What is NODDI and what is its role in Parkinson’s assessment?

Dendrites and axons, known as neuronal processes or neurites, constitute the basic cellular units of the brain’s computational circuits. Dendrite function is reflected by their branching complexity (hence density). Cortical areas with simple and complex dendritic structures are involved in the early and late stages, respectively, of information processing [1]. Neurite structure dynamically changes with development [2] and aging [3], and degenerates in various neurological disorders including Alzheimer’s disease [4] and amyotrophic lateral sclerosis [5]. Dendritic spine loss and shortening in the substantia nigra (SN) and striatum are reported in Parkinson’s disease (PD), a representative neurodegenerative disorder [6]. Hence these changes, which may be the primary causes of motor symptoms in PD, are thought to reflect the PD disease state.

Magnetic resonance imaging (MRI) is an excellent tool for noninvasive in vivo brain assessment. However, the spatial resolution with conventional MRI is at best a few millimeters, while microscopic nerve tissue evaluation in the brain requires submicron scale resolution. Therefore, diffusion MRI (dMRI) was developed to resolve the micron-scale displacement of water molecules diffusing in tissue and to evaluate neural tissue microstructure. Water molecule displacement in dMRI-targeted tissues is strongly influenced by the direction, permeability, and number of barriers (cell membranes and myelin sheaths) in the tissue, and by differences in intracellular proteins and organelles in neurons, dendrites, axons, neurofilaments, and microtubules. Thus, dMRI can indirectly infer the microstructures of these barriers and organelles [7].

Although many dMRI methods are proposed, the most representative models are diffusion tensor imaging (DTI) [8], diffusional kurtosis imaging (DKI) [9], and neurite orientation dispersion and density imaging (NODDI) [10]. DTI, which is widely utilized in clinical and research settings, has two major limitations. First, the assumption in DTI that water molecule diffusion follows a Gaussian distribution may be inappropriate because water molecule diffusion in actual living tissue is restricted by cell membranes and organelles, thereby showing a non-Gaussian distribution of displacement. Second, interpreting DTI parameter estimations, e.g. fractional anisotropy (FA), which reflects neuronal density, fiber orientation dispersion (OD), axonal diameter, and myelination degree, is difficult. Because FA changes are nonspecific, there are limits for using this parameter to infer pathological changes [11]. DKI was proposed to quantify the non-Gaussian distribution of water molecules in complicated structures where diffusion is restricted by cell membranes and organelles [9], and made it possible to quantify structural complexity. DKI parameters were at least as sensitive as DTI parameters, if not more so, to microstructural changes; however, DKI parameter changes are difficult to interpret because they are nonspecific. Lately, NODDI was proposed to overcome these limitations (of the Gaussian distribution model assumed in DTI and the nonspecific changes in DTI and DKI parameters). NODDI assumes a three-compartment biophysical tissue model including intracellular, extracellular, and cerebrospinal fluids in a single voxel, which enables the inference and quantification of the direction and structure of neurites (axons and dendrites) by the orientation-dispersed cylinder model and the Watson distribution. Representative NODDI parameters include the intracellular volume fraction (Vic), indicating neurite density, and the OD index indicating the neurite OD.

For reference, we indicate the current NODDI imaging conditions at our facility. Diffusion-weighted data for NODDI were acquired with a spin-echo planar imaging sequence along 32 isotropic diffusion gradient directions. For each direction, diffusion-weighted data were acquired with three b values (0, 1000, and 2,000 s/mm²). Sequence parameters were as follows: repetition time, 9810 ms; echo time 100 ms; field of view, 256 mm; matrix, 128 × 128; slice thickness, 2 mm; and imaging time, 13 min 8 s. Furthermore, using the NODDI Matlab Toolbox5 (http://www.nitrc.org/projects/noddi_toolbox), the diffusion-weighted data were fitted to an NODDI model. In assessing Vic maps calculated by NODDI, the very bright voxels (indicating unusually high-neurite densities) appearing near the border between the cerebrospinal fluid (= iso diffusion) and cerebral parenchyma required attention because these are voxels in which the tissues were not accurately inferred. Accelerated Microstructure Imaging via Convex Optimization (AMICO) [12], a recently proposed analysis algorithm, ameliorates this problem and greatly reduces the NODDI calculation time approximately 1/40–1/80 of the original time. Hence, AMICO use is recommended. To date, dMRI studies of PD are typically performed using DTI and DKI techniques. However, reliability sufficient for introducing these techniques in routine clinical practice has yet to be achieved [13,14]. Therefore, we investigated and compared the degeneration of the substantia nigra pars compacta (SNc) and basal ganglia in 58 PD patients [15] to 36 healthy subjects. We found significant reductions in the OD and Vic in the SNc and a significant decrease in the putamen OD. Because more than half of the SNc dopamine neurons are lost by the time of PD onset, the reduction in Vic is thought to reflect a decrease in the SNc neurite density. A reduction in the SNc and putamen ODs are thought to reflect reductions in dendritic length and number of spines. Expectedly, reductions of Vic and OD in the SNc and putamen were more pronounced on the side contralateral to the symptom-dominant side. Receiver operating characteristic (ROC) analyses indicated that the SNc Vic on the side contralateral to the symptomatic side showed the best diagnostic ability (areas under the ROC curve (AUC), 0.92; mean cutoff, 0.62; sensitivity, 0.88; specificity, 0.83), suggesting this index to be a useful tool for future PD diagnosis. Furthermore, because the Vic and OD in the SNc and putamen were significantly negatively correlated with disease duration and the unified Parkinson’s disease rating scale-3 motor score, these indices may be useful to assess disease progression.

