基于增强AlexNet的深度学习的阿尔茨海默病的早期诊断

发布时间：2018-03-05 23:32

本文选题：阿尔茨海默病　切入点：轻度认知障碍　出处：《计算机科学》2017年S1期 　论文类型：期刊论文

【摘要】：在中国,越来越多的老人正在忍受着阿尔茨海默病(Alzheimer's Disease,AD)的痛苦。阿尔茨海默病俗称老年痴呆症,临床上表现为失忆、丧失语言能力等。目前,中国的阿尔茨海默病患者人数已居世界第一。因此,早期诊断阿尔茨海默病变得十分急迫。研究表明,轻度认知障碍(Mild Cognitive Impairment,MCI)转化为阿尔茨海默病的概率很高,它是介于阿尔茨海默病和正常(Healthy Control,HC)之间的一种状态。随着大数据时代的来临,机器学习方法在疾病诊断方面受到热捧。所以,研究提出使用深度学习方法实现对阿尔茨海默病、轻度认知障碍和健康人群的诊断。数据库来自网络公开数据库ADNI。原始的核磁共振图像(Magnetic Resonance Imaging,MRI)的预处理得到首都医科大学附属北京天坛医院的指导。使用卷积神经网络对降维后的实验数据进行训练。因为目前的网络模型不是针对医学图像的,所以实验的重点在于改进现有网络模型,使之达到良好的诊断效果。改进的网络模型是在图像分类方面十分出色的AlexNet网络模型。实验根据阿尔茨海默病的特点提出改进原始模型的4种算法,采用并行方式计算,使用曙光W780-G20服务器,利用8块NVIDIA Tesla K80进行GPU计算,获得4个分类器:AD vs.HC,AD vs.MCI,MCI vs.HC和AD vs.MCI vs.HC。数据集中图像总数量超过7万张,耗时不超过30分钟。最终,通过绘制ROC曲线,计算敏感度、特异度、精确度,对测试结果进行评估,得到了较好的测试结果。
[Abstract]:In China, more and more old people are suffering from Alzheimer's disease (Alzheimer's disease), commonly known as Alzheimer's disease, which has clinical manifestations of amnesia, loss of language, and so on. China has the highest number of people with Alzheimer's disease in the world. Therefore, early diagnosis of Alzheimer's disease is urgent. Studies have shown that mild cognitive impairment has a high probability of converting to Alzheimer's disease. It is a state between Alzheimer's disease and normal healthy control HC.With the advent of big data's time, machine learning methods have become popular in disease diagnosis. Diagnosis of mild cognitive impairment and healthy population. The database is from the online open database ADNI. the preprocessing of the original magnetic Resonance imaging is guided by Beijing Temple of Heaven Hospital affiliated to Capital Medical University. The use of convolutional nerve. The network trains the experimental data after dimensionality reduction because the current network model is not for medical images. So the focus of the experiment is to improve the existing network model, The improved network model is a very good AlexNet network model for image classification. According to the characteristics of Alzheimer's disease, four algorithms for improving the original model are proposed in this experiment. Using the dawning W780-G20 server and using 8 NVIDIA Tesla K80 to carry out GPU calculation, four classifiers:: AD vs.HCN AD vs.MCIvs.MCIvs.HC.The total number of images in the dataset is more than 70,000, and the total number of images in the dataset is no more than 30 minutes. Finally, by drawing the ROC curve, Calculated sensitivity, specificity, accuracy, evaluation of the test results, obtained a better test results.
【作者单位】：中国科学院计算机网络信息中心;中国科学院大学;
【基金】：国家高技术研究发展计划(863计划)(2014AA01A302) 国家自然科学基金重点项目(91530324) 北京市自然科学基金重点项目(4161004) 中国科学院计算机网络信息中心主任基金(CNIC_ZR_201502)资助
【分类号】：R749.16;TP18

【相似文献】