基于ICA的X射线医学图像目标提取

X射线医学成像能观察到患者体内病变组织，对医学诊断有重要参考价值。针对传统医学X射线图像噪声强、层次感差和器官组织重叠的问题，提出利用多能谱X射线成像结合独立成分分析(independent component analysis, ICA)进行图像去噪和目标提取。首先ICA结合稀疏编码收缩法对图像降噪预处理以保证目标提取精度;然后根据图像中各目标组成特性，分离图像中每个像素对应的目标厚度矩阵;最后ICA以盲分离理论获得收敛矩阵重建出目标对象。在ICA算法中，借助于主观评价标准，发现当收敛次数大于40时目标分离成功;当幅值尺度在[25, 45]区间内，目标图像对比度高且失真较小。同时，通过观测实验得到的三维峰值信噪比图表明：ICA算法中收敛次数和幅值对图像质量有较大影响，当重建图像的对比度和边缘信息均达到较好效果时，收敛次数与幅值为85和35。

Object Separation from Medical X-Ray Images Based on ICA

X-ray medical image can examine diseased tissue of patients and has important reference value for medical diagnosis. With the problems that traditional X-rayimages have noise，poor level sense and blocked aliasing organs, this paper proposes a method for the introduction of multi-spectrum X-ray imaging and independent component analysis (ICA) algorithm to separate the target object. Firstly image de-noising preprocessing ensures the accuracy of target extraction based on independent component analysis and sparse code shrinkage. Then according to themain proportion of organ in the images, aliasing thickness matrix of each pixel was isolated. Finally independent component analysis obtains convergence matrix to reconstruct the target object with blind separation theory. In the ICA algorithm, it found that when the number is more than 40, the target objects separate successfully with the aid of subjective evaluation standard. And when theamplitudes of the scale are in the [25, 45] interval, the target images havehigh contrast and less distortion. The three-dimensional figure of Peak signal to noise ratio (PSNR) shows that the. different convergence times and amplitudes have a greater influence on image quality. The contrast and edge informationof experimental images achieve better effects with the convergence times 85 andamplitudes 35 in the ICA algorithm.