Although our previous study suggested that NODDI is useful for PD diagnosis and disease progression assessment, it remains the only study to apply NODDI to PD; thus, evidence is still lacking, and result verification on large-scale studies by multicenter collaborations and other arrangements will be necessary. Because NODDI is merely a model to estimate the neurite structure and orientation using the orientation-dispersed cylinder model and Watson distribution, comparisons with pathological data from postmortem brains or PD model mice will be necessary. Furthermore, the degree to which differences in the manufacturer/model or static magnetic field intensity (1.5 T versus 3 T) of the MRI scanner affect the NODDI parameters’ reproducibility is not adequately tested. Therefore, testing reliability across MRI scanners also remains an issue because differences in MRI scanners may impact the ability to diagnose PD.

Additionally, numerous previous DTI/DKI studies have confirmed a high degree of diagnostic accuracy of the technique for PD. Using DTI, Vaillancourt et al. showed FA reduction in the SN caudal region in PD, and reported that the technique could perfectly differentiate early untreated PD patients (n = 14) from healthy controls [16]. Using DKI, Wang et al. showed that the mean kurtosis in the SN and putamen (derived from DKI data) for PD patients (n = 30) increased compared with those on the same side among healthy controls (n = 30), thereby achieving high diagnostic ability (AUC = 0.95) [17]. These studies demonstrated greater diagnostic ability than that in our study (AUC = 0.92). However, the report by Vaillancourt et al. was based on a small number of cases, and the report by Wang et al. showed greater changes in the SN and putamen on the side ipsilateral to the symptom-dominant side, contradicting the clinical hypothesis. Moreover, instead of decreasing, FA in the SN increased in the report by Wang et al. Hence, DTI/DKI/NODDI will need to be verified in large-scale multicenter studies of PD cohorts. Furthermore, it may be possible to achieve even better ability to diagnose PD in our research by measuring the Vic and OD in the caudal region, which shows pronounced pathological degeneration in PD.

NODDI is likely to provide effective biomarkers for early PD diagnosis and disease progression monitoring as well as the evaluation of drug efficacy and elucidation of various disease states. For instance, PD is treated by L-dopa (dopamine precursor) administration and dopamine release promotion from the residual neurons. However, L-dopa administration itself may cause dyskinesia and marked impairments in the activities of daily living. Recently, L-dopa-induced dyskinesia (LID) was found to be associated with the structural plasticity of striatal dendritic spines [18]. Striatum evaluation by NODDI may be useful to elucidate the disease state and predict LID onset.

Furthermore, the development of disease-modifying drugs and neuroprotective agents is highly anticipated because most current PD therapies involve improving symptoms by administering L-dopa and dopamine agonists [19]. If such agents can be used in clinical settings, NODDI, which makes it possible to infer neurite density and structure in the SN and basal ganglia, may also become useful to evaluate drug efficacy. If cell transplantation therapy becomes possible by pluripotent stem cell inductions [20], it may also become possible to evaluate the survival of transplanted cells or neurite outgrowth in vivo. In conclusion, NODDI may be a promising MRI technique for monitoring disease progression and providing biomarkers for early PD diagnosis, and may also be used to provide surrogate markers for elucidating the LID disease state and evaluating the treatment efficacy.

Financial and competing interests disclosure

This work was supported by a research grant from Hitachi, Ltd.; the program for Brain Mapping by Integrated Neurotechnologies for Disease Studies (Brain/MINDS) of the Japan Agency for Medical Research and Development (AMED); Technology of Japan; and by MEXT/JSPS KAKENHI Grant Number 26893266. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.