利用机器学习方法对神经肌肉罕见病DMD进行分类预测

Classification Prediction of Duchenne Muscular Dystrophy with a Machine Learning Method

为早期诊断和检测神经肌肉罕见病——杜兴氏肌营养不良(DMD),设计了一组分类预测试验.首先,利用小波变换对DMD患者组和健康对照组的磁共振图像(MRI)进行小波分解;其次,从所得的分解图像中提取出若干纹理特征参数并进行降维处理;最后,再基于这些纹理特征参数,利用支持向量机算法(SVM)对试验图像进行分类预测.试验结果显示,若选择适当的小波分解尺度、分类器核函数和相关参数组合,则MRI图像的分类灵敏度、特异度和准确率分别可达96.9%,97.3%和97.1%.该处理方法有望为临床提供客观有效的辅助诊断手段,可作为DMD罕见病无创检测的尝试探索.

To diagnose and test the orphan neuromuscular disease———Duchenne muscular dystrophy （DMD）in early stage,an experimental plan was designed.First,the magnetic resonance images （MRI）for DMD patients and healthy persons were decomposed into wavelets by using wavelet transform technique. Then,the dimension reduction was conducted with respect to the texture feature parameters extracted from the decomposed images.In the end,on the basis of the texture feature parameters,the classification and prediction of these images were carried out by using support vector machines （SVM）.The results show that if the suitable combination of wavelet function,decomposed scale,kernel function and related parameters were selected,the classification sensitivity,specificity and overall correct classification rate of the MRI images can reach 96.9%,97.3% and 97.1% respectively.This plan might provide an objective and effective auxiliary method for clinical diagnoses.