Investigation of changes in anomalous diffusion parameters in a mouse model of brain tumour

In this paper, we investigate anomalous diffusion models in a mouse model of glioblastoma, a grade IV brain tumour, and study how the anomalous diffusion model parameters reflect the change in tumour tissue microstructure. Diffusion-weighted MRI data with multiple b-values at 9.4T was acquired from mice bearing U87 brain tumour cells at four time points. Voxel-level fitting of the MRI data was performed on the classical mono-exponential model, and four anomalous diffusion models, namely, the stretched exponential model, the sub-diffusion model, the continuous time random walk model and the fractional Bloch-Torrey equation. The performance of the anomalous diffusion parameters for differentiating the three-concentric layers of tumour tissue (i.e., core; intermediate zone; peripheral and hyper-vascularised tumour layer) was evaluated with multinomial logistic regression and multi-class classification analysis. We found that parameter (\alpha) from the stretched exponential model, parameter (\beta) from the subdiffusion model and parameter (\beta) from the continuous time random walk model provide a clear delineation of the three layers of tumour tissue. The analysis revealed that the combination of diffusion coefficient D and anomalous diffusion parameters greatly improved the classification power in terms of F1-scores compared with the current approach in clinics, in which D is used alone. Hence, our mouse brain tumour study demonstrated that anomalous diffusion model parameters are useful for differentiating different tumour layers and normal brain tissue.