基于深度学习的阿尔兹海默症多模态辅助诊断研究

阿尔兹海默症(Alzheimer's Disease,AD)是一种在老年人群中常见的痴呆疾病,由于病程不可逆且无法治愈,常会对病人的生活质量产生极大影响,因此尽早诊断病情并对病程加以干预是唯一有效的手段。由于良好的实验效果,深度学习模型在医学图像领域受到了越来越多研究者的关注,但深度学习方法常需要较大的数据量作为支撑,而医学图像由于设备成本以及病例数量的限制,常存在着数据量不足的问题,因而在某些情况下会出现过拟合的问题。提出一种参数高效的深度学习模型,引入了可分离卷积、全局平均池化、残差结构,使得模型参数量成倍地减少,同时引入多模态数据,增大了输入样本的信息量,以求减少过拟合问题。最后,通过对照试验,验证了该文所提出模型的优越性。

Multi-Modal Data Computer-Aided Diagnosis of AD Based on Deep Learning Model

Alzheimer's Disease(AD)is a common dementia disease in the geriatric population which is irreversible and cannot be cured,and it often affects the life quality of the patients.Early detection and early treatment may be the only effective way for helping the patients of AD.Due to excellent performance,deep learning models have attracted more and more researchers'attention in the field of medical images.However,deep learning methods often require a large amount of data for training model,and because of the limitations of the patient number and the high equipment costs,the data of medical images are always insufficient,and overfitting problems may occur in some cases.To avoid the overfitting problems,this paper attempts to propose a parameter-efficient deep learning model.It introduces separable convolution,global average pooling,and residual structure into the model,and the number of model parameters are reduced by several times.At the same time,with using multimodal data,the information of the input is increased.Finally,the superiority of the model proposed in this paper is verified through controlled experiments.