Typically quantified by the apparent diffusion coefficient (ADC), diffusion MRI measures water diffusion in biological tissues which in turn is seen as a probe of microstructural changes derived from pathology. However, the ADC is confounded by a number of processes and phenomena rendering it sub-optimal for the characterisation of brain tissues. Additionally, the even more complex micro-environments of pathologies like brain tumours do not lend themselves to be properly summarised by a single value. In this work, two methods for the acquisition and analysis of multi-b-value dMRI data were developed, expanding the ADC formalism and keeping in mind clinical feasibility. The first is a unified model for the measurement of perfusion, and Gaussian and non- Gaussian diffusion. This is achieved by sampling the tissue at 16 unique b-values, from 0 to 3,000 s/mm2, while including the clinically established b = 1,000 s/mm2. With an acquisition time of 4 minutes and 19 seconds, four quantities can be derived from this protocol: pseudo-diffusivity and perfusion fraction, characterising the perfusion regime; apparent diffusivity, describing Gaussian diffusion; and apparent kurtosis, as a measure of non-Gaussian diffusion. This method was validated in vivo in a brain tumour patient cohort by pinning the derived quantities against their counterparts from established protocols. The metrics characterising perfusion did not correlate with their canonical counterparts, which suggests in part that these reflect different phenomena. Conversely, the Gaussian and non-Gaussian quantities proposed were highly correlated with those from the canonic protocols, indicating that the clinical usefulness of the canonic quantities is preserved when using the proposed metrics. The second method is based on a parameter-free fitting of dMRI data with b-values ranging up to 10,000 s/mm2, using non-negative least squares. This method provides with a diffusivity spectrum, as opposed to a fixed set of diffusivity pools and their relative fractions, allowing for a richer characterisation of the different diffusivity regimes in the tissue. Data was also fitted using established methods such as mono-, bi-, and tri-exponential decays in order to establish the added value of the proposed method. This method was demonstrated in a broad brain tumour cohort, where tissue specific spectra were derived: grey matter, white matter, cerebrospinal fluid, oedema, tumour, and restricted diffusion lesions. The latter develops as a consequence of the disease but it is yet poorly understood. The proposed method is applied to try to characterise this lesion. The suggested method was shown to better fit the data than the fixed-term models and the derived spectra show distinct differences in both healthy and pathologic tissue. In conclusion, this work developed methods for the multi-parametric characterisation of diffusion in biological tissues, evaluated in brain tumour patients.
Ricardo Loução (Wed,) studied this question.