基于低秩学习的阿尔茨海默病诊断方法

Alzheimer′s disease diagnosis method based on low-rank learning

阿尔茨海默病是老年痴呆症中最普遍的一种,目前大量国内外老年人的生活受其困扰。然而,目前阿尔茨海默病在我国仍然处于就诊率低、误诊率高、治疗率低的局面,因此亟需探索阿尔茨海默病的早期诊断方案,以及时发现潜在病患并进行干预治疗。当前,计算机辅助诊断十分流行,能帮助医生快速、高效地进行疾病诊断。因此,文中提出基于低秩学习的阿尔茨海默病诊断方法(LRL)。该方法利用MRI图像数据提取多模板特征并进行融合;然后针对MRI数据样本数量少、特征维度高的特点,采用低秩学习方法进行特征选择,得到最具代表性的特征子集;最后将选择后的特征输入到支持向量机(SVM)分类器,执行三分类和四分类任务。实验结果表明,提出的LRL模型优于其他几种经典的特征选择方法。在两个主要评价指标准确率ACC和F1上,LRL模型在三分类实验中分别达到了74.94%和75.80%,在四分类实验中分别达到了63.88%和59.99%。

Alzheimer's disease is the most common one of senile dementias,which currently affects a significant number of elderly individuals both in China and around the world.However,Alzheimer's disease is still characterized by low diagnosis rate,high rate of misdiagnosis,and limited treatment efficacy in China.Therefore,there is an urgent need to explore early diagnostic solutions for Alzheimer's disease to detect potential patients promptly and provide intervention and treatment.At present,computer-aided diagnosis(CAD)is very popular,enabling healthcare professionals to efficiently and rapidly diagnose diseases.Hence,an Alzheimer's disease diagnostic method based on low-rank learning(LRL)is proposed.In this method,MRI image data is utilized to extract multiple template features and fused them.To address the challenge of the limited samples and the high feature dimension in MRI data,the LRL approach is employed for feature selection to obtain the most representative feature subset.Subsequently,the selected features are input into a support vector machine(SVM)classifier for both three-class and four-class classification tasks.Experimental results demonstrate that the proposed LRL model outperforms the other classical feature selection methods.On both primary evaluation index accuracy(ACC)and F1 score,the LRL model achieves 74.94%and 75.80%in the three-class classification task,and 63.88%and 59.99%in the four-class classification task,respectively.