The mean kurtosis value reflects the underlying micro-structures. "To date there is no well-defined physiological model which makes it possible to simulate the results of kurtosis measurements due to the complexity of the problem and the high number of factors intervening.  There is no clear description or simple link between kurtosis and the micro-elements that form the structure of tissues (obstacles). However, measurements carried out in different situations, and among different populations, make it possible to see that the kurtosis varies locally in some areas of the brain", states Evelyne Balteau. The idea consists in using the mean kurtosis as a biomarker which makes it possible to more reliably diagnose diseases such as Alzheimer’s or Parkinson’s and to assess more accurately the stage the disease has reached thanks to the more sensitive and potentially more relevant information that this technique offers with regard to the integrity of white matter. 

"Kurtosis measurements among different populations make it possible to create a dictionary of kurtosis identity patterns for each population. Kurtosis could therefore be used as a diagnostic tool in conjunction with and complementary to other existing tools. Yet kurtosis is not the only biomarker of interest in diffusion MRI, other models are focusing on e.g. the 3D tracking of myelin fibers which connect the different parts of the brain to each other and to the rest of the body. Kurtosis is one aspect of the pattern, complementary to other features and parameters collected during cognitive assessments, and helping building a more general picture of a pathological situation and its evolution", states Evelyne Balteau.

Quiet, no noise please...

The major interest of Elodie André’s study lies in her contribution to image quality which makes it possible to go further, acquiring more valuable data in a higher number of diffusion directions. "For each diffusion direction, an image of the brain is obtained. The signal drops as a function of the diffusion weighting we impose: in DKI, a more important weighting is applied. The signal received is therefore weaker, more ‘noisy’. The objective of my research is to find a method for improving the quality of the signal and, therefore, to increase the quality of the result" 

Her idea consisted in extracting the signal from this noise by using what are called the moments of the distribution. “On one hand, we measure the noise, on the other, we remove it from the signal”, she clarifies. To succeed, she designed a quite rapid method. "It is not yet perfect” she says, “but it already gives a more precise signal”. The two experiments she carried out in her study confirmed this.

Eliminating dependency

The first experiment was carried out by positioning the head of the subject in different ways, more or less close to the MRI coil (the radio-frequency (RF) antenna collecting the signal). "This made it possible to produce a distinct spatial distribution of the signal-to-noise ratio (SNR) for each acquisition, because the elements of the RF coil have a sensitivity that decreases with distance. The signal collected is therefore higher in the areas of the brain closer to one of these elements and weaker for a more distant region. In principle, for a given individual, the results obtained in two different positions should be identical. But in practice, due to the difference in SNR , this is not the case. However, with my correction, we get the same parameters. This indicates that with this method, we are no longer dependent on the SNR and we see identical kurtosis maps after correction".

The second experiment was aimed at the variability of results between subjects, in order to verify whether the correction reduced the variability due to the low signal-to-noise ratio. Twenty-five individuals took part in the experiment. As figure 9 presented in the study shows, without correction, kurtosis is artificially higher and variability is more important (see the first column) in particular at the front of the brain (top of the slice), where the signal-to-noise ratio was lower. On the other hand, when high-performance correction methods were used (columns 2 and 3), the contrast was better (the white matter is represented in yellow) and the variability due to the low signal-to-noise ratio is reduced. "We can conclude that using the correction made it possible to discard differences due to noise, while demonstrating the finer and more interesting differences between groups of subjects. It is then possible to use fewer subjects to draw solid conclusions and avoid false positives (with erroneous results due to noise)", explains Elodie André.