Modelling of Subjective Radiological Assessments with Objective Image Quality Measures of Brain and Body CT Images

In this work we determine how well the common objective image quality measures (Mean Squared Error (MSE), local MSE, Signal-to-Noise Ratio (SNR), Structural Similarity Index (SSIM), Visual Signal-to-Noise Ratio (VSNR) and Visual Information Fidelity (VIF)) predict subjective radiologists’ assessments for brain and body computed tomography (CT) images.A subjective experiment was designed where radiologists were asked to rate the quality of compressed medical images in a setting similar to clinical. We propose a modified Receiver Operating Characteristic (ROC) analysis method for comparison of the image quality measures where the “ground truth” is considered to be given by subjective scores. The best performance was achieved by the SSIM index and VIF for brain and body CT images. The worst results were observed for VSNR.We have utilized a logistic curve model which can be used to predict the subjective assessments with an objective criteria. This is a practical tool that can be used to determine the quality of medical images.