Multimodal MR imaging model to predict tumor infiltration in patients with gliomas.

Introduction: Gliomas remain difficult to treat, in part, due to our inability to accurately delineate the margins of the tumor. The goal of our study was to evaluate if a combination of advanced MR imaging techniques and a multimodal imaging model could be used to predict tumor infiltration in patients with diffuse gliomas.
Methods: Institutional review board approval and written consent were obtained. This prospective pilot study enrolled patients undergoing stereotactic biopsy for a suspected de novo glioma. Stereotactic biopsy coordinates were coregistered with multiple standard and advanced neuroimaging sequences in 10 patients. Objective imaging values were assigned to the biopsy sites for each of the imaging sequences. A principal component analysis was performed to reduce the dimensionality of the imaging dataset without losing important information. A univariate analysis was performed to identify the statistically relevant principal components. Finally, a multivariate analysis was used to build the final model describing nuclear density.
Results: A univariate analysis identified three principal components as being linearly associated with the observed nuclear density (p values 0.021, 0.016, and 0.046, respectively). These three principal component composite scores are predominantly comprised of DTI (mean diffusivity or average diffusion coefficient and fractional anisotropy) and PWI data (rMTT, Ktrans). The p value of the model was <0.001. The correlation between the predicted and observed nuclear density was 0.75.
Conclusion: A multi-input, single output imaging model may predict the extent of glioma invasion with significant correlation with histopathology.