基于深度学习的CT脑影像分类方法用于阿尔茨海默病的初步筛查

摘要

目的 本研究旨在探讨卷积神经网络(Convolutional Neural Network,CNN)深度学习在脑CT影像分类中的应用,达到提高影像分类智能化程度的目的,为临床筛查阿尔茨海默病(Alzheimer Disease,AD)提供便利.方法 收集2014~2016年3个类别的脑CT影像资料,其中包含AD、器质性病变(如肿瘤、脑出血等)和正常老年化的受试者的数据.由于本组CT脑图像高度方向(z轴,层厚5 mm)单位长度相对水平方向大的特点,本研究将CT二维轴位CNN图像和三维分割组块进行融合运算分类后对照已有的诊断.结果 AD、器质性病变和正常老年化的分类准确率分别为84.2%、73.9%和88.9%,平均为82.3%.结论 本研究为初筛AD提供了新的方法.%Objective The study aims to discuss the application of deep leaning based on the convolutional neural network (CNN) in the CT imaging classification, so as to improve the intelligent image classification for clinical screening of Alzheimer disease (AD). Methods Three categories of brain CT image data, including the data from AD patients, organic lesion patients (eg. tumor, cerebral hemorrhage) and normal aging patients were collected. For the reason that the relative horizontal direction in CT brain image was high (z axis, seam thickness 5 mm), we fused the two dimensional and three dimensional CNN data in this study, and the results were compared with the diagnostic results. Results The accuracy rates of diagnosis for AD patients, organic lesion patients and normal aging patients were 84.2%, 73.9% and 88.9% respectively, with mean rate of 82.3%. Conclusion Our results supply a new method for preliminary screen of AD.