An anatomically curated fiber clustering white matter atlas for consistent white matter tract parcellation across the lifespan

Abstract

This work presents an anatomically curated white matter atlas to enable consistent white matter tract parcellation across different populations. Leveraging a well-established computational pipeline for fiber clustering, we create a tract-based white matter atlas including information from 100 subjects. A novel anatomical annotation method is proposed that leverages population-based brain anatomical information and expert neuroanatomical knowledge to annotate and categorize the fiber clusters. A total of 256 white matter structures are annotated in the proposed atlas, which provides one of the most comprehensive tract-based white matter atlases covering the entire brain to date. These structures are composed of 58 deep white matter tracts including major long range association and projection tracts, commissural tracts, and tracts related to the brainstem and cerebellar connections, plus 198 short and medium range superficial fiber clusters organized into 16 categories according to the brain lobes they connect. Potential false positive connections are annotated in the atlas to enable their exclusion from analysis or visualization. In addition, the proposed atlas allows for a whole brain white matter parcellation into 800 fiber clusters to enable whole brain connectivity analyses. The atlas and related computational tools are open-source and publicly available. We evaluate the proposed atlas using a testing dataset of 584 diffusion MRI scans from multiple independently acquired populations, across genders, the lifespan (1 day-82 years), and different health conditions (healthy control, neuropsychiatric disorders, and brain tumor patients). Experimental results show successful white matter parcellation across subjects from different populations acquired on multiple scanners, irrespective of age, gender or disease indications. Over 99% of the fiber tracts annotated in the atlas were detected in all subjects on average. One advantage in terms of robustness is that the tract-based pipeline does not require any cortical or subcortical segmentations, which can have limited success in young children and patients with brain tumors or other structural lesions. We believe this is the first demonstration of consistent automated white matter tract parcellation across the full lifespan from birth to advanced age.

Figure Appendix C.1:

(a) Overview of the computation of the TAP_atlas of one atlas cluster c. This process used the full clustered tractography data (approximately 1 million fibers per subject) from the 100 HCP-atlas subjects (Section 2.1.1). For each Freesurfer segmented region, the percentage of fibers in c that intersect this region (across the 100 subjects) is computed. A high percentage value indicates that a Freesurfer segmented region is highly intersected by the cluster across all subjects. Then, a threshold (see Figure Appendix C.2) is applied to this percentage to determine the set of most highly intersected regions, i.e. the TAP of the atlas cluster (TAP_atlas(c)). As an example, the sample atlas cluster has a TAP containing five Freesurfer regions (as listed in the figure). (b) Illustration of the computation of TAPC (TAP coherence) for a subject-specific cluster. For each fiber f within the cluster, the set of Freesurfer regions through which the fiber passes is measured. This gives the TAP of the fiber (TAP(f)). Then, a Dice score between the fiber’s TAP (TAP(f)) and the atlas cluster’s TAP (TAPatlas) is computed. For example, the sample fiber (red) has a Dice score of 0.73. Then, the TAPC is calculated as the mean Dice score across all fibers in the subject-specific cluster.

Figure Appendix C.2:

Threshold determination for defining tract anatomical profile. Th = 40% generated the highest Dice scores across the different settings under different parcellation scales.

Figure 1:

Method overview. Sub-figure (a) shows the data-driven fiber clustering atlas generation process. Given the input tractography (a.1) from the 100 atlas subjects, a groupwise whole brain tractography registration for simultaneous joint alignment of tractography across all subjects (fibers from different subjects colored differently) is conducted (a.2). Spectral clustering is performed to generate the fiber clustering atlas (each cluster has a unique color as shown in a.3). Three example individual fiber clusters are displayed. Sub-figure (b) displays the corticospinal tract (CST) in the atlas, as curated by an expert neuroanatomist. Several example clusters belonging to the CST are displayed (b.1), where each cluster represents a specific subdivision of the whole CST (b.2). Sub-figure (c) demonstrates whole brain white matter tract parcellation for a new subject. The new subject’s tractography (c.1) is first registered to the atlas tractography (colored in pink in c.2). Fiber clustering of the aligned tractography is then conducted according to the fiber clustering atlas for whole brain white matter tract parcellation (c.3). Sub-figure (d) illustrates the subject-specific anatomical tract identification. Identification of the CST (d.2) in the new subject is conducted by finding the corresponding subject-specific clusters (d.1) to those annotated as the CST in the atlas.

Figure 2:

Overview of the fiber cluster anatomical annotation method, including two initial annotation computation steps, followed by expert judgment. Sub-figure (a) shows the initial cluster annotation using White Matter Query Language (WMQL) that provides anatomical definitions of fiber tracts based on their intersected Freesurfer regions. The corticospinal tract (CST) is selected for illustration. Any fiber clusters that have fibers meeting the WMQL CST definition (a.1) are initially identified to belong to CST. Two example clusters (a.2) are displayed. Sub-figure (b) shows the initial cluster annotation using tract anatomical profile (TAP) that is defined as a set of Freesurfer regions through which a cluster passed. The TAP of the example yellow cluster contains six Freesurfer regions (b.1). TAP-based initial annotation for the cerebellar tract is used for illustration. Any fiber clusters that have TAP containing the cerebral-cortex or cerebral-white-matter Freesurfer regions are initially identified to belong to the cerebellar tract. Two example clusters are displayed (b.2). After the initial cluster annotation using the above two steps, expert judgment is performed for a final cluster annotation, as illustrated in sub-figure (c). For the two potentially CST clusters (c.1 and c.2), the yellow cluster is accepted; however, the green cluster is rejected because most of its fibers do not touch the precentral or postcentral gyri, and a corrected annotation of the corona-radiata-frontal (CR-F) tract is provided. For the two clusters potentially belonging to the cerebellar tract (c.3 and c.4), the yellow cluster is accepted and an annotation of a sub-category, i.e. the cortico-ponto-cerebellar (CPC) tract, is provided; however, the green cluster is rejected because the white matter connections between the cerebellum and the cortex should cross the hemispheres, and thus this cluster is categorized as a false positive tract.

Figure 3:

Quantitative evaluations for whole brain white matter parcellations at different scales (k = 200 to 4000) using the two HCP datasets. At each parcellation scale, WMPG (a), ISPV (b) and TAPC (c) were computed, as described in Section 2.5. WMPG was used to measure if all fiber clusters could be generally detected in the population. ISPV was computed to measure if the number of fibers in the corresponding clusters were similar. TAPC was used to measure if fibers within a cluster commonly passed through the same Freesurfer regions.

Figure 4:

Population-based visualizations for several example fiber tracts. For each tract, individual fiber clusters are displayed in different colors. For AF, SLF II, UF, TF and CPC, a left view is displayed; for IoFF, a superior view is displayed; for CC4, CST and Sup-P, a anterior view is displayed.

Figure 5:

(a) Voxel-based tract heatmaps of three example fiber tracts based on the 100 HCP-atlas subjects. The value of a voxel in the heatmaps represents the number of subjects that have fibers passing through the voxel. The background image is the population mean T1 image from all HCP-100 subjects (Section 2.2). (b) The corresponding tracts from one individual HCP-atlas subject, overlaid on the subject’s T1 image.

Figure 6:

Visualizations of example subject-specific fiber tracts identified using the proposed method (left) and the WMQL method (right). For the CPC and Sup-P tracts, only results from our method are displayed because there were not corresponding definitions in WMQL. Individuals from the multiple datasets under study were selected as follows. For each of the HCP-atlas and the HCP-test datasets, an adult who had a population mean age (29 years) was selected. For the dHCP dataset, the youngest (1 day) and the oldest (27 days) neonates were selected. For the ABIDE-II dataset, the youngest AUT (5.1 years) and healthy control (5.9 years) were selected. For the CNP dataset, four 29-year-old adults (same as the population mean age of the HCP-atlas subjects), respectively, from the ADHD, BP, SZ and healthy groups were selected. For the PPMI dataset, two individuals who had the highest ages in the PD and healthy groups (82 and 80 years, respectively) were selected. For the BTP dataset, tracts from two patients (36 and 66 years old) were selected. The tumor (green) and surrounding edema (gray) are visualized for each patient.