摘要
人工智能辅助的医学图像识别诊疗系统应用面非常广泛,然而在当前医疗环境和社会背景下难以收集到足够多的数据来训练模型,利用数据增强技术对已有样本进行处理可以显著缓解训练数据缺乏的问题。本文就近年来出现的较为常用的图像增强技术进行简要概述,根据样本处理数量,将已有的数据增强技术分为单样本数据增强和多样本数据增强两大类。其中单样本数据增强又根据处理水平分为像素水平处理和整体水平几何处理;而多样本数据增强则主要介绍了以合成少数过采样算法(SMOTE)和生成对抗网络(GAN)等为代表的以整个样本集为操作对象,通过调整采样比例缓解类不平衡现象的算法。此外,本文还对每种方法的优缺点进行分析,总结了近年来上述算法的实现和改进,以期为相关从业者提供新思路。
Artificial intelligence-assisted medical image recognition diagnosis and treatment system has been used extensively.However,it is difficult to collect enough data to train the model under the current medical environment and social background.By data augmentation technology on processing existed samples,shrotage of data can be greatly eased.This article briefly summarized the common image augmentation techniques used in recent years.According to the number of samples processed,the existing data augmentation techniques are divided into two categories:single-sample data augmentation and multi-sample data augmentation.The former is further divided into pixel-level processing and overall horizontal geometric processing according to the different processing levels;while the latter mainly introduces the object of operation in the entire sample set represented by synthetic minority oversampling technique(SMOTE)algorithm and generative adversarial networks(GAN),and solves the imbalance by adjusting the sampling ratio.Additionally,this article also analyzed the advantages and disadvantages of each method,and summarized the performance and improvement of the above algorithms in recent years to provide new ideas for related practitioners.
作者
王天任
李伊宁
王弘熠
康健
赵爽
柳岸
WANG Tian-ren;LI Yi-ning;WANG Hong-yi;KANG Jian;ZHAO Shuang;LIU An(Department of Dermatology,the Third Xiangya Hospital of Central South University,Hu′nan Province,Changsha410013,China;Department of Otorhinolaryngology,Head and Neck Surgery,the Third Xiangya Hospital of Central South University,Hu′nan Province,Changsha410013,China)
出处
《中国当代医药》
CAS
2021年第3期34-37,44,共5页
China Modern Medicine
关键词
人工智能
数据增强
合成少数过采样算法
生成对抗网络
Artificial intelligence
Data augmentation
Synthetic minority oversampling technique
Generative adversarial networks
