Model-free cluster analysis for multi-b-value diffusion-weighted imaging of the inferior alveolar nerve

Objectives: Multi-b-value diffusion-weighted imaging (DWI) is useful for the detection of pathological tissues. The signal decay curve of each voxel is generally analyzed using an exponential model; however, the most appropriate curve for an analysis does not necessarily follow a single function. We used model-free analysis to classify data from the inferior alveolar nerve (IAN) into clusters according to the pattern of decay curves. This clustering should be able to classify groups of voxels representing different tissue properties in the neurovascular bundle of the IAN. Materials and Methods: DWI with eight b-values was acquired from the IANs of 13 normal volunteers. K-means cluster analysis was used to classify the data. Silhouette analysis was performed to define the optimum number of clusters. The suitability of single and double exponential functions was evaluated for each cluster. The fitting and spatial distributions of parameters associated with diffusion and perfusion using the double exponential model were tested. Results: The optimum number of clusters was three. The cluster that exhibited the steepest decay curve showed a higher apparent diffusion coefficient than the others, and was affected by the perfusion component. The cluster with the most gradual decay curve showed the best fit to the double exponential function, and contained the highest volume fraction of the slow diffusion component, indicating a different distribution to the other clusters. Conclusions: Clustering of DWI was accomplished without model fitting and was presumably based on the diffusivity properties of IAN, which may be influenced by microcirculation and fascicles.