Click on image to view larger version.
Fig. 3. —Graph shows positive predictive value for mammographers
([UNK]) and trainees (
). Positive predictive value is function of time
to decision for final-decision phase and takes into account both true-positive
responses (TP) and false-positive responses for cases with normal
findings (FPn). Positive predictive value is calculated as [TP /
(TP + FPn)]. False-positive responses for abnormal cases were not
included, which is common usage. Positive predictive value performance begins
high and levels off for both mammographers and trainees. Each set of positive
predictive value data are fit by two linear-regression lines. These lines
cross at approximately 25 sec for mammographers and at approximately 40 sec
for trainees. These lines divide performance over time course of viewing into
what Christensen et al. [1]
labeled rapid phase and slow phase. We hypothesize that rapid phase reflects
global discovery of lesions by Gestalt process and that slow phase reflects
detection of lesions by focal search process.
