Optimizing Texture Retrieving Model for Multimodal MR Image-Based Support Vector Machine for Classifying Glioma.

Area of Science:

• Radiology and Medical Imaging
• Machine Learning in Healthcare
• Oncology

Background:

• Accurate glioma grading is crucial for effective patient treatment planning.
• Glioma grading relies on histopathological assessment, which can be invasive.
• Non-invasive imaging techniques offer potential for improved glioma grading.

Purpose of the Study:

• To evaluate the impact of different texture retrieval models on Support Vector Machine (SVM) based glioma grading.
• To identify the optimal texture model for classifying glioma grades using MRI data.

Main Methods:

• A retrospective study analyzed 117 glioma patients with WHO 2007 grades II, III, and IV.
• Texture attributes were extracted using Global, Gray-Level Co-occurrence Matrix (GLCM), Gray-Level Run-Length Matrix (GLRLM), and Gray-Level Size-Zone Matrix (GLSZM) models.
• Radial Basis Function SVM (RBF-SVM) with SVM-Recursive Feature Elimination (SVM-RFE) was employed for classification, with data augmentation using SMOTE.

Main Results:

• The GLSZM model, utilizing gray-level 64, demonstrated superior performance in retrieving texture attributes.
• The optimized SVM model achieved an accuracy of 0.875 and an AUC of 0.971 on the independent test set.
• SVM-RFE effectively reduced attribute redundancy and improved classification efficacy.

Conclusions:

• The GLSZM model, combined with specific gray-level selection and attribute reduction, represents an optimized solution for texture-based SVM glioma classification.
• This approach shows significant potential for accurate and non-invasive glioma grading.