Voxel-based clustered imaging by multiparameter diffusion tensor images for glioma grading.

Area of Science:

• Neuroimaging
• Radiology
• Oncology

Background:

• Gliomas are common brain tumors, and accurate pre-operative grading is crucial for treatment planning.
• Current methods for determining glioma grade pre-operatively are insufficient.

Purpose of the Study:

• To develop an unsupervised method using multiple parameters from pre-operative diffusion tensor images (DTI) for glioma visual grading.
• To create a clustered image enabling visual assessment of glioma grade.

Main Methods:

• Extracted seven DTI features (including diffusion-weighted imaging, fractional anisotropy, eigenvalues, and mean diffusivity) from 33 patients (14 low-grade, 19 high-grade gliomas).
• Employed a two-level clustering approach: Self-Organizing Map (SOM) followed by K-means algorithm for unsupervised clustering.
• Developed a voxel-based diffusion tensor-based clustered image and assessed its efficacy in a supervised manner using Support Vector Machine (SVM).

Main Results:

• The 16-class diffusion tensor-based clustered images achieved high performance in differentiating glioma grades: sensitivity 0.848, specificity 0.745, accuracy 0.804, and AUC 0.912.
• Specific clusters (14, 15, 16) showed significantly higher log-ratio values in high-grade gliomas compared to low-grade gliomas.
• These differentiating clusters represented distinct patterns of the seven DTI-derived parameters.

Conclusions:

• Multiple DTI parameters integrated into voxel-based clustered images can effectively differentiate between low- and high-grade gliomas.
• This unsupervised clustering method offers a promising tool for pre-operative glioma grading.
• The visual grading approach using clustered DTI shows potential to improve therapeutic strategy selection.