The development and validation of a privacy-preserving model based on federated learning for diagnosing severe pediatric pneumonia.

Abstract

Background: There is a challenge of in diagnostic testing of pneumonia in children, especially severe pneumonia. Thus, developing an auxiliary diagnostic model to help identify severe pneumonia in pediatric patients at an early stage would be highly valuable to address the issues. To overcome the issue of privacy protection, we applied a privacy-preserving machine learning framework to build a multicenter diagnostic model based on federated learning technology.

Methods: Based on Arya, a novel privacy computing platform developed by Hangzhou Healink Technology Corporation, several privacy-preserving federated learning models were developed using datasets from one, two, or four medical centers. A total of 5,091 records were included in this multicenter retrospective study, with 2,484 pediatric patients with severe pneumonia and 2,607 with common pneumonia. Among the records, 80% were used in model training for the diagnosis of severe pneumonia, with 11 common indicators, including white blood cell count (WBC), high-sensitivity C-reactive protein (hs-CRP), hemoglobin (Hb), platelet count (PLT), lymphocyte percentage (L%), monocyte percentage (M%), neutrophil percentage (N%), prothrombin time (PT), alanine aminotransferase (ALT), aspartate aminotransferase (AST), and lactic dehydrogenase (LDH), while the other 20% records were used for model efficacy evaluation. During the process, the original data were stored in the individual hospitals without transmission.

Results: Based on privacy-preserving federated learning technology, the developed models provided reliable diagnostic efficacy for severe pneumonia. Among these models, the four-center model achieved the highest diagnostic efficacy (95.10% sensitivity, 82.70% specificity, and 85.80% accuracy). Although the two-center models achieved a relatively low diagnostic efficacy, they still surpassed the diagnostic efficacy of the single-center model (88.10% sensitivity, 74.60% specificity, and 81.00% accuracy).

Conclusions: Privacy-preserving federated learning technology can facilitate the performance of multicenter studies and was used to develop a high-performance diagnostic model for severe pneumonia in pediatric patients, which can benefit doctors and patients as an auxiliary diagnostic tool.