Neuroimaging enables detailed investigation into the evolution of the structure and function of the human brain over its lifespan^1-2. Diffusion Tensor Imaging (DTI)³ in particular has provided unique insight into the lifespan changes of white matter (WM) through its sensitivity to tissue microstructure^4-9. While DTI is sensitive, the technique lacks in specificity to tissue composition. For example, changes to fractional anisotropy (FA), a common index derived from DTI, might be attributable to alterations in axonal density or axonal orientation distribution. Advances in diffusion MRI acquisition and modeling now support much more specific characterization of tissue microstructure, offering an exciting opportunity to refine our understanding of WM microstructural evolution.

The primary aim of this work is to chart the lifespan trajectory of WM microstructure using Neurite Orientation Dispersion and Density Imaging (NODDI)¹⁰, a state-of-the-art multi-compartment model of diffusion, using HCP lifespan dataset (a state-of-the-art multi-shell diffusion MRI and high spatial resolution data). Our second aim is to evaluate Accelerated Microstructure Imaging via Convex Optimization (AMICO)¹¹, a new framework to accelerate the fitting of advanced multi-compartment diffusion models (e.g., NODDI) in a practical neuroimaging application.

Subjects

6 subjects per age group (age 8-9 and 14-15) and 5 subjects per age group (age 25-35, 45-55, and 65-75) from HCP lifespan Phase 1a dataset.

MRI acquisition

The acquisition parameters were: voxel size 1.5mm isotropic; b=1000 and 2500 s/mm2, each with 75 directions; Acquisition time 21 minutes¹².

Image Processing

The diffusion MRI images underwent standard HCP preprocessing pipeline¹³. The NODDI indices, including neurite density index (NDI), orientation dispersion index (ODI) and volume fraction of free water (FISO), were computed separately using NODDI matlab toolbox¹⁴, which implements the standard nonlinear least square algorithm, and AMICO. Spatial normalization was achieved with DTI-TK¹⁵.

Image Analysis

We extracted the mean NODDI indices of WM at two spatial levels, on WM skeleton created with FSL TBSS¹⁶ and over the core WM ROI defined as the sum of the 48 ROIs in the JHU WM altas¹⁷. The average of these indices for each age group was used to determine the lifespan trajectory. Correlation analysis and Bland-Altman plot were used to evaluate the NODDI indices from AMICO against the standard fitting.

Lifespan trajectory of NODDI indices on the WM skeleton and core WM ROI

At both spatial levels (Figures 1a-1b), NDI showed a clear pattern of inverted U shape over lifespan, increasing from the childhood, plateauing around the early adulthood, then decreasing following the late adulthood. ODI showed a more stable pattern throughout the lifespan (Figures 1e-1f). FISO varied little before 60s and its subsequent increase (Figures 1c-1d) is consistent with the onset of aging.

Evaluation of AMICO fitting across subjects

Results at both spatial level showed NODDI indices estimated by AMICO were highly consistent with the values determined by the standard fitting. Most importantly, AMICO reproduced the same lifespan trajectory as the standard fitting. For all NODDI indices on the WM skeleton, all of the R2 were over 0.97. For NDI and FISO in the core WM ROI, both of the R2 were over 0.98 while for ODI, the R2 was 0.88. Bland-Altman plots showed that AMICO estimates consistently underestimated the NODDI indices compared to the standard fitting. Nevertheless, this did not prevent AMICO from recovering the same lifespan trajectory as the standard fitting.

Required time for NODDI fitting

For the HCP lifespan dataset, the standard NODDI fitting required nearly 230 hours to complete the parameter estimation of a single subject while AMICO took about one hour. This clearly demonstrated the benefits of AMICO for large-scale studies.

Taking advantage of the state-of-the-art HCP lifespan dataset, we showed that neurite density estimated with the NODDI model changes across the human lifespan in an inverted U shape. This is consistent with DTI findings over lifespan^1-5 and extends the recent NODDI findings for brain maturation¹⁸. This study additionally demonstrated, in a neuroimaging application, that AMICO provides NODDI indices consistent with the standard fitting. Despite a slight underestimation bias, it can recover the same lifespan trajectory as the standard fitting. The results support the use of AMICO for NODDI fitting in large cohort studies where the standard NODDI fitting is impractical.