Imaging data are provided by the Biomedical Magnetic Resonance Laboratory at Mallinckrodt Institute of Radiology, Washington University.

The D-Histo imaging data used in this study consists of 79 patients samples and 9 non-patient samples, where patient samples record data of PCa and non-patient samples record reference data of non-PCa.

Format of Data

Data of each sample contains a 3D matrix for each D-Histo map, with one slice used to calculate D-Histo values of the corresponding physical meaning, and two masking matrices, one annotating the entire tissue and one annotating the cancer area. 12 D-Histo maps are selected to be used based on their physical meanings.

Definition of D-Histo Maps

  • DTI: diffusion tensor imaging, traditional MRI method
  • ADC: apparent diffusion coefficiency
  • axial diffusivity: water diffusion along fiber
  • radial diffusivity: water diffusion perpendicular to fiber
  • FA: fractional anisotropy
  • restricted ratio 1 map: highly restricted isotropic diffusion portion, proposed to reflect water diffusion within smaller cells like lymphocytes
  • restricted ratio 2 map: restricted isotropic diffusion portion, proposed to reflect water diffusion within larger cells like cancer cells
  • hindered ratio map: hindered isotropic diffusion portion, proposed to reflect water diffusion in extracellular space or brain tumor necrosis regions
  • water ratio map: free isotropic diffusion portion, proposed to reflect water diffusion in CSF regions
  • fiber ratio map: anisotropic diffusion portion, proposed to reflect water diffusion in white matter axon

Analysis of Data

Distribution of voxels in all samples; red: benign tissues, blue: cancerous tissues

Between PCa and non-PCa, some D-Histo maps turn out to have significant difference in values. This suggests that the separation between PCa and non-PCa data can be relatively easy.

Distribution of voxels in all patient samples with different primary Gleason scores

However, no specific pattern can be observed from the statistics of PCa data with different Gleason scores. The mean values of maps of each class are very similar and all have large variance, which makes it hard to distinguish among cancerous classes.