Non-reference Evaluation of Hyperspectral X-CT Images Based on CdTe Photon Counting Detector

With the wide application of CT imaging technology in the medical field, CT reconstruction images can visually show the internal structure of the testing object. Different from natural images, the main reasons affecting the quality of CT reconstruction images are noise and artifacts. In addition, due to the lack of original reference images, there is currently a lack of a general standard for evaluating the quality of CT reconstruction images. The CT reconstruction algorithms include filtering back-projection method (FBP), iterative method ART, SART, and iterative algorithm ARTTV and SARTTV with total variation (TV) minimization optimization. The evaluation indexes are Tenengrad gradient function, gray variance (SMD) function, energy gradient function, point sharpness arithmetic function (EVA) algorithm function and the combined evaluation of noise and blur which are generally used in no-reference perceptual quality assessment. The results show that in the case of complete projection data, for CdTe hyperspectral CT, FBP algorithm has the shortest reconstruction time and the best reconstruction effect, and TV minimization optimization can indeed reduce the blur and noise of the image.