IEEE Paper Practice

\documentclass[conference]{IEEEtran}
\IEEEoverridecommandlockouts
\usepackage{cite}
\usepackage{amsmath,amssymb,amsfonts}
\usepackage{algorithmic}
\usepackage{algorithm}
\usepackage{graphicx}
\usepackage{textcomp}
\usepackage{xcolor}
\usepackage{url}
\usepackage{hyperref}
\hypersetup{
    colorlinks=true,
    linkcolor=black,
    citecolor=blue
}

\def\BibTeX{{\rm B\kern-.05em{\sc i\kern-.025em b}\kern-.08em
    T\kern-.1667em\lower.7ex\hbox{E}\kern-.125emX}}

\begin{document}

\title{A Meta Survey on Federated Learning for Medical Image Analysis}

\author{
\IEEEauthorblockN{ Kazi Namira Meyheg Sanam}
\IEEEauthorblockA{
  ID: C231450\\
  Department of Computer Science and Engineering\\
  \textit{International Islamic University Chittagong}\\
  Chittagong 4318, Bangladesh\\
  c231450@ugrad.iiuc.ac.bd\\
  \vspace{5pt} 
}
}

\maketitle

%-----------------------------------------------------------------------
\begin{abstract}
In recent years, machine learning and deep learning techniques have become widely applied in medical image analysis. These techniques help doctors analyze medical images for tasks such as disease classification, image segmentation, and object detection. Nevertheless, the volume of data required to build reliable medical imaging models is considerable and access to such data is frequently constrained by regulations designed to protect patient privacy, such as HIPAA and GDPR. As a result, federated learning (FL) has attracted growing interest in the medical imaging community as a privacy-aware training strategy. This meta-survey conducts an extensive examination of recent research on the application of federated learning to medical image processing. We review studies from the past five years covering imaging of organs including the liver, stomach, colon, prostate, breast, and lungs. The survey discusses techniques for medical image classification, investigates the impact of non-IID data distributions on federated models, addresses data diversity and privacy challenges, and outlines the key benefits and limitations of applying federated learning to the medical imaging domain.
\end{abstract}

\begin{IEEEkeywords}
medical image processing, federated learning, deep learning, CNN
\end{IEEEkeywords}

%-----------------------------------------------------------------------
\section{Introduction}

Medical image processing relies on a variety of imaging modalities—including CT scans \cite{ref1}, Magnetic Resonance Imaging \cite{ref2}, and X-rays \cite{ref3}—to acquire visual data that allow clinicians to detect, classify, and monitor a wide spectrum of diseases, from breast cancer \cite{ref4} and lung cancer \cite{ref5} to brain tumors \cite{ref6} and COVID-19. Radiomics, an emerging discipline within medical imaging, focuses on the extraction of quantitative features from clinical images to capture subtle information about tissue shape, texture, and signal intensity \cite{ref7}. The radiomics pipeline encompasses image acquisition, region-of-interest segmentation, feature extraction, and predictive model construction.

Artificial intelligence—particularly machine learning (ML) \cite{ref8} and deep learning (DL)—has proven indispensable for the timely and accurate identification and classification of diseases observable on medical images. Computer vision methods based on ML and DL have become central to this domain. However, traditional ML approaches are based on manually engineered features, which can introduce errors and reduce performance. DL models, in contrast, require substantial volumes of labeled training data to achieve satisfactory accuracy \cite{ref9}.

Collecting large medical image datasets is difficult because patient data must remain private. Regulations such as the General Data Protection Regulation (GDPR) and the Health Insurance Portability and Accountability Act (HIPAA) restrict how hospitals share medical information \cite{ref10}. These restrictions make it difficult to aggregate data from multiple institutions into a unified global training set. To address this fundamental limitation, McMahan et al. introduced federated learning (FL) \cite{ref11, ref12, ref13}—a distributed training paradigm in which data held in multiple hospitals are used to train local neural network models, while a central server coordinates the overall training process. In each communication round, participating institutions update their local models and transmit only model parameters—not raw patient data—to the central server \cite{ref14, ref15}. The server then combines these local updates into a new global model \cite{ref16}, which is redistributed to all participating clients for the next training round \cite{ref17, ref18}.

Given its inherent privacy-preserving properties, FL has attracted considerable attention in the medical research community. During the past decade, a large body of work has been published, examining various datasets (including images, electronic health records, and patient records), tackling various challenges (such as improving model accuracy, managing multi-modal data and handling non-IID data distributions), and exploring a wide range of learning methodologies \cite{ref19, ref20, ref21, ref22, ref23, ref24, ref25, ref26, ref27, ref28, ref29}.

Despite a search query restricted to ``federated AND learning AND images,'' the Scopus database returns more than 1000 results, reflecting the rapid growth of this field and the increasing difficulty of maintaining a comprehensive view of all medical sub-disciplines involved.

\subsection*{A. Contributions}

This paper presents a meta-survey—a survey of existing surveys—on federated learning applied to medical imaging. Its objective is to offer a unified and consolidated perspective on the current body of literature. To our knowledge, this work represents the first systematic meta-analysis dedicated to federated learning for medical image analysis.

By examining and comparing prior survey articles, this work aims to expose longitudinal trends and developments within the field. It also highlights the limitations and gaps present in existing reviews, identifies topics requiring further investigation, and offers a structured summary of current state-of-the-art methodologies. In addition, it compares the effectiveness of federated learning against conventional centralized ML and DL baselines, explores model aggregation strategies, collaborative training paradigms, and personalized medicine approaches.

This survey is intended as a practical reference for both academic researchers and industry practitioners. It is designed to support the development of robust, privacy-preserving models for medical image recognition, classification, and segmentation.

The remainder of the paper is structured as follows: Section II describes the methodology employed. Section III presents the analysis and findings corresponding to the defined research questions. Section IV concludes the study.

%-----------------------------------------------------------------------
\section{Methodology}

\subsection{Search Strategy}

This study adopts the Preferred Reporting Items for Systematic Reviews (PRISMA) framework \cite{ref13} as its guiding methodology. PRISMA is a widely recognized standard for presenting and reporting evidence in systematic reviews and it has been adopted by numerous journals and professional organizations in the medical imaging field. The literature search was conducted on Google Scholar, covering publications from January 1, 2018 to April 1, 2023. The search queