Tommaso Pavan

Biophysical diffusion MRI models such as Neurite Exchange Imaging (NEXI) are essential for probing gray matter microstructure, estimating compartment diffusivities (Di, De), neurite fraction (f), and exchange time (tex). However, NEXI multi-shell, multi-diffusion-time requirements lead to prohibitively long acquisitions. Leveraging the Connectome 2.0 ultra-high gradient scanner, we developed a time-efficient protocol using an Explainable AI (XAI) framework. Combining XGBoost, SHAP, and Recursive Feature Elimination trained on synthetic signals, XAI identified an optimal 8-feature subset, halving scan time from 27 to 14 minutes. Validated in vivo in 7 healthy participants, the XAI protocol was benchmarked against the full 15-feature acquisition, a Cramér-Rao Lower Bound (CRLB) theoretical optimum, and 2 expert-designed heuristics (“Mid-Range” and “Corner”). The XAI protocol robustly reproduced parameter estimates and maintained full protocol test–retest reproducibility. Remarkably, the data-driven XAI selection converged to the theoretical CRLB optimum. This validates XAI mathematical optimality while highlighting its primary advantage: achieving gold-standard optimization without requiring complex analytical Jacobians, making it effortlessly adaptable to numerically defined models or complex noise distributions where CRLB becomes intractable. Furthermore, XAI demonstrated superior in vivo robustness compared with heuristics: “Mid-Range” sampling yielded biased tex estimates due to insufficient temporal diversity, while “Corner” sampling yielded unstable Di estimates (fivefold higher coefficient of variation) due to noise sensitivity. Ultimately, this robust 14-minute NEXI protocol accelerates exchange-sensitive microstructural mapping, establishing a generalizable, model-agnostic optimization framework adaptable to future ultra-high gradient systems (e.g., Siemens Cima.X or GE MAGNUS) and existing clinical scanners.

While Blood Oxygenation Level-Dependent (BOLD) fMRI remains the gold standard for mapping functional brain networks with MRI, its vascular origins inherently conflate haemodynamic effects with neural activity, limiting its sensitivity in white matter (WM) or its interpretation in neurovascular diseases. Apparent Diffusion Coefficient (ADC) fMRI offers an alternative, diffusion-based contrast that is theoretically more sensitive to neuromorphological coupling and therefore more specific to neuronal activation, though investigated primarily during task-based conditions. This study aimed to comprehensively evaluate the efficacy of isotropic ADC-fMRI in detecting established resting-state networks (RSNs) and to extend this methodology to the investigation of grey-to-white matter (GM-WM) functional connectivity. Our analyses revealed a gradient of ADC detectability shaped by the degree of static functional cohesion and structural tethering of each network. The visual and somatomotor networks, being both highly segregated and strongly anchored to underlying structural pathways, yielded the most robust detection. The default mode network (DMN) and dorsal attention network (DAN) reached group-level significance but with lower effect sizes, and their detection proved fragile across analytical approaches. The frontoparietal network (FPN) and salience network (SAN), whose functional identity is defined by dynamic cross-network reconfiguration, did not reach significance. This gradient partially mirrors the established hierarchy of network segregation observed in BOLD, while further suggesting that ADC sensitivity depends on the structural grounding of each network. Furthermore, ADC demonstrated superior sensitivity to GM-WM functional coupling compared to BOLD. GM-WM functional connectivity profiles derived from ADC were significantly more aligned with underlying structural WM architecture across subjects. Taken together, these findings position isotropic ADC-fMRI as a viable complementary modality to BOLD, offering a more direct window into the neural and structural foundations of brain connectivity.

Biophysical models in diffusion MRI (dMRI) hold promise for characterizing gray matter tissue microstructure. Yet, the reliability of their parameter estimates remains largely under-studied, especially in models that incorporate water exchange. In this study, we investigate the accuracy, precision, and presence of degeneracy of two recently proposed gray matter models, NEXI and SANDIX, using established acquisition protocols, on both simulated and in vivo data. We employ µGUIDE, a Bayesian inference framework based on deep learning, to quantify parameter uncertainty and detect degeneracies, enabling a more interpretable assessment of model fits. Our results show that while some microstructural parameters, such as extra-cellular diffusivity and neurite signal fraction, are robustly estimated, others, including exchange time and soma radius, are often associated with high uncertainty and estimation bias, particularly under realistic noise conditions and reduced acquisition protocols. Comparison with non-linear least squares fitting highlights the critical advantage of uncertainty-aware methods: the ability to flag and filter out unreliable estimates. Together, these findings emphasize the need to report uncertainty and account for model degeneracies when interpreting model-based estimates. Our study advocates for the integration of probabilistic fitting approaches into imaging pipelines to improve reproducibility and biological interpretability.

Recent research highlights the critical role of white matter (WM) alterations in psychosis and schizophrenia (SZ), reporting volumetric and structural brain changes in affected individuals. In this study, we explored the role of astroglia in SZ, which is believed to play a role in white matter integrity. We investigated for the first time the associations between advanced diffusion Magnetic Resonance Imaging (dMRI) measures of WM microstructure and Magnetic Resonance Spectroscopy (MRS)-derived glial markers in 30 subjects with early psychosis (EP, mean age 24 ± 6) versus 49 healthy controls (HC, mean age 25 ± 6). We focused on two metabolites involved in glia: myo-Inositol (myo-Ins) and total Choline (tCho), measured in the medial prefrontal cortex (mPFC), relating them to quantitative dMRI metrics derived from Diffusion Kurtosis Imaging (DKI) and WM Tract Integrity-Watson (WMTI-W) biophysical model, including mean diffusivity and kurtosis, axonal water fraction and extra-axonal diffusivities in the whole white matter. Our findings reveal a difference between EP and HC in WM diffusivities, specifically in the extra-axonal parallel direction, but not in MRS metabolites. However, we found that the mPFC myo-Ins concentrations in EP are exclusively and strongly associated with proximal WM microstructure features, in the form of a positive correlation with axonal water fraction, a proxy for axonal density, and a negative correlation with extra-axonal parallel diffusivity, suggesting the white matter alterations could be linked to astrocytic changes in early psychosis.

The conduction velocity of neuronal signals along axons is a key neurophysiological property that can be altered in various disease processes. While cortico-cortical evoked potentials (CCEPs) can be measured in presurgical assessment to provide information about conduction delay between a subset of brain regions, it is currently not possible to efficiently and systematically estimate conduction velocity in vivo across the whole brain. Given the established link between conduction velocity and axon morphology (most notably axon diameter but also myelination), mapping a reliable and quantitative metric linked to axon properties could fill the gap of inferring conduction velocity across the entire human brain. By integrating multiple MRI-derived microstructural measures – including axon radius, axonal water fraction, extra-axonal perpendicular diffusivity, and longitudinal relaxation time – and conduction velocity estimates obtained from a large database of CCEPs, we developed a whole-brain prediction model of conduction velocity. Our multivariate MRI-based model explained 29% of variance in neurophysiological conduction velocity, making it possible to partially predict whole-brain conduction velocity and delay matrices along connections for which no direct measurement is commonly available from epilepsy surgery investigations. This integrative MRI-based approach could provide a non-invasive framework for comprehensively characterising conduction delays in vivo across the human brain white matter.

Recent research highlights the critical role of white matter (WM) alterations in psychosis and schizophrenia (SZ), reporting volumetric and structural brain changes in affected individuals. In this study, we explored the role of astroglia in SZ, which is believed to play a role in white matter integrity. We investigated for the first time the associations between advanced diffusion Magnetic Resonance Imaging (dMRI) measures of WM microstructure, and Magnetic Resonance Spectroscopy (MRS)-derived glial markers in 30 subjects with early psychosis (EP, 24±6 y.o.) versus 41 healthy controls (HC, 25±6 y.o.). We focused on two metabolites involved in glia: myo-Inositol (myo-Ins) and total Choline (tCho), measured in the medial prefrontal cortex (mPFC), relating them to quantitative dMRI metrics derived from Diffusion Kurtosis Imaging (DKI) and WM Tract Integrity-Watson (WMTI-W) biophysical model including mean diffusivity and kurtosis, axonal water fraction and extra-axonal diffusivities in the whole white matter. Our findings reveal a difference between EP and HC in WM diffusivities, specifically in extra-axonal parallel direction, but not in MRS metabolites. However, we found that the mPFC myo-Ins concentrations in EP are exclusively and strongly associated with proximal WM microstructure features, in the form of a positive correlation with axonal water fraction, a proxy for axonal density, and negative correlation with extra-axonal parallel diffusivity, suggesting the white matter alteration could be linked to astrocytic changes in subset of early psychosis.

Biophysical models of diffusion tailored to characterize gray matter (GM) microstructure are gaining traction in the neuroimaging community. NEXI, SMEX, SANDI, and SANDIX represent recent efforts to incorporate different microstructural features,such as soma contributions and inter-compartment exchange, into the diffusion MRI (dMRI) signal. In this work, we present a comparative evaluation of these four gray matter models on a single, publicly available in vivo human dataset, the Connectome Diffusion Microstructure Dataset (CDMD), acquired with two diffusion times. Using the open-source Gray Matter Swiss Knife toolbox, we estimate cortical microstructure metrics in 26 healthy subjects and evaluate goodness of fit, anatomical patterns and consistency with previous studies. CDMD data yielded GM parameter estimates consistent with values reported in previous studies. This retrospective cross-model analysis establishes the feasibility of estimating exchange models from only two diffusion times and highlights trade-offs in biological specificity, model complexity, and fitting robustness, critical considerations when choosing a model for future clinical and research applications.

Biophysical models of diffusion in gray matter (GM) can provide unique information about microstructure of the human brain, in health and disease. Therefore, their compatibility with clinical settings is key. Neurite Exchange Imaging (NEXI) is a two-compartment model of GM microstructure that accounts for inter-compartment exchange, whose parameter estimation requires multi-shell multi-diffusion time data. In this work, we report the first estimates of NEXI in human cortex obtained on a clinical MRI scanner. To do that, we establish an acquisition protocol and fitting routine compatible with clinical scanners. The model signal equation can be expressed either in the narrow-pulse approximation, NEXI, or accounting for the actual width of the diffusion gradient pulses, SMEX. While NEXI enables a faster analytical fit and is a valid approximation for data acquired on high-performance gradient systems (preclinical and Connectom scanners), on which NEXI was first implemented, SMEX has significant relevance for data acquired on clinical scanners with longer gradient pulses. We establish that, in the context of broad pulses, SMEX estimates were more comparable to previous literature values. Furthermore, we evaluate the repeatability of NEXI estimates in the human cortex on a clinical MRI scanner and show intra-subject variability to be lower than inter-subject variability, which is promising for characterizing healthy and patient cohorts. Finally, we analyze the relationship of NEXI parameters on the cortical surface to the Myelin Water Fraction (MWF), estimated using an established multicomponent T2 relaxation technique. Indeed, although it is present in small quantities in the cortex, myelin can be expected to decrease permeability. We confirm a strong correlation between the exchange time (tex) estimates and the MWF, although the spatial correspondence between the two is brain-region specific and other drivers of tex than myelin density are likely at play.