Blind CT Image Quality Assessment Model Based on CT Image Statistics

Abstract

Image quality assessment is widely used in many image processing tasks, which can help researchers adjust image processing algorithms, design imaging systems, and evaluate image processing systems. Generally, CT image quality assessment can be categorized into task-specific and general image quality evaluation. Task-specific image quality assessment evaluates the performance of the imaging system or the detectability of the tumor. These IQA index, for example, are modulation transfer function (MTF), Signal-to-Noise Ratio (SNR), observer model, etc. General image quality assessment measures the general reconstruction image quality under different reconstruction algorithms. SSIM (Structural Similarity), Mean Squared Error (MSE), etc. are the traditional general image quality assessment indexes widely used in nowadays CT image quality assessment. The drawback of these indexes is the demand for reference images, which is not practical in the real CT system. In this paper, we design a CT image dataset, and by using this dataset, and we propose a blind image quality assessment (BIQA) model based on CT image statistics, which can be employed to measure the algorithms under no reference image situation. Different from other image datasets, we recruited no-converged images of the reconstruction process in designing datasets, which enables our BIQA model to evaluate non-converged images during the iterations. Hence, the BIQA model can be embedded in the reconstruction process to monitor reconstructed image quality during iterations.