A model free, sparse acquisition approach (« Sindex ») to investigate brain tissue microstructure from diffusion MRI data in the human brain
Denis Le Bihan1,2,3, Tetsuya Yamamoto2, Masaki Fukunaga2, Tomohisa Okada3, and Norihiro Sadato2

1NeuroSpin/Joliot, CEA-Saclay, Gif-sur-Yvette, France, 2System Neuroscience, National Institute for Physiological Sciences, Okazaki, Japan, 3Graduate School of Medicine, Human Brain Research Center, Kyoto University, Kyoto, Japan


Diffusion MRI provides noninvasively information of tissue microstructure. Current models allow to empirically analyze data or to provide more insightful information on the tissue features. However, those models require strong assumptions on the underlying tissues and the acquisition of large image data sets with different acquisition parameters. We have investigated a new, model free approach which enables classification of tissue types from the “proximity” or resemblance of their diffusion MRI signal profile at a sparse set of key b values (maximizing sensitivity to tissue microstructure) to a library of “signature” signal profiles (e.g. typical brain grey and white matter).


Diffusion MRI (dMRI) provides noninvasively information of tissue microstructure, especially non-Gaussian diffusion (1). Many models have been suggested to empirically analyze dMRI data, such as the polynomial or Kurtosis model (2). Other models have been designed to provide more insightful, ‘explanatory’ information on the tissue features, such as the axon diameter (AxCaliber model (3)) or the neurite distribution (NODDI model (4)). However, those refined models require strong assumptions on the underlying tissue and the acquisition of multiple images with a large range of b values and/or diffusion times, resulting in long acquisition times. Indeed, a great number of tissue features (the values are most often unknown and highly variable) interact with water diffusion (eg cell shape, size or cell density), one may consider that accurately modeling tissues might be ill-posed and elusive. To cut on acquisition and processing times we have investigated a new, model free approach which enables direct classification of tissue types: A signature index (5) is calculated from the “proximity” or resemblance of the diffusion MRI signal profile of a examined tissue (obtained using a sparse set of key b values chosen for the higher sensitivity to underlying tissue microstructure) to a database or library of “signature” signal profiles acquired or simulated once for all.


dMRI data were acquired on 8 normal subjects using 7T MRI scanners (Siemens Healthinners) at 2 institutions using a 32 channel head RF coil. Data were first acquired on 3 subjects with 11 b values [0-4000 s/mm²] (64 directions, 1.2x1.2x2mm3 voxels, 30 slices, TR/TE=6000/91ms, PAT2, 4 averages) to establish typical signature decay signals (S=f(b)) for grey, SG, and white matter, SW, and to determine corresponding key b values from a set of differential equations (5). dMRI data were then collected on the other subjects with the same parameters but using only those 2 key b values (Lb=200 and Hb=1800s/mm²). The S index, SI(V) was calculated from the direction-averaged, normalized signals, SV(b) in each voxel at the key b values, as the algebraic distance between the vector made of these signals and those of the signature tissue signals for each key b value:


with dSV,W,G(b)=[SV,W,G(b)- SN(b)]/SN(b). SN is taken as an intermediate signal between SW and SG. SI was then further linearly scaled as Sindex=(SI+1)*25+25 which is now centered at 50, so that Sindex=75 for a typical white matter tissue and Sindex=25 for a typical gray matter tissue. The S index represents a continuous scale. Color-encoded maps and 3D renderings of the segregated tissues based on the voxel-by-voxel Sindex were generated. Beside mean Sindex statistics in ROIs were also generated with histograms to assess local heterogeneity (texture analysis). DTI images were also produced for comparison.


A typical Sindex map is shown in Fig.1 with a comparison with DTI Mean Diffusivity map. White matter appears with high S index values (>60) but present important local variations. Gray matter Sindex is around 40 and much more homogeneous. CSF spaces appear in blue (Sindex close to 0). To further understand the nature of the new contrast generated in the Sindex maps further processing was done to segregate tissues based on the Sindex values. The “white matter” map (Fig.2) globally reflects the presence of diffusion anisotropy, but small local differences can be found with the Fractional Anisotropy map (shown for comparison). The “gray matter” (Fig.3 and 4) map clearly shows basal ganglia and the cortical ribbon. Strikingly important variations in Sindex values are visible along the cortical ribbon (Fig.3 and 4). It is well known that the brain is a spatially very inhomogeneous organ. This new and simple Sindex approach has the potential to generate in vivo maps of cyto and myeloarchitecture in the human brain without making assumptions about underlying tissue structure. Further work is obviously necessary, to link the nature of the Sindex values with known tissue features, but one may envision that the Sindex might reveal differences related to the functional areas along the cortical surface (6).


With this “signature index” tissue patterns and classification can be readily obtained with accuracy from a limited set of dMRI images without calculating any parameter model. Potentially, this Sindex can also be tuned to reveal more specific features, for instance to provide an estimation on the amount of neurogenesis in specific brain regions following brain irradiation (7). Beside cyto- and myeloarchitectony of the normal human brain this information might be very useful to investigate the brain of patients with neurological or psychiatric disorders, or under therapy, revealing possible alterations in local brain tissue microstructure.


No acknowledgement found.


1.Le Bihan D. EMBO Mol Med.2014 May 1;6(5):569-73.

2.Jensen JH, et al. Magnetic Resonance in Medicine 2005;53(6):1432-1440.

3.Assaf Y, et al. MRM 2008 59(6) :1347-1354.

4.Zhang H et al. Neuroimage 2012 61(4) : 1000-1016.

5.Iima M, Le Bihan D. Radiology. 2016 Jan;278(1):13-32.

6.Glasser MF, et al. Nature. 2016; 536: 171-178.

7.Pérès EA, et al. Int J Radiat Oncol Biol Phys. 2018 Feb 2. pii: S0360-3016(18)30184-6.


Sindex map (left) and Mean Diffusivity map (right)

Sindex map tuned to “White Matter” (left) and FA map (right) (same subject/slice as in Fig.1)

Sindex map tuned to “Gray Matter” (same subject/slice as in Fig.1)

Sindex 3D rendering maps of gray and white matter (different subject)

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)