Yanhua Zheng, Ruilin Ren, Teng Zuo, Xuan Chen, Hanxuan Li, Cheng Xie, Meiling Weng, Chunxiao He, Min Xu, Lili Wang, Nainong Li, Xiaofan Li
{"title":"Prediction of early-phase cytomegalovirus pneumonia in post-stem cell transplantation using a deep learning model.","authors":"Yanhua Zheng, Ruilin Ren, Teng Zuo, Xuan Chen, Hanxuan Li, Cheng Xie, Meiling Weng, Chunxiao He, Min Xu, Lili Wang, Nainong Li, Xiaofan Li","doi":"10.3233/THC-240597","DOIUrl":null,"url":null,"abstract":"<p><strong>Background: </strong>Diagnostic challenges exist for CMV pneumonia in post-hematopoietic stem cell transplantation (post-HSCT) patients, despite early-phase radiographic changes.</p><p><strong>Objective: </strong>The study aims to employ a deep learning model distinguishing CMV pneumonia from COVID-19 pneumonia, community-acquired pneumonia, and normal lungs post-HSCT.</p><p><strong>Methods: </strong>Initially, 6 neural network models were pre-trained with COVID-19 pneumonia, community-acquired pneumonia, and normal lung CT images from Kaggle's COVID multiclass dataset (Dataset A), then Dataset A was combined with the CMV pneumonia images from our center, forming Dataset B. We use a few-shot transfer learning strategy to fine-tune the pre-trained models and evaluate model performance in Dataset B.</p><p><strong>Results: </strong>34 cases of CMV pneumonia were found between January 2018 and December 2022 post-HSCT. Dataset A contained 1681 images of each subgroup from Kaggle. Combined with Dataset A, Dataset B was initially formed by 98 images of CMV pneumonia and normal lung. The optimal model (Xception) achieved an accuracy of 0.9034. Precision, recall, and F1-score all reached 0.9091, with an AUC of 0.9668 in the test set of Dataset B.</p><p><strong>Conclusions: </strong>This framework demonstrates the deep learning model's ability to distinguish rare pneumonia types utilizing a small volume of CT images, facilitating early detection of CMV pneumonia post-HSCT.</p>","PeriodicalId":48978,"journal":{"name":"Technology and Health Care","volume":null,"pages":null},"PeriodicalIF":1.4000,"publicationDate":"2024-07-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Technology and Health Care","FirstCategoryId":"5","ListUrlMain":"https://doi.org/10.3233/THC-240597","RegionNum":4,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q4","JCRName":"ENGINEERING, BIOMEDICAL","Score":null,"Total":0}
引用次数: 0
Abstract
Background: Diagnostic challenges exist for CMV pneumonia in post-hematopoietic stem cell transplantation (post-HSCT) patients, despite early-phase radiographic changes.
Objective: The study aims to employ a deep learning model distinguishing CMV pneumonia from COVID-19 pneumonia, community-acquired pneumonia, and normal lungs post-HSCT.
Methods: Initially, 6 neural network models were pre-trained with COVID-19 pneumonia, community-acquired pneumonia, and normal lung CT images from Kaggle's COVID multiclass dataset (Dataset A), then Dataset A was combined with the CMV pneumonia images from our center, forming Dataset B. We use a few-shot transfer learning strategy to fine-tune the pre-trained models and evaluate model performance in Dataset B.
Results: 34 cases of CMV pneumonia were found between January 2018 and December 2022 post-HSCT. Dataset A contained 1681 images of each subgroup from Kaggle. Combined with Dataset A, Dataset B was initially formed by 98 images of CMV pneumonia and normal lung. The optimal model (Xception) achieved an accuracy of 0.9034. Precision, recall, and F1-score all reached 0.9091, with an AUC of 0.9668 in the test set of Dataset B.
Conclusions: This framework demonstrates the deep learning model's ability to distinguish rare pneumonia types utilizing a small volume of CT images, facilitating early detection of CMV pneumonia post-HSCT.
期刊介绍:
Technology and Health Care is intended to serve as a forum for the presentation of original articles and technical notes, observing rigorous scientific standards. Furthermore, upon invitation, reviews, tutorials, discussion papers and minisymposia are featured. The main focus of THC is related to the overlapping areas of engineering and medicine. The following types of contributions are considered:
1.Original articles: New concepts, procedures and devices associated with the use of technology in medical research and clinical practice are presented to a readership with a widespread background in engineering and/or medicine. In particular, the clinical benefit deriving from the application of engineering methods and devices in clinical medicine should be demonstrated. Typically, full length original contributions have a length of 4000 words, thereby taking duly into account figures and tables.
2.Technical Notes and Short Communications: Technical Notes relate to novel technical developments with relevance for clinical medicine. In Short Communications, clinical applications are shortly described. 3.Both Technical Notes and Short Communications typically have a length of 1500 words.
Reviews and Tutorials (upon invitation only): Tutorial and educational articles for persons with a primarily medical background on principles of engineering with particular significance for biomedical applications and vice versa are presented. The Editorial Board is responsible for the selection of topics.
4.Minisymposia (upon invitation only): Under the leadership of a Special Editor, controversial or important issues relating to health care are highlighted and discussed by various authors.
5.Letters to the Editors: Discussions or short statements (not indexed).