SU-D-218-01: Support Vector Machine Tissue Classification of Multiparametric MRI Tumor Data.

Area of Science:

• Neuroimaging
• Machine Learning
• Oncology

Background:

• Glioblastoma Multiforme (GBM) is an aggressive brain tumor.
• Accurate tumor segmentation is crucial for treatment planning and monitoring.
• Multiparametric MRI offers rich data for characterizing tumor tissue.

Purpose of the Study:

• To evaluate the accuracy of Support Vector Machines (SVM) for classifying Glioblastoma Multiforme tumor and cyst voxels.
• To assess the performance of SVM using multiparametric MRI data.
• To explore the potential of SVM in differentiating tumor recurrence.

Main Methods:

• Collected multiparametric MRI data (T1, T2, diffusion, perfusion, hypoxia) from GBM patients.
• Generated Apparent Diffusion Coefficient (ADC) and corrected rCBV maps.
• Utilized a Support Vector Machine trained on radiologist-confirmed labels for 'cyst', 'tumor', and 'normal tissue'.

Main Results:

• The SVM achieved high specificity for both tumor (0.9841) and cyst (0.9825) classification.
• Sensitivity for tumor classification was 0.7498 and for cyst classification was 0.9414.
• Model performance improved with the inclusion of additional MRI features.

Conclusions:

• Support Vector Machines demonstrate capability in classifying tumor and cyst voxels in GBM.
• The SVM model shows promise for aiding in the early detection of tumor recurrence.
• Future research will focus on cross-patient validation and optimizing SVM for early recurrence detection.