We recently proposed a method for fast estimation of the tissue microstructural parameters such as intra/extraaxonal volume fraction and diffusivities based on a Bayesian approach and machine learning [1]. The major novelty of the method lies in the fast and direct estimation of the volume fractions for multi-compartmental tissue models with minimal constraints and without the detour via the cumulants such as the diffusion and kurtosis tensors. In this work, we report the application to acute ischemic stroke cases.

The microstructure determination method is explained in detail in the abstract with submission number 2120. In brief, the method relies on a machine learning algorithm based on features derived directly from the DWI signal. The features are chosen such that they are invariant to fiber orientation and dispersion. With this method, parameters of the tissue microstructure such as intra / extraaxonal volume fractions and diffusivities can be obtained within seconds. The presently used microstructural model includes an intraaxonal compartment, an extraaxonal compartment, and a fraction of CSF (Figure 1).

We present measurements of 3 acute stroke patients, performed on a SIEMENS Prisma, 3Tesla. Perfusion was measured with a DSC protocol (TE=35ms, TR=1800ms, 2.3x2.3x5mm3, 0.1mmol/Kg Gadovist). ADC was calculated from the standard DWI protocol (TE=85ms, TR=4200ms, 0.6x0.6x5mm3, acquisition time 1min 20s). An additional multishell DWI scan was performed with TE=87ms, TR=3200ms, 1x1x5mm3, 30 diffusion-directions in a hexagonal q-space scheme, acquisition time 2min. The microstructural parameters (intraaxonal, extraaxonal and CSF volume fractions, intra- and extraaxonal diffusivities) were calculated with the method described in [1].

Figure 2-4 show exemplary results for three cases. Compared to the standard ADC, the infarct core stands out much better in the microstructural parameter maps, especially in Dax_intra, the intraaxonal diffusivity parallel to the fibers. This emphasizes the capability of the proposed method to separate the tissue compartments. In all cases, obviously, the infarct core shows a very strong decrease in the intraaxonal diffusivity, and increase in intraaxonal volume. The biophysical origin for this somewhat counterintuitive feature has been under debate for decades. Our findings are in line with previous measurements of mean kurtosis in stroke [2] and support the relatively new picture of axonal beading [3, 4]. In this picture, it is assumed that the change in cellular osmolarity due to cell death leads to a non-uniform swelling of the axons (see Figure 5). This leads to an increase in the intraaxonal volume fraction, and a decrease in diffusivity parallel to the fibers, as their former cylindrical shape is strongly distorted.

The presented microstructural parameters should be currently regarded as generic biomarkers until their precise biophysical correlates and possible interdependencies are investigated deeper.

The robustness of the method allows for applications to data with only few diffusion directions, and can hence be applied to protocols as the presented one with measurement times on the order of 2 minutes. The data processing time is also very short (order of seconds), which is crucial in acute stroke.

In this work, we presented the first application of a fast method for estimating tissue microstructure parameters to stroke. With the proposed method, the diffusional properties of intra and extraaxonal space can directly be separated. As in stroke, the major diffusional changes happen in the intraaxonal space only, this allows for a much clearer and more natural identification of the infarct core compared to standard the ADC. The results are in line with the common picture of axonal beading. Due to the short measurement and data processing time, the method is readily applicable in clinical routine.