相关实验视频

Updated: Jun 20, 2026

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography

Published on: November 30, 2022

KD-SqueezeNet:一种高效的深度学习策略,用于新生儿肺部疾病的多任务诊断

Jie Li^1,2, Renyi Pan³, Fa Tian⁴

¹The Department of Radiology, Second Affiliated Hospital and Yuying Childrens Hospital of Wenzhou Medical University, Xueyuan West Road, Wenzhou, 325027, China.

Pediatric radiology

|February 20, 2026

概括

这项研究介绍了KD-SqueezeNet,这是一种轻量级的深度学习模型,用于从胸部X射线分类新生儿肺部疾病,实现高精度和效率,以改善诊断.

科学领域:

医疗成像医学成像
人工智能的人工智能
新生儿医疗保健新生儿医疗保健

背景情况:

深度学习模型在医学图像分类方面展示了专家级别的性能.
从胸部X射线图中准确分类新生儿肺部疾病对于及时干预至关重要.

研究的目的:

开发KD-SqueezeNet,一种轻量级和可解释的深度学习模型,用于分类新生儿肺部疾病.
为了提高新生儿呼吸状况评估的诊断准确性和效率.

主要方法:

对2,089名新生儿的胸部X射线进行了回顾性分析.
胸部X射线成像被分为五组:支气管肺功能障碍,肺炎,肺胸,呼吸困扰综合征和正常.
该KD-SqueezeNet模型使用8:1:1数据分割进行训练和验证,使用准确性,精度,回忆,F1分数和AUROC来评估性能.

主要成果:

在分类新生儿肺部疾病方面,KD-SqueezeNet的表现优于EfficientNet和Vision Transformer等其他模型.
对于二进制分类 (健康/疾病),KD-SqueezeNet实现了93%的准确性,精度,回忆和F1评分,AUROC为0.97.
在四个类别的分类中,该模型获得了86%的准确性,精度,回忆和F1分数,高类别特定的AUROC从0.94到0.97.

结论:

KD-SqueezeNet为诊断新生儿肺部疾病提供了一个计算效率高,准确的解决方案.
该模型的可解释性和性能使其适合快速临床部署,帮助决策和改善患者的结果.
在大规模查中,KD-SqueezeNet表现出显著的临床价值,节省时间和资源,同时支持临床医生.

关键词:

深度学习是一种深度学习.诊断诊断诊断诊断诊断新生儿肺部疾病新生儿肺部疾病放射学放射学放射学放射学胸部胸部胸部胸部

相关概念视频

Respiratory Syncytial Virus Disease

Respiratory Syncytial Virus Disease

Human respiratory syncytial virus (RSV) is a widespread pathogen that primarily targets infants and young children but also poses a serious health risk to elderly and immunocompromised individuals. Belonging to the Pneumoviridae family, RSV is a negative-sense, single-stranded RNA virus within the Pneumovirus genus. Its global health burden is significant, with millions of cases annually resulting in hospitalizations and mortality, particularly in resource-limited settings. Although most...