AI-driven prediction of severe respiratory sequelae in COVID-19 patients

人工智能驱动的COVID-19患者严重呼吸系统后遗症预测

阅读:1

Abstract

BACKGROUND: In coronavirus disease 2019 (COVID-19) patients, respiratory sequelae are common complications that significantly impact health outcomes. Hence, early identification of patients at risk is essential for improving prognosis and care. MATERIALS AND METHODS: We enrolled 516 COVID-19 patients and applied K-means algorithms to cluster them into subtypes based on clinical characteristics and risk profiles. ResNet-50 was employed to analyze and extract features from chest X-rays, accurately identifying COVID-19-related lesions. The extracted imaging data were integrated with clinical data to develop a predictive model aimed at stratifying post-COVID-19 patients by risk and identifying those likely to develop severe respiratory sequelae. RESULTS: We identified two distinct COVID-19 subtypes, one of which was associated with severe respiratory sequelae. The convolutional neural networks (CNNs) accurately detected COVID-19-related lesions on chest X-rays. The predictive model showed excellent subtype discriminative ability, achieving an area under the curve (AUC) of 0.949 and 0.958 in the training and validation cohorts, respectively. CONCLUSIONS: Our AI-driven predictive model demonstrates strong potential for the early identification of respiratory sequelae in COVID-19 patients. By applying the K-means algorithm to cluster patients based on clinical characteristics, in combination with feature extraction from chest X-rays using the ResNet-50 deep learning model, we accurately stratified patients by their risk of severe respiratory outcomes. However, to evaluate its performance in clinical settings, further validation using larger, independent datasets is essential to confirm the model's reliability and generalizability across diverse populations.

特别声明

1、本页面内容包含部分的内容是基于公开信息的合理引用;引用内容仅为补充信息,不代表本站立场。

2、若认为本页面引用内容涉及侵权,请及时与本站联系,我们将第一时间处理。

3、其他媒体/个人如需使用本页面原创内容,需注明“来源:[生知库]”并获得授权;使用引用内容的,需自行联系原作者获得许可。

4、投稿及合作请联系:info@biocloudy.com。