Predicting detection performance with model observers: Fourier domain or spatial domain?

Predicting detection performance with model observers: Fourier domain or spatial domain?