查询结果:   金祝新,秦飞巍,方美娥.深度迁移学习辅助的阿尔兹海默氏症早期诊断[J].计算机应用与软件,2019,36(5):171 - 177.
中文标题
深度迁移学习辅助的阿尔兹海默氏症早期诊断
发表栏目
图像处理与应用
摘要点击数
667
英文标题
DEEP TRANSFER LEARNING-ASSISTED EARLY DIAGNOSIS OF ALZHEIMER’S DISEASE
作 者
金祝新 秦飞巍 方美娥 Jin Zhuxin Qin Feiwei Fang Meie
作者单位
杭州电子科技大学计算机学院 浙江 杭州 310018 广州大学计算机科学与教育软件学院 广东 广州 510000    
英文单位
School of Computer and Technology, Hangzhou Dianzi University, Hangzhou 310018, Zhejiang, China School of Computer Science and Educational Software, GuangZhou University, Guangzhou 510000, Guangdong, China    
关键词
MRI医学图像 图像处理 深度学习 迁移学习
Keywords
MRI medical image Image processing Deep learning Transfer learning
基金项目
国家自然科学基金项目(61502129);浙江省自然科学基金项目(LY17F020025,LQ16F020004);广东省高校省级平台创强项目(2017KTSCX143)
作者资料
金祝新,硕士生,主研领域:人工智能,医学图像处理。秦飞巍,副教授。方美娥,教授。 。
文章摘要
目前越来越多的老人正在遭受着阿尔茨海默氏症AD(Alzheimer’s Disease)带来的痛苦。临床研究显示,轻度认知障碍MCI(mild cognitive impairment)转化为阿尔兹海默氏症的概率很高,但是若能在MCI阶段对其进行药物治疗是可以康复的,因此提高根据核磁共振图像MRI医学图像诊断的准确率很有必要有。由于医学领域的特点,构建带标注信号的一定规模的数据集非常困难,导致现有机器学习/深度学习方法难以应用于医学影像分析以至于深度学习的结果并不理想。采用数据增广方式对原来数据集的规模进行一定数量的扩充。然后采用一种针对MRI识别的端到端的深度神经网络分类器(MCINet)。结合迁移学习方式对MCINet模型进行有效训练而不至于过拟合。实验结果表明,该方法在较少带标记训练样本的情形下,也获得较高的准确率。
Abstract
At present, more and more elderly people are suffering from Alzheimer’s Disease(AD). Clinical studies have shown that mild cognitive impairment(MCI) has a high probability of transforming into AD, but if it can be treated with medication at the stage of MCI, it can be recovered. Therefore, it is necessary to improve the diagnostic accuracy of medical images based on MRI. Because of the characteristics of medical field, it is very difficult to construct a certain scale data set with annotated signals, which makes the existing machine learning/deep learning difficult to apply to medical image analysis, so that the results of deep learning are not ideal. We used data augmentation to expand the size of the original data set. Then we used an end-to-end deep neural network classifier(MCINet) for MRI recognition. And we combined transfer learning to train the MCINet model effectively without over-fitting. The experimental results show that the method has a high accuracy in the case of fewer labeled training samples.
下载PDF全文