Evaluation of rigid registration methods for whole head imaging in diffuse optical tomography

Abstract

Functional brain imaging has become an important neuroimaging technique for the study of brain organization and development. Compared to other imaging techniques, diffuse optical tomography (DOT) is a portable and low-cost technique that can be applied to infants and hospitalized patients using an atlas-based light model. For DOT imaging, the accuracy of the forward model has a direct effect on the resulting recovered brain function within a field of view and so the accuracy of the spatially normalized atlas-based forward models must be evaluated. Herein, the accuracy of atlas-based DOT is evaluated on models that are spatially normalized via a number of different rigid registration methods on 24 subjects. A multileveled approach is developed to evaluate the correlation of the geometrical and sensitivity accuracies across the full field of view as well as within specific functional subregions. Results demonstrate that different registration methods are optimal for recovery of different sets of functional brain regions. However, the "nearest point to point" registration method, based on the EEG 19 landmark system, is shown to be the most appropriate registration method for image quality throughout the field of view of the high-density cap that covers the whole of the optically accessible cortex.

Keywords: atlas-based tomography; diffuse optical tomography; functional connectivity brain imaging; registration; sensitivity analyses; whole head imaging.

Fig. 1

Workflow of creating a registered atlas-based mesh.

Fig. 2

Set of different Landmark systems used for registration: (a) basic 4, (b) EEG 19, (c) EEG 40, (d) full head, (e) line, and (f) sphere 19.

Fig. 3

High-density source-and-detector cap for an example head surface.

Fig. 4

An example of geometry error based on three registration methods for an example subject: (a) posterior view, (b) right view, and (c) top view.

Fig. 5

Evaluation of geometrical errors based on 24 subjects. The central (red) lines represent the median, the box plots represent the 25th and 75th percentiles, whereas the whiskers present $\pm 2.7$ standard deviations. Outliers are presented as red crosses.

Fig. 6

Brain functional regions used for geometrical representation. (a) posterior view, (b) right view, and (c) top view.

Fig. 7

An example of gray matter geometry errors based on three registration methods for an example subject: (a) posterior view, (b) right view, and (c) top view.

Fig. 8

An example of sensitivity percentage error of the cortex based on three registration methods for an example subject: (a) posterior view, (b) right view, and (c) top view.

Fig. 9

Evaluation of sensitivity errors of the cortex based on 24 subjects.

Fig. 10

Correlation between geometry error and sensitivity errors based on 24 subjects and registration methods.

Fig. 11

Outline of the EEG 10/20-based head regions within the ROI for geometrical and sensitivity analysis.

Fig. 12

Region 2 variation based on Fig. 11 showing a high correlation and medium strength (slope). (a) Evaluation of geometrical errors in region 2, (b) evaluation of sensitivity errors of the cortex in region 2, and (c) correlation between geometry error and sensitivity errors in region 2.

Fig. 13

Region 6 variation based on Fig. 11 showing a low correlation and high strength (slope). (a) Evaluation of geometrical errors in region 6, (b) evaluation of sensitivity errors of the cortex in region 6, and (c) correlation between geometry error and sensitivity errors in region 6.

Fig. 14

Correlation between geometrical and sensitivity errors in all EEG 10/20-based head regions.

Fig. 15

Strength of geometrical and sensitivity errors in all EEG 10/20-based head regions.