gms | German Medical Science

Objective: In low-grade gliomas (LGG), early postoperative MRI (epMRI) has been shown to frequently over-estimate the amount of residual tumor on fluid-attenuated inversion recovery (FLAIR) sequences. Therefore, residual tumor is usually defined according to follow-up MRI 3 months after surgery (fuMRI). This study sought to evaluate if integration of apparent diffusion coefficient (ADC)-maps permits an accurate estimation of residual tumor also on epMRI.

Methods: Since 2004, 28 consecutive cases with primary surgery for WHO°II astrocytomas and complete epMRI and fuMRI including ADC-maps were retrospectively identified. To account for imaging confounders, histology was adjusted and cases with previous radiation therapy were excluded. Residual FLAIR hyperintense tumor was manually segmented on epMRI. After co-registration with corresponding ADC-maps, residual tumor segments were probabilistically clustered into areas of either low-ADC (edema, ischemia), high-ADC (residual tumor) or partial volume. Clustering was based on an expectation maximization (EM) algorithm fitting a mixture model of three respective Gaussians to the normalized ADC histogram. EpMRI residual tumor and results from ADC-clustering were compared with respective residual tumor on fuMRI.

Results: Median residual FLAIR tumor was significantly greater on epMRI compared to intMRI (17.0ccm, range 1.6-122.7ccm vs. 1.8ccm, range 0-168.0ccm, p<0.001). ADC-clustering of epMRI subclassified FLAIR-hyperintense segments into proportions of: 27±23% residual tumor (median volume 1.7ccm, range 0.2-79.2ccm), 45±20% ischemia (6.2ccm, range 0.4-47.2ccm) and 28±20% partial volume (2.9ccm, range 0.2-64.9ccm). After ADC-clustering, the amount of residual tumor on epMRI was comparable to fuMRI (p=0.45) and showed significant correlation (Spearman r=0.65, p<0.0001). The amount of low-ADC clusters on epMRI suggestive for surgical trauma strongly correlated with the difference in FLAIR-hyperintensity between epMRI and fuMRI (Spearman r=0.86, p<0.0001).

Conclusion: With additional information conveyed in ADC-maps, an accurate quantification of residual tumor could already be achieved on epMRI in this series of low-grade astrocytomas. Probabilistic segmentation of ADC-maps helped to reassess the amount of residual tumor under suppression of signal alterations caused by surgical trauma which was the driving force for bias on epMRI. Reliability of ADC-clustering results based on the proportional partial volume has to be evaluated in the future. Larger case-series are needed to evaluate the prognostic impact of ADC-clustering in LGG.