Innovative fusion: MRSI-guided brain tumour classification via integrated image segmentation and GLCM feature extraction

Abstract

Accurate brain tumour classification is crucial for effective diagnosis and treatment planning. This study aims to develop an advanced classification approach by integrating Magnetic Resonance Spectroscopy Imaging (MRSI) segmentation with Grey Level Co Occurrence Matrix (GLCM) feature extraction to improve glioma classification. Using MR images from 52 subjects, we examined various glioma metabolites and computed metabolite ratios within segmented tumour regions. GLCM features provided insights into pixel intensity spatial distributions and metabolic alterations, aiding in differentiation of tumour types and grades. Our method, employing image preprocessing, Otsu's thresholding for segmentation, and CNN Efficient Net for classification, achieved nearly 90% accuracy in high-grade glioma identification. The confusion matrix demonstrated strong classification performance, with a precision of 0.89 for healthy subjects and 0.92 for glioma cases, while recall values were 0.92 and 0.88, respectively. The F1-scores of 0.91 (healthy) and 0.90 (glioma) further indicate the model's balanced predictive capability. Additionally, the Receiver Operating Characteristic (ROC) curve yielded an AUC of 0.95, emphasising the effectiveness of the proposed approach. Significant correlations were found between model efficacy, metabolite ratios, and GLCM features, highlighting the efficiency of our method in brain tumour classification and its potential clinical relevance.