基于神经网络的压缩后医学图像的质量评估

Quality Evaluation of Compressed Medical Image Based on Neural Network

目的研究如何对压缩后的医学图像的质量进行客观的评估。方法针对用单个客观参数(如均方差)对压缩后的医学图像进行质量评估效果很差的问题,本文先应用神经网络进行参数优化,从众多的客观参数中选择出若干参教,然后作为误差反传神经网络(BPNN)的输入,再采用神经网络来拟合医生的主观评估,实现压缩后的医学图像的质量评估。结果实验结果表明,该方法对肺窗和纵膈窗的测试准确率平均达到84.72％和85.49％,而且具有很强的适应性。结论所提出的用神经网络评估压缩后的医学图像总体质量的方法,能实现图像的按质量自动归类和比较客观的质量评估。

Objective To study the method of quality evaluation of compressed medical images. Method Single objective parameter such as MSE, performs badly in classing and evaluating the quality of compressed medical images. Firstly, we select out several parameters from a group of objective parameters including traditional parameters such as MSE, parameters based on HVS model and a new images quality index by using a neural network. Then these parameters are inputted to another back propagation neural network to mimic the doctor' s subjective evaluation and assess the quality of compressed medical images. Result The proposed method achieves much better results than the method only using single objective parameter. The average test accuracy is 84.72% and 85.49% for lung and mediastinum respectively, and the method has very good adaptability as well. Conclusion The proposed method is able to classify the compressed medical images according to their quality automatically and evaluate their quality more objectively.