去噪扩散概率模型在人工智能医疗器械影像数据增广中的应用
Application of DDPM in artificial intelligence image data augmentation of medical device
郝鹏飞 1李庆雨 1柴蕊 1陈曦 2宋庆华 1韩乃水 1张克1
作者信息
- 1. 山东省医疗器械和药品包装检验研究院医用电器质量评价中心 济南 250101
- 2. 道普云(山东)智能科技有限公司研发部 济南 250101
- 折叠
摘要
医疗器械影像数据增广是一种通过生成新的数据样本来扩展现有数据集的方法,对提高人工智能(AI)医疗器械相关模型性能和临床应用效果具有重要意义.传统的数据增广方法通常受限于生成样本质量、真实感和多样性.去噪扩散概率模型(DDPM)是一种基于噪声扩散过程的生成模型,其主要思想是通过将目标分布的采样过程建模为从噪声分布中逐步去噪的过程,从而生成具有高质量的样本.综述DDPM基本原理和工作机制,分析该方法在AI医疗器械数据增广中的应用场景,探讨其优势、挑战和未来发展方向,为AI医疗器械数据增广领域提供参考.
Abstract
Medical device imaging data augmentation is a method of expanding existing datasets by generating new data samples,which is of great significance for improving the performance of artificial intelligence(AI)medical device-related models and clinical application effects.However,traditional data augmentation methods are usually limited by the quality,realism,and diversity of generated samples.Denoising diffusion probabilistic model(DDPM)is a generative model based on the noise diffusion process,and its main idea is to generate samples with high quality by modelling the sampling process of the target distribution as a process of progressive denoising from the noise distribution.The basic principles and working mechanisms of DDPM were reviewed,the application scenarios of this method in AI medical device data augmentation were analyzed,and its advantages,challenges,and future development directions were explored to provide a reference for the field of AI medical device data augmentation.
关键词
去噪扩散概率模型(DDPM)/医疗器械/人工智能(AI)/数据增广Key words
Denoising diffusion probabilistic model(DDPM)/Medical device/Artificial intelligence,Data augmentation引用本文复制引用
基金项目
国家重点研发计划(2020YFC2007105)
出版年
2024
国家科技期刊平台
NSTL
万方数据