Clinically, biopsy is considered as gold standard for grading brain tumors. Knowledge of tumor grade determines treatment decisions. Perfusion imaging is currently used to infer tumor grade. However this requires contrast agent injection. Apparent Diffusion Co-efficient (ADC) maps derived from Diffusion MRI provide image contrast (without contrast injection) through measurement of the diffusion properties of water within tissues. ADC maps are expected to provide different intensities to low grade and high grade tumors¹. However, these intensity differences are not easily distinguishable by visual interpretation. In a previous study² a method based on “texture analysis” of ADC maps was proposed for tumor grading, with validation on limited 1.5 T data. In this study, we use a modified method and show additional results on 3.0 T data.

Data Acquisition: 3.0 Tesla MR data from 46 patients (28 high grade and 18 low grade) with brain tumor was retrospectively analyzed. Necessary IRB and Ethics approvals were obtained. Routine anatomic imaging was performed using standard T1W, T2W, FLAIR and post-contrast T1W sequences. Diffusion images were acquired by standard EPI-based diffusion weighted imaging (b-values of 0 and 1000). ADC maps were generated using product software on the scanner console.

Heterogeneity Quantification: A method that quantifies spatial heterogeneity in brain tumors from ADC maps (Bhat et al.²) has been used for post-processing and analysis of 3.0 T MR data. The method is based on modified co-occurrence matrices and yields a heterogeneity map, which shows the local variations within the tumor and then produces a global measure of the tumor, which we call as “heterogeneity measure”. This measure is used for brain tumor grading. The “contrast” of co-occurrence matrices was computed for each voxel of the image to generate the heterogeneity map. A heterogeneity measure is then computed as kurtosis of the histogram (fixed x-axis) of the entire tumor in the heterogeneity map.

An example of heterogeneity map from low and high grade tumors is shown in figure 1. Bar chart showing comparison of simple ROI mean, ROI kurtosis and heterogeneity measure in low and high grade tumors (mean±SE) is shown in figure 2. There was significant difference between high and low grade tumors using heterogeneity measure (p-value < 0.05). 39 out of 46 patients were found to be correctly classified using a threshold in-between mean values of high and low grade tumors. In the other seven patients, the tumors were either very small or had undergone surgical interventions.In this study, we showed that instead of simple methods such as ROI mean or kurtosis of ADC, the heterogeneity measure of ADC values within gliomas may be more useful for grading brain tumors. In the present study, we tested a texture analysis- based method that incorporates information on the local spatial heterogeneity within each imaging slice through the tumor (heterogeneity map). We observed that values in the heterogeneity map were widely distributed in the high grade tumors, whereas in low grade tumors only lower values were seen in the heterogeneity map. This indicates that the more heterogeneous behavior of high grade tumors was captured using the method proposed in this study. As a limitation of this study, tumors with interventions or very small size tumors (only a few voxels) were not classified correctly. Such cases can be identified by the clinicians and these should not be assessed by the method used in this study.