We estimated human observer linear templates underlying the detection of a realistic, spherical mass signal with mammographic backgrounds. Five trained naïve observers participated in two-alternative forced-choice (2-AFC) detection experiments with the signal superimposed on synthetic, clustered lumpy backgrounds (CLBs) in one condition and on nonstationary real mammographic backgrounds in another. Human observer linear templates were estimated using a genetic algorithm. A variety of common model observer templates were computed, and their shapes and associated performances were compared with those of the human observer. The estimated linear templates are not significantly different for stationary CLBs and real mammographic backgrounds. The estimated performance of the linear template compared with that of the human observers is within 5% in terms of percent correct (Pc) for the 2-AFC task. Channelized Hotelling models can fit human performance, but the templates differ considerably from the human linear template. Due to different local statistics, detection efficiency is significantly higher on nonstationary real backgrounds than on globally stationary synthetic CLBs. This finding emphasizes that nonstationary backgrounds need to be described by their local statistics.
© 2007 Optical Society of AmericaPDF Article