Search research articles

关于 JoVE

概览领导团队博客 JoVE 帮助中心

作者

出版流程编辑委员会范围与政策同行评审常见问题投稿

图书馆员

用户评价订阅访问资源图书馆顾问委员会常见问题

研究

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments 存档

教育

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual 教师资源中心教师网站

使用条款与条件

Search research articles

相关实验视频

Updated: Mar 10, 2026

Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities

Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities

Published on: October 27, 2023

适应性多教师知识蒸框架,用于医学图像分析的基础模型.

Dudu Liu¹, Yuan Gao², Ningyi Zhang¹

¹Intelligent Medical Computing Laboratory, Faculty of Applied Sciences, Macao Polytechnic University, Rua de Luís Gonzaga Gomes, 999078, Macao Special Administrative Region of China.

Computerized medical imaging and graphics : the official journal of the Computerized Medical Imaging Society

|March 9, 2026

概括

相关概念视频

您也可能阅读

相关文章

通过共同作者、期刊和引用图与本文相关的文章。

排序

Same author

Characteristics of gut microbiota structure and composition in diabetic patients from the Chaoshan region of China.

Journal of medical microbiology·2026

Same author

Efficacy of Jishi Shuanghua Granules in preventing radiation-induced esophagitis in non-small cell lung cancer patients undergoing concurrent chemoradiotherapy: protocol for a multicenter, randomized, double-blind, placebo-controlled trial.

Frontiers in medicine·2026

Same author

Modified sini powder for the management of postoperative depression in non-small cell lung cancer patients: a multicenter, randomized, double-blind, placebo-controlled trial protocol.

Frontiers in pharmacology·2026

Same author

Characteristics of dermal fibroblasts and their biotherapeutic value in musculoskeletal system diseases.

Frontiers in bioengineering and biotechnology·2026

Same author

Brain and knee joint degeneration following anterior cruciate ligament injury in a mouse model of Alzheimer's disease.

JBMR plus·2026

Same author

Soil Fungal Communities' Characteristics of the <i>Lamiophlomis rotata</i> Root-Zone to Altitude and Their Relationship with Environmental Factors.

Microorganisms·2026

医疗成像的基础模型 (FMs) 现在可以使用MultiMedDistill.ll将其蒸成一个单一的轻量级学生模型. 这种框架可以在边缘设备上有效地部署,克服计算限制.

科学领域:

人工智能的人工智能
医疗成像医学成像
机器学习机器学习

背景情况:

医学成像中的基础模型 (FMs) 面临着在资源有限的边缘设备上部署的挑战,原因是高计算需求和大参数大小.
重新培训FM需要稀缺的数据和大量的资源,而知识转移往往导致信息丢失.

研究的目的:

引入MultiMedDistill,这是一个适应性的多教师蒸框架,用于将异质FM集成到轻量级学生模型中.
为了实现高效的模型协调,能力迁移,知识保存和医疗成像FM的实际边缘部署.

主要方法:

开发了MultiMedDistill,一个适应性的多教师蒸框架.
将多个异质的FM集成到一个轻量级的学生模型中.
实施了一种双层门机制,用于动态教师协调,以及一个返回解码器,用于语义保证.

主要成果:

在6个基准数据集 (超声波,内镜, fundus成像,CT,MRI) 中实现了94.77%的Dice在BUSI和97.06%的Dice在Kvasir-SEG.
与基线相比,表现出25.76%和13.04%的显著改善.
将学生模型压缩到8.8M参数 (18×缩小),具有自适应性门和基于重建的知识保存,贡献了显著的收益.

关键词:

适应性关门机制适应性关门机制深度学习是一种深度学习.基金会模型基金会模型医疗图像分析医学图像分析模型的压缩压缩.多个教师的知识蒸.

相关实验视频

Last Updated: Mar 10, 2026

Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities

Author Spotlight: An Efficient and Robust Software for Automated Fusion of Multiple Preclinical Imaging Modalities

Published on: October 27, 2023

结论:

MultiMedDistill有效地以最小的计算成本转移FM功能.
该框架允许在临床边缘设备上实际部署强大的医学成像模型.
这种方法克服了再培训和跨模型知识转移的局限性,用于医疗保健中的边缘AI.