Diffusion MRI-based cortical connectome reconstruction: dependency on tractography procedures and neuroanatomical characteristics

Abstract

Diffusion MRI (dMRI)-based tractography offers unique abilities to map whole-brain structural connections in human and animal brains. However, dMRI-based tractography indirectly measures white matter tracts, with suboptimal accuracy and reliability. Recently, sophisticated methods including constrained spherical deconvolution (CSD) and global tractography have been developed to improve tract reconstructions through modeling of more complex fiber orientations. Our study aimed to determine the accuracy of connectome reconstruction for three dMRI-based tractography approaches: diffusion tensor (DT)-based, CSD-based and global tractography. Therefore, we validated whole brain structural connectome reconstructions based on ten ultrahigh-resolution dMRI rat brain scans and 106 cortical regions, from which varying tractography parameters were compared against standardized neuronal tracer data. All tested tractography methods generated considerable numbers of false positive and false negative connections. There was a parameter range trade-off between sensitivity: 0.06-0.63 interhemispherically and 0.22-0.86 intrahemispherically; and specificity: 0.99-0.60 interhemispherically and 0.99-0.23 intrahemispherically. Furthermore, performance of all tractography methods decreased with increasing spatial distance between connected regions. Similar patterns and trade-offs were found, when we applied spherical deconvolution informed filtering of tractograms, streamline thresholding and group-based average network thresholding. Despite the potential of CSD-based and global tractography to handle complex fiber orientations at voxel level, reconstruction accuracy, especially for long-distance connections, remains a challenge. Hence, connectome reconstruction benefits from varying parameter settings and combination of tractography methods to account for anatomical variation of neuronal pathways.

Keywords: Brain; Brain connectomics; Constrained spherical deconvolution; Diffusion MRI; Diffusion tractography; Neuronal tracers; Rats.

Conflict of interest statement

Conflict of interest

None of the authors has any conflict of interest to disclose in relation to this work.

Research involving animals

All animal procedures were approved by the Animal Experiments Committee of the University Medical Center Utrecht and Utrecht University, and experiments were performed in accordance with the guidelines of the European Communities Council Directive.

Figures

Fig. 1

Comparison of connectivity networks from neuronal tracer database and diffusion tractography algorithms. Neuronal tracer-based (left column) and diffusion tractography-based (middle column) connectivity networks represented as network graphs (top), in which nodes represent cortical atlas regions (N = 106) and edges represent connections, and as adjacency matrix (bottom), in which rows and columns represent cortical regions and dark squares represent connections. Diffusion tractography-based connectivity networks were compared against the neuronal tracer-based network as ground truth, which yielded true positives (green lines and squares), false positives (red lines and squares), false negatives (dotted red lines and squares), and true negatives (no line and color-coding) (right column).

Fig. 2

Tractography from high-resolution diffusion MRI of postmortem rat brain. Top: coronal rat brain slice displaying fiber orientation distributions, with an enlarged view of the dorsal hippocampal area. Bottom: representative examples of tract reconstructions in the dorsal hippocampal area, computed with diffusion tensor-based (DT left), constrained spherical deconvolution-based (CSD middle) and global tractography algorithms (GT right)

Fig. 3

Connectome reconstruction sensitivity, specificity and Jaccard index of DT-based (left), CSD-based (middle) and global tractography (GT) (right). Left and middle graphs: reconstruction sensitivity (true positive rate; TPR) versus 1-specificity (false positive rate; FPR) (top) and Jaccard index (bottom) over FA thresholds (DT-based tractography) and over FOD thresholds (CSD-based tractography), for different angle thresholds (line color) with default step size and 250,000 streamlines. Right graphs: GT-based reconstruction sensitivity versus 1-specificity (top) and Jaccard index (bottom) over connection potentials for different particle potentials (line color). All parameters are plotted for interhemispheric (solid lines) and intrahemispheric (dashed lines) connections separately

Fig. 4

Connectome reconstruction sensitivity, specificity and Jaccard index of DT-based and CSD-based tractography, with and without SIFT correction. Reconstruction sensitivity (true positive rate; TPR) versus 1-specificity (false positive rate; FPR) (top) and Jaccard index (bottom) over FA thresholds (DT and DT-SIFT) and over FOD thresholds (CSD and CSD-SIFT) for different angle thresholds (line color) with default step size and 250,000 streamlines. All parameters are plotted for interhemispheric (solid lines) and intrahemispheric (dashed lines) connections separately

Fig. 5

Reconstruction sensitivity, specificity and Jaccard index of DT-based (left), CSD-based (middle) and global tractography (GT) (right). Sensitivity (top), specificity (middle) and Jaccard index (bottom) over Euclidean distance (mm) for DT-based (step size = 15 µm, FA threshold = 0.15) and CSD-based tractography (step size = 75 µm, FOD threshold = 0.125) with 250,000 streamlines and different angle thresholds (line color), and for GT (connection potential = 1) with different particle potentials (line color)

Fig. 6

Reconstruction sensitivity, specificity and Jaccard index of DT-based tractography at different group-based incidence thresholds and streamline thresholds. Left graph: reconstruction sensitivity (true positive rate; TPR) versus 1-specificity (false positive rate; FPR) (top) and Jaccard index (bottom) over group incidence thresholds with different angle thresholds (line color) for DT-based tractography (step size = 15 µm, FA threshold = 0.15 and 250,000 streamlines) (left graphs). Right graph: Reconstruction sensitivity (true positive rate; TPR) versus 1-specificity (false positive rate; FPR) (top) and Jaccard index (bottom) over streamline thresholds for DT-based tractography (red; step size = 15 µm, FA threshold = 0.15), CSD (green; step size = 75 µm, FOD threshold = 0.125), with an angle threshold of 40° and 250,000 streamlines, and for global tractography (GT) (blue; connection potential = 1, particle potential = 0.01). All parameters are plotted for interhemispheric (solid lines) and intrahemispheric (dashed lines) connections separately