基于典型相关分析与双模态数据融合的抑郁症辅助诊断

为了充分提取抑郁症患者的磁共振影像信息，提高抑郁症的诊断准确率，本研究将功能磁共振图像与结构磁共振图像作为研究对象，提出一种双模态数据融合的抑郁症分类算法。首先构建4种不同尺度的功能脑网络，提取功能磁共振图像的数据特征，然后使用迁移学习处理的三维密集连接卷积神经网络，提取结构磁共振图像的数据特征，接着使用典型相关分析方法融合两种特征，最后使用支持向量机对融合特征进行分类，从而将受试者识别为健康者或抑郁症患者。实验结果表明，本文提出的方法可获得89.56%的分类准确率与95.48%的召回率，与单模态数据分类相比，基于双模态数据的分类方法具有更好的分类性能。此外，典型相关分析法可以有效融合双模态的图像特征。

Aided diagnosis of major depressive disorder based on canonical correlation analysis and bimodal data fusion

Abstract: In order to fully extract the magnetic resonance imaging (MRI) information of patients with major depressive disorder for improving the diagnostic accuracy of the disease, taking functional MRI data and structural MRI data as the research objects, a bimodal data fusion algorithm is proposed for the classification of major depressive disorder. After extracting the functional MRI data features by 4 kinds of functional brain networks with different scales, and obtaining the structural MRI data features by three-dimensional densely connected convolutional neural network processed by transfer learning, canonical correlation analysis method is used to fuse the two kinds of features, and finally support vector machine is utilized to classify the fusion features, thereby identifying the subjects as healthy or depressed. The experimental results show that the proposed method can obtain a classification accuracy of 89.56% and a recall rate of 95.48%. Compared with classification methods based on unimodal data, the classification algorithm based on bimodal data fusion has better classification performance. In addition, canonical correlation analysis is effective to fuse bimodal image features.