Fig. 2

The habitats (from H = 2 to H = 16) formed over the patient cohort using the apparent diffusion coefficient (ADC) and maximum slope of increase (MSI) at V1 and V2 are visualized on the center axial tumor slice of an example patient. As the indices assigned by k-means clustering are random, we order the indices in terms of increasing mean V1 ADC to more easily visualize changes in the maps as we increase H. While the habitats are not formed using any spatial information, note that they do seem to be spatially connected. As we increase h, the resulting habitats capture increasing intra-tumoral heterogeneity. For example, when H = 2, 3, or 4 we identify a habitat covering much of the edge of the tumor on the left side of this slice. Investigating the same area with a larger H, such as 16, the habitat covering the left edge of the tumor is divided into additional habitats. As the habitats are used to inform regions in which a tumor cell proliferation rate is locally calibrated, the additional heterogeneity captured with increasing H can also result in more heterogeneity in the calibrated proliferation rates.