Figure 1.
Example input and output data for two subjects from Study II.
The DWIs were randomly chosen by a computer but are from the same gradient for each subject. DWIs and bo are shown at different intensity scales for viewing purposes. The FA color scale ranges from [blue-red] and colors map to FA values [0 1]. Inter-subject registration was not performed and the axial slices are from approximate matching locations. The FA map of subject 2 contains an anomalous bright region in the middle-right mid-brain/hemisphere (patient right is image right). This bright region is not clearly traced to DWI artifacts, but upon comparison to a re-scan of the same individual is associated with flow artifacts rather than pathology.
Table 1.
DTI Data.
Figure 2.
Flowchart of QA pipeline showing major processing steps (blue boxes) and outputs stored or graphed in the QA report (red ovals).
Not shown is conversion of image data to NIfTI files (immediately after ‘Parse Header’), the creation of two brain masks after ‘rotate gradients’, one mask for CAMINO and a second slightly more restrictive mask for statistical analysis, and the implied additional calculation of in ‘Analyze Residuals’.
Figure 3.
The four pages of the QA report.
All data is shown from the same subject. The first page (P.1) parses the header (gray and blue boxes), plots patient movement in rotation and translation, plots RESTORE outliers per DWI (three upper right plots), plots (color plot), shows five ‘best’ and ‘worst’ DWI slices (bottom left), and plots a histogram of
with an automatically determined magnification around the ‘noise’ lobe region as described in Papadakis et. al.. (P.2) shows a model of the 25 segmented regions (right column) and plots regional distributions of MD, FA,
, and BFA (blue boxplots) adjacent to corresponding distributions from the Multi-modal dataset (black boxplots). (P.3) Page three displays mid-axial slice views for MD, FA,
, and BFA (top row, left to right), select power curves for FA (middle column), and full mid coronal, axial, and sagittal slices for the vector colormap (R = right-left, G = anterior-posterior, B = foot-head). (P.4) Page 4 shows the vector directions (while lines, not clearly discernible at figure size) overlaid on the vector colormap for a mid-axial (left column) and mid-coronal (right column) slice. Three different enlargements are shown for improved viewing.
Figure 4.
Comparing three pieces of the QA report for four subjects.
Each piece is selected from a different page of the report. Subject 1 and subject 2 are from Study II and are the same subjects in Figure 1. Subject 3 is from Study I and subject 4 is from Study IX. (A) Colorplot of. Within each colormap, one column represents an entire normalized diffusion weighted volume, and each row represents the same axial slice. High
correspond to poor data with 0.2 being definitively bad. (B) MD distributions for four segmented regions, left cerebellar gray matter (GM), right cerebellar GM, left cerebellar white matter (WM), and right cerebellar WM. Blue boxplots indicate the distribution of each subject’s MD value while the black box-plot is the reference data from the multi-modal study. (C) Power curves for theoretical sample sizes of n = 5 (black), n = 15 (red), and n = 30 (blue). Curves include bias estimates and pertain to voxels in the cerebral WM.
Table 2.
Information presented on each page of the QA report.
Table 3.
Stored Pipeline Outputs.
Table 4.
Rubric Questions.
Figure 5.
Statistical metrics stored by pipeline evaluated using PCA.
Each of the 567 DTI datasets was characterized by a 112 element vector of stored outputs from the pipeline. PCA analysis was performed on the resulting data. (A) DTI dataset locations in the first two dimensions of the PCA analysis. Data is symbolized by study Roman numeral (Table 1). Single arrow points to a data quality outlier from study I; subject 3 in Figure 4 and in (B). A double headed arrow points to a cluster representing an isolated protocol sub-group from study VI. (B) FA maps from similar sagittal slice locations in two subjects from Study I. Subject 4 is the indicated outlier in (A) and subject 5 was selected from the center of the study I cluster seen in (A).
Table 5.
Summary Metrics for Rubric Evaluations.
Table 6.
Rubric Evaluation Using Non-parametric Hypothesis Testing.
Table 7.
Time Evaluation.