Applying Transparent Thresholding to Visualize Data and Highlight Results in suma

1933 

Abstract Submission 

Peter Lauren¹, Daniel Glen², Paul Taylor¹, Richard Reynolds¹

¹National Institute of Mental Health, Bethesda, MD, ²NIMH, Bethesda, MD

Peter Lauren  
National Institute of Mental Health
Bethesda, MD

Daniel Glen  
NIMH
Bethesda, MD

Paul Taylor  
National Institute of Mental Health
Bethesda, MD

Richard Reynolds  
National Institute of Mental Health
Bethesda, MD

AFNI's [1] 3D visualization tool, suma [2], can display neuroimaging data on the anatomical surfaces of an individual's brain or a reference template. To date, thresholding of this overlaid data at some value T has only been performed in a standard all-or-nothing (AON) or "opaque" manner: only information with magnitude >T would be displayed, and none below it. However, there may be times when the user would want to show information below T, and get a sense of how far the data is below T, from a single view without having to interactively adjust T. For example, Allen et al [3] showed that graphical displays can become less informative as data complexity increases; but one can make them more informative by transparently displaying, rather than hiding, subthreshold areas. AON thresholding also hid the degree of uncertainty throughout much of the brain. Showing all areas but making subthreshold ones increasingly transparent provided substantially more information, such as near-significant effects and overall quality of the data. Taylor et al [4] further showed that using AON thresholding in neuroimaging had several negative consequences in reproducibility and understanding and leads to selection bias, giving undue influence to arbitrary filter values. Highlighting, rather than hiding, produced more complete results as information contained in sub-threshold locations is often extremely useful in understanding the supra-threshold ones such as by providing context.

Those previous studies only implemented volume-based transparent thresholding, not on surface viewers. suma can now apply transparent thresholding to surface datasets. The opacity for the thresholded value is assigned the maximum value of 1, but can also be between 0 and 1. This allows the sub-T overlay colors to fade, rather than disappear, as their values fall below T.

An "A" checkbox has been added to the suma surface control menu to turn on transparent threshold (or "Alpha" thresholding, since it is the alpha channel of the displayed coloration that is affected). Consider an underlying surface that has a color U, to which a user adds a color overlay dataset X and threshold dataset Y, which is thresholded at value T. If "A" is unchecked (standard thresholding), then the blended overlay-surface value Z at any location is: X, if |Y|>=T; or U, if |Y|<T. If "A" is checked (transparent thresholding), then the blended value Z at any location is: X, if |Y|>=T (as before); or a*X+(1-a)*U, if |Y|<T, where a=(|Y|/T)**2 is the alpha value, which is always between 0 and 1. In the latter scenario the displayed value in a subthreshold region is a blend of the overlay and underlay, where the blending depends on the square of the ratio of the local value to the global threshold T.

Figure 1 shows two applications of transparent thresholding in suma (each row is a separate overlaid dataset). Within each row, when AON thresholding is applied (Panel C),only suprathreshold overlay values are shown. There is no visible overlay information on the rest of the surface. When transparent thresholding is applied (Panel D), the suprathreshold overlay coloration remains the same, but subthreshold values are also visible globally and interpretable: near-threshold values are nearly opaque, while much lower ones are nearly transparent. This enables all overlay areas to be understood, while the suprathreshold ones remain focally highlighted.

A new option, added to SUMA allows fading, rather than hiding, of subthreshold data on surface overlays. This allows significant (at or above threshold) data to be enhanced while not removing its context which may otherwise be hidden by an arbitrary threshold that does not account for uncertainty or processing errors. For example, this is particularly useful for enhancing understanding when reporting statistics results. Future work will allow multiple overlays to be blended based upon their strength relative to their respective thresholds.

Methods Development ¹

Other Methods ²

Computational Neuroscience

Computed Tomography (CT)

Data analysis

Data Organization

FUNCTIONAL MRI

Modeling

Open-Source Code