Automatic glioma characterization from dynamic susceptibility contrast imaging: brain tumor segmentation using...

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Oncology

Background:

• Accurate glioma volume delineation is crucial for diagnosis and treatment planning.
• Manual segmentation of tumors from dynamic susceptibility contrast (DSC) MRI is time-consuming and subject to inter-observer variability.
• Knowledge-based fuzzy c-means (FCM) clustering offers a potential automated solution.

Purpose of the Study:

• To evaluate the diagnostic efficacy of automated glioma volumes derived from knowledge-based FCM clustering against manually defined volumes.
• To compare the performance of both segmentation methods in glioma grading and patient survival prediction using DSC imaging.

Main Methods:

• Fifty patients with newly diagnosed gliomas underwent DSC MRI at 1.5 Tesla.
• Glioma volumes were segmented using knowledge-based FCM clustering and by four independent neuroradiologists.
• Diagnostic efficacy was assessed using receiver operator characteristic (ROC) curve analysis and Kaplan-Meier survival analysis.

Main Results:

• Automated and manual volumes yielded similar diagnostic efficacy for glioma grading, with comparable areas under the ROC curves (P = 0.576-0.970).
• Kaplan-Meier survival analysis showed a significantly higher log-rank value for automated volumes (14.403; P < 0.001) compared to manual volumes (10.650-12.761; P = 0.001-0.002) in distinguishing high-risk from low-risk patients.
• The automated method demonstrated superior performance in stratifying patients based on expected survival.

Conclusions:

• Knowledge-based FCM clustering provides an automated, user-independent method for glioma volume segmentation in DSC imaging.
• This automated approach achieves diagnostic efficacy comparable to manual segmentation for glioma grading.
• The automated method offers improved accuracy in predicting patient survival, aiding presurgical glioma characterization.