Research On Low-dose CT Image Quality Assessment And Its Performance

Computer tomography has the advantage of high-resolution and imaging fastly. But the high-dose X-ray may result in risk of disease cancer. Low-dose CT can reduce the X-ray radiation hazard for clients, but the image quality will be decreased accordingly. On the premise of guarantee the quality of CT imaging, scholars conducted deeply-research in various aspects about how to lower the radiation dose effectively. Research on low-dose CT medical image quality assessment method can test the researchers on the condition of using low dose X-ray images quality of imaging.The disadvantage of Subjective quality evaluation method is time-consuming and poor real-time performance. Traditional objective quality evaluation method of Mean Square Error and Peak signal-to-noise ratio is making simply pixel error caculation to the distortion image and the reference image, but not considering local correlation between neighborhood pixels in the image, and with a deviation between evaluation result and subjective feeling. Structure Similarity is proposed to comparing two structural differences for the quality of the image, the method is simple and effective, but the performance is not good for serious distortion and cross distortion images. Based on Human Visual System algorithm to simulate low characteristics of the Human eye, the accuracy is improved, the disadvantage is that the complex modeling, large amount of calculation.This paper deeply analyzes the traditional image quality assessment method and the structural similarity method, and obtained a common disadvantage that without considering the local correlation of image pixels, so in many cases, the evaluation result is not consistent with subjective feeling. Since the image of adjacent pixels are not isolated, there is an association between them, and Markov random field is often used to analyze the correlation between physical phenomena in time or space, and can be a effective simulation of the image pixels partial correlation. This paper uses MRF for modeling and simulating the local correlation of adjacent pixels.Espen Volden combined the Mutual Information with Markov random vector field, put forward the Clique-vector model, and achieved good results with the index evaluation of two images redundancy achieved good results. However, the index has not test as image quality evaluation index of performance on comprehensive database. In this paper, choose the LIVE image library of Gaussian Blur, Fast Fading and White Noise image to validate and analyze the performance of various image indexes. Then evaluate the different low dose phantom images by the metrics. Experimental results show that the mutual information based on MRF gets the best performance for Gaussian Blur, Fast Fading and White Noise image, when evaluating low dose CT image has obvious advantages.