Cognitive and White-Matter Compartment Models Reveal Selective Relations between Corticospinal Tract Microstructure and Simple Reaction Time

Abstract

The speed of motor reaction to an external stimulus varies substantially between individuals and is slowed in aging. However, the neuroanatomical origins of interindividual variability in reaction time (RT) remain unclear. Here, we combined a cognitive model of RT and a biophysical compartment model of diffusion-weighted MRI (DWI) to characterize the relationship between RT and microstructure of the corticospinal tract (CST) and the optic radiation (OR), the primary motor output and visual input pathways associated with visual-motor responses. We fitted an accumulator model of RT to 46 female human participants' behavioral performance in a simple reaction time task. The non-decision time parameter (T_er) derived from the model was used to account for the latencies of stimulus encoding and action initiation. From multi-shell DWI data, we quantified tissue microstructure of the CST and OR with the neurite orientation dispersion and density imaging (NODDI) model as well as the conventional diffusion tensor imaging model. Using novel skeletonization and segmentation approaches, we showed that DWI-based microstructure metrics varied substantially along CST and OR. The T_er of individual participants was negatively correlated with the NODDI measure of the neurite density in the bilateral superior CST. Further, we found no significant correlation between the microstructural measures and mean RT. Thus, our findings suggest a link between interindividual differences in sensorimotor speed and selective microstructural properties in white-matter tracts.SIGNIFICANCE STATEMENT How does our brain structure contribute to our speed to react? Here, we provided anatomically specific evidence that interindividual differences in response speed is associated with white-matter microstructure. Using a cognitive model of reaction time (RT), we estimated the non-decision time, as an index of the latencies of stimulus encoding and action initiation, during a simple reaction time task. Using an advanced microstructural model for diffusion MRI, we estimated the tissue properties and their variations along the corticospinal tract and optic radiation. We found significant location-specific correlations between the microstructural measures and the model-derived parameter of non-decision time but not mean RT. These results highlight the neuroanatomical signature of interindividual variability in response speed along the sensorimotor pathways.

Keywords: NODDI; along tract analysis; cognitive model; microstructure; non-decision time; simple reaction time.

Figure 1.

A, Experimental paradigm of the SRT task. Participants were instructed to respond when a filled circle appeared over the index finger in the hand picture. B, Exemplar time course of the LBA model. The LBA assumes that accumulated evidence for an action decision is accumulated linearly over time, and a decision is made once the accumulated evidence reaches a threshold. In each trial, the starting point of the accumulation process is sampled from a uniform distribution. The rate of accumulation is sampled from a Gaussian. The model predicted RT includes the duration of the accumulation process and a non-decision time (T_er), that accounts for the latencies of non-decision processes such as stimulus encoding and action initiation, which are shown as shaded area. C, Violin plots (mean and density) of the mean RT and T_er across participants. D, The fit of the LBA model to RT quantiles in the SRT task. Error bars denote the 95% confidence intervals.

Figure 2.

Analysis pipeline of along-tract microstructural metrics. A, After preprocessing, whole-brain voxelwise NODDI (NDI and ODI) and DTI (FA and MD) metrics were calculated from the two-shell DWI data. B, Probabilistic tractography of CST and OR were calculated and thresholded in individual participant's native space. Seed (green), waypoint (purple), and termination masks (blue) used for the tractography are overlaid for both tracts. C, Individual tractography results of CST and OR were normalized to the MNI space and united to obtain group-based tract images. The voxel intensity in the group-based tract images denotes the proportion of overlapping across participants. D, Isolated voxels (red) with no second-order neighbors were removed from the group-based tract volumes. E, Three-dimensional group-based tract volume image was skeletonized using thinning algorithms. Subsidiary branches of the skeleton were trimmed. F, The main skeleton was smoothed with cubic spline interpolation. The central portion of the group-based tract volume (between the two blue planes shown in C) and the skeleton were clipped for further processing. G, The skeletons of the CST and OR were divided into 30 segments with equal lengths and 20% overlap between adjacent segments. Voxels in each group-based tract volume were assigned to the nearest segment based on Euclidean distance, resulting in 30 equidistant sub-volumes along the principal skeleton. Each sub-volume was transformed back to the individual's native space to calculate microstructural metrics along tracts.

Figure 3.

A, Group-based image of the CST. The voxel intensity denotes the proportion of overlapping across participants. B, Visualization of the CST with the skeleton and the approximate locations of segments 1, 10, 20, and 30. C, The means of NDI, ODI, MD, and FA profiles of the left and right CST across participants. Shaded areas represent 95% CI. The segments 1–30 were from CP to SCR as in B.

Figure 4.

A, Group-based image of the OR. The voxel intensity denotes the proportion of overlapping across participants. B, Visualization of the OR with the skeleton and the approximate locations of segments 1, 10, 20, and 30. C, The means of NDI, ODI, MD, and FA profiles of the left and right OR across participants. Shaded areas represent 95% CI. The segments 1–30 were from V1 to posterior LGN as in B.

Figure 5.

Standardized linear regression coefficients between microstructural metrics and T_er (A) and mean RT (B). The segments with significant associations after TFCE (p < 0.05; 10,000 permutations) are presented with increased line thickness. C, CST segments that had significant negative NDI-T_er correlations were highlighted (blue) over the group-based tract volume.