Medical image quality assessment is critical for comparing and optimizing medical imaging system. In the frame of task-based approach, numerical observers proposed for the objective medical image quality assessment have several limitations: e.g. for the multi-signal localization, most need a priori knowledge of signal parameters; multi-slice MOs narrow the task down to the detection of one symmetrical signal at the image centre. In this thesis, we propose novel numerical observers: the CJO for the detection of one parametric signal with random amplitude, orientation and size on single-slice; the PCJO and the msPCJO for the detection-localization of multiple parametric signals with random amplitude, orientation, size and location on single-slice and multi-slice, respectively. Two observer studies were designed and implemented for their validation on MR images with MS lesions, and the JAFROC FOM results indicate that they are promising for predicting radiologists' task performance, ultimately for evaluating medical imaging systems.
