基于自适应非对称卷积网络的精神分裂症MRI图像分类研究

Research on MRI image classification of schizophrenia based on adaptive asymmetric convolutional networks

针对精神分裂症的发病机制尚不清楚,难以精准治疗的问题,首次提出了一种自适应非对称卷积网络用于精神分裂症MRI图像分类研究.首先,基于熵值对三种切片图像进行数据筛选,以获取更为全面、丰富的特征信息.然后,利用自适应非对称卷积灵活地确定特征之间的连接状态,并在不增加额外参数的情况下增强特征表示能力.最后,将高阶特征投影到分类目标空间进行疾病分类.在两个公开数据集上评估了该模型的有效性,分别获得了95.77%和96.99%的平均准确率,与传统的卷积神经网络相比提高了3%~6%的分类准确率.分析表明,提出的方法可以动态地关注局部特征和全局特征,从而挖掘更有效的特征信息,为精神分裂症的识别诊断提供了一种新的思路和方法.

To address the problem that the pathogenesis of schizophrenia is still unclear and difficult to treat precisely,an adaptive asymmetric convolutional network was proposed for the first time for MRI image classification study of schizophrenia.Firstly,the data of three kinds of slice images were screened based on entropy value to obtain more comprehensive and rich feature information.Then,adaptive asymmetric convolution was used to flexibly determine the connection state between features and enhance the feature representation ability without adding additional parameters.Finally,the higher-order features were projected into the classification target space for disease classification.The effectiveness of the proposed model was evaluated on two public datasets,and the average accuracy was 95.77%and 96.99%,respectively.Compared with the traditional convolutional neural network,the classification accuracy of the proposed model was improved by 3%-6%.The analysis showed that the method proposed in this study could dynamically focus on local features and global features,so as to mine more effective feature information and was expected to provide important assistance to physicians in disease diagnosis.