Subjective Experiments And Perception Models Of PET/CT Image Quality Assessment

PET/CT(Positron Emission Tomography /Computed Tomography, PET/CT) medical image quality assessment has been a research hotspot in the field of imaging processing. Medical image quality assessment includes two major catalogs: subjective assessment and objective assessment.Subjective assessment, which is based on human visual psychological experiments, obtains high reliable mean-opinion score(MOS) of digital images directly. Objective assessment predicts digital image quality with a metric or an algorithm modeling. IQA metrics or algorithms are evaluated by the results of subjective assessment image database.The research of image quality assessment is highly dependent on subjective assessment results. However, no subjective assessment database is presently available for medical image, which poses huge challenge for the evaluation of medical IQA algorithms. In this paper, the first PET/CT medical image subjective assessment database based on double-stimulus impairment scale method(DSIS) is proposed. The database includes 288 test pictures(12 original pictures, 6 types of distortion for each original pictures, 4 levels for each type of distortion).Performances of thirteen commonly used IQA algorithms are compared on the database. Moreover, effects of different image distortions on IQA algorithms are analyzed. Experimental results show that the Feature Similarity(FSIM) model outperforms other IQA methods, including PSNR, the most commonly used algorithms in the medical field.Experimental results indicates that traditional evaluation index cannot express the algorithm performances objectively. A system of evaluation perception model quality based on PET/CT medical image is proposed. The system comprehensively evaluates the predicting result by veracity, monotony and consistency, which objectively reflects the performances of the perception model.