Shift-Invariant Shear let Transform (SIST), adaptive segmentation, feature extraction, and categorization of tumours. At first, image registration and SIST are carried out to improve image quality. Adaptive segmentation is then carried out utilizing Improved Fuzzy C-Means clustering. Next, Grey Level Co-occurrence Matrix, Discrete Wavelet Transform is utilized for the extraction of features in brain MRI data. Finally, MGNN is introduced for the detection of anomalous tumour-infected MR and actual MR brain images. The findings are demonstrated that the proposed model leads in higher accuracy levels for both classification and segmentation.

Keywords

• Gliomas
• image registration
• shift-invariant shear let transform (SIST)
• improved fuzzy c-means (IFCM) clustering
• grey level co-occurrence matrix (GLCM) and discrete wavelet transform (DWT)
• modified graph neural network (MGNN)