Current State of Federated Learning in Cardiology: A Narrative Review

سال انتشار: 1402
نوع سند: مقاله کنفرانسی
زبان: انگلیسی
مشاهده: 104

نسخه کامل این مقاله ارائه نشده است و در دسترس نمی باشد

استخراج به نرم افزارهای پژوهشی:

لینک ثابت به این مقاله:

شناسه ملی سند علمی:


تاریخ نمایه سازی: 1 مرداد 1402

چکیده مقاله:

Background and aims: Artificial intelligence (AI) has become increasingly important in themedical industry, including cardiology. Federated learning (FL) is a promising approach for creatingAI models in cardiology that allows access to a diverse set of patient data while simultaneouslyaddressing privacy concerns. This narrative review aims to present a summary of the currentstatus of FL in cardiology.Method: We searched Google Scholar to find papers on the use of federated learning in cardiology.Relevant articles were selected and summarized, showcasing current use cases of FL incardiovascular medicine.Results: Federated learning has been employed in cardiology for several applications, includingbut not limited to arrhythmia classification, cardiac disease prediction, and image segmentation.Multiple studies have demonstrated the efficacy of federated learning in ECG data analysis, includingarrhythmia classification and cardiac disease prediction. For instance, one study examinedthe performance and effectiveness of FL using three benchmark datasets, one of which wasthe PhysioNet ECG dataset. The researchers carried out federated experiments on balanced andimbalanced data between clients. In both cases, they achieved an F۱-score of ۰.۸۰۷. Furthermore,they noted that the performance of FL and centralized learning were not significantly different. Inother studies, FL has been used for the multi-label, multi-class classification of ECG data that isheterogeneously distributed; researchers proposed several FL scenarios for this purpose.In the field of cardio-imaging, several studies have demonstrated the effectiveness of FL for imagesegmentation and diagnosis of heart conditions. For example, one study incorporated differentpriors to the model by leveraging ground truth masks to improve performance for deep learning-based multi-center cardiovascular magnetic resonance diagnosis. They experimented withdiverse data augmentations followed by different convolutional neural network settings to assessmodel robustness. They also studied a federated learning algorithm with equal votes assigned toevery training center. In their study, federated learning achieved comparable results with traditionalcentralized learning.Other studies have investigated the use of FL in cardiology for different tasks like predicting coronaryartery calcification scores, risk stratification of diseases such as ischemic heart disease, andbinary supervised classification model based on EHR data for predicting hospitalizations due tocardiac events. The studies have consistently reported similar performance results for the distributedand centralized methods, further indicating that FL has potential as an alternative approachto analyzing cardiology data.Conclusion: Federated learning is a relatively new approach to developing more robust AI modelsby enabling access to diverse patient data repositories while addressing privacy concerns. Iteliminates the need for centralizing the data and allows multiple parties to share their local datawithout exchanging it. The studies we reviewed demonstrated the effectiveness of FL in variouscardiology applications, such as arrhythmia classification, cardiac disease prediction, image segmentation,and risk stratification of diseases like ischemic heart disease. Notably, current applicationsof FL in cardiology have primarily focused on ECG datasets and cardio-imaging. The results consistently indicate that FL performs similarly to traditional centralized learning, highlighting itspotential as a collaborative approach for analyzing cardiology data on a larger scale. The currentstate of FL in cardiology represents a noteworthy advancement in the field that offers new pathwaysto explore novel solutions for improving patient outcomes.


Hamid Ariannejad

School of Advanced Technologies in Medicine, Iran University of Medical Sciences, Tehran, Iran- Tehran Heart Center, Tehran University of Medical Sciences, Tehran, Iran

Zeynab Barzegar

School of Advanced Technologies in Medicine, Iran University of Medical Sciences, Tehran, Iran

Masoud Aliyali

School of Advanced Technologies in Medicine, Iran University of Medical Sciences, Tehran, Iran

Hamidreza Sadeghsalehi

School of Advanced Technologies in Medicine, Iran University of Medical Sciences, Tehran, Iran

Ashkan Vatankhah

School of Advanced Technologies in Medicine, Iran University of Medical Sciences, Tehran, Iran