Using Gaze-tracking Data and Mixture Distribution
Analysis to Support a Holistic Model for the
Detection of Cancers on Mammograms
1
Harold L. Kundel, MD, Calvin F. Nodine, PhD, Elizabeth A. Krupinski, PhD, Claudia Mello-Thoms, MSEE, PhD
Rationale and Objectives. Use data collected independently at three institutions to compare time to first fixate the true
lesion in searching for cancers on mammograms. Examine the fit of the results to a holistic model of visual perception.
Materials and Methods. The time required to first fixate a cancer on a mammogram was extracted from 400 eye-tracking
records collected independently from three institutions. The time was used as an indicator of the initial perception of can-
cer. The distribution of first fixation times was partitioned into two normally distributed components using mixture distri-
bution analysis. The true-positive fraction of each component was calculated.
Results. About 57% of the cancers had a 95% chance of being fixated in the first second of viewing. The remainder took
longer (range, 1.0 to 15.2 seconds). The true-positive fraction was larger for the lesions hit immediately for most of the
readers (TPF = 0.63 vs. 0.52, F = 5.88, P = .02) in 68% (13/19) of the readers.
Conclusions. The initial detection occurs before visual scanning and, therefore, must be the result of a parallel “global”
analysis of the image resulting in an initial holistic, gestalt-like perception. The development of expertise in medical im-
age analysis may consist of a shift in the recognition mechanism from scan-look-detect to look-detect-scan.
Key Words. Mammography; breast cancer; perception; gaze tracking; observer performance; holistic perception
©
AUR, 2008
A previous gaze-tracking study of mammographers, mam-
mography fellows, and radiology residents searching
mammograms for cancer showed that more than half of
the cancers fixated by the observers were visually in-
spected within 1.1 seconds of the onset of viewing (1).
This result was attributed to a global response that syn-
thesizes a complete perception and identifies perturbations
in the image. The gaze is then directed to the perturba-
tions, and local features are analyzed using input from the
high-resolution central vision performing, what we term
“checking fixations.” After a covert task-based decision is
made about the nature of a perturbation (eg, is it a cancer
or not), the eyes are either moved to another location
based on information from the global response or begin a
more general discovery scanning of the image. Discovery
scanning can be cognitively determined. For example, a
scan path can be geometric when the target abnormality is
very small and seemingly randomly positioned like micro-
calcifications on a mammogram or it can be anatomic
when the target is a rib fracture. During discovery scan-
ning, there is continued input from the peripheral retina
and the scan path can be interrupted by checking fixa-
tions. Additionally, perturbations that are considered to be
target locations are revisited resulting in increased dwell
time (2) on locations that will be reported as positive and
Acad Radiol 2008; 15:881– 886
1
Department of Radiology, University of Pennsylvania, 3600 Market Street,
Suite 370, Philadelphia, PA 19444 (H.L.K., C.F.N.); Department of Radiol-
ogy Research, University of Arizona, Tucson, AZ (E.A.K.); Department of
Radiology, University of Pittsburgh, Pittsburgh, PA (C.M.-T.). Received Au-
gust 24, 2007; accepted January 17, 2008. Address correspondence to
C.F.N. e-mail: nodine@oasis.rad.upenn.edu
©
AUR, 2008
doi:10.1016/j.acra.2008.01.023
881