Multiparametric analysis of magnetic resonance images for glioma grading and patient survival time prediction.

BACKGROUND: A systematic comparison of magnetic resonance imaging (MRI) options for glioma diagnosis is lacking. PURPOSE: To investigate multiple MR-derived image features with respect to diagnostic accuracy in tumor grading and survival prediction in glioma patients. MATERIAL AND METHODS: T1 pre- and post-contrast, T2 and dynamic susceptibility contrast scans of 74 glioma patients with histologically confirmed grade were acquired. For each patient, a set of statistical features was obtained from the parametric maps derived from the original images, in a region-of-interest encompassing the tumor volume. A forward stepwise selection procedure was used to find the best combinations of features for grade prediction with a cross-validated logistic model and survival time prediction with a cox proportional-hazards regression. RESULTS: Presence/absence of enhancement paired with kurtosis of the FM (first moment of the first-pass curve) was the feature combination that best predicted tumor grade (grade II vs. grade III-IV; median AUC = 0.96), with the main contribution being due to the first of the features. A lower predictive value (median AUC = 0.82) was obtained when grade IV tumors were excluded. Presence/absence of enhancement alone was the best predictor for survival time, and the regression was significant (P < 0.0001). CONCLUSION: Presence/absence of enhancement, reflecting transendothelial leakage, was the feature with highest predictive value for grade and survival time in glioma patients.

Glioma grading by using histogram analysis of blood volume heterogeneity from MR-derived cerebral blood volume maps.

PURPOSE: To retrospectively compare the diagnostic accuracy of an alternative method used to grade gliomas that is based on histogram analysis of normalized cerebral blood volume (CBV) values from the entire tumor volume (obtained with the histogram method) with that of the hot-spot method, with histologic analysis as the reference standard. MATERIALS AND METHODS: The medical ethics committee approved this study, and all patients provided informed consent. Fifty-three patients (24 female, 29 male; mean age, 48 years; age range, 14-76 years) with histologically confirmed gliomas were examined with dynamic contrast material-enhanced 1.5-T magnetic resonance (MR) imaging. CBV maps were created and normalized to unaffected white matter (normalized CBV maps). Four neuroradiologists independently measured the distribution of whole-tumor normalized CBVs and analyzed this distribution by classifying the values into area-normalized bins. Glioma grading was performed by assessing the normalized peak height of the histogram distributions. Logistic regression analysis and interobserver agreement were used to compare the proposed method with a hot-spot method in which only the maximum normalized CBV was used. RESULTS: For the histogram method, diagnostic accuracy was independent of the observer. Interobserver agreement was almost perfect for the histogram method (kappa = 0.923) and moderate for the hot-spot method (kappa = 0.559). For all observers, sensitivity was higher with the histogram method (90%) than with the hot-spot method (55%-76%). CONCLUSION: Glioma grading based on histogram analysis of normalized CBV heterogeneity is an alternative to the established hot-spot method, as it offers increased diagnostic accuracy and interobserver agreement.

Predictive modeling in glioma grading from MR perfusion images using support vector machines.

The advantages of predictive modeling in glioma grading from MR perfusion images have not yet been explored. The aim of the current study was to implement a predictive model based on support vector machines (SVM) for glioma grading using tumor blood volume histogram signatures derived from MR perfusion images and to assess the diagnostic accuracy of the model and the sensitivity to sample size. A total of 86 patients with histologically-confirmed gliomas were imaged using dynamic susceptibility contrast (DSC) MRI at 1.5T. Histogram signatures from 53 of the 86 patients were analyzed independently by four neuroradiologists and used as a basis for the predictive SVM model. The resulting SVM model was tested on the remaining 33 patients and analyzed by a fifth neuroradiologist. At optimal SVM parameters, the true positive rate (TPR) and true negative rate (TNR) of the SVM model on the 33 patients was 0.76 and 0.82, respectively. The interobserver agreement and the TPR increased significantly when the SVM model was based on an increasing sample size (P < 0.001). This result suggests that a predictive SVM model can aid in the diagnosis of glioma grade from MR perfusion images and that the model improves with increasing sample size.

Automatic glioma characterization from dynamic susceptibility contrast imaging: brain tumor segmentation using knowledge-based fuzzy clustering.

PURPOSE: To assess whether glioma volumes from knowledge-based fuzzy c-means (FCM) clustering of multiple MR image classes can provide similar diagnostic efficacy values as manually defined tumor volumes when characterizing gliomas from dynamic susceptibility contrast (DSC) imaging. MATERIALS AND METHODS: Fifty patients with newly diagnosed gliomas were imaged using DSC MR imaging at 1.5 Tesla. To compare our results with manual tumor definitions, glioma volumes were also defined independently by four neuroradiologists. Using a histogram analysis method, diagnostic efficacy values for glioma grade and expected patient survival were assessed. RESULTS: The areas under the receiver operator characteristics curves were similar when using manual and automated tumor volumes to grade gliomas (P = 0.576-0.970). When identifying a high-risk patient group (expected survival <2 years) and a low-risk patient group (expected survival >2 years), a higher log-rank value from Kaplan-Meier survival analysis was observed when using automatic tumor volumes (14.403; P < 0.001) compared with the manual volumes (10.650-12.761; P = 0.001-0.002). CONCLUSION: Our results suggest that knowledge-based FCM clustering of multiple MR image classes provides a completely automatic, user-independent approach to selecting the target region for presurgical glioma characterization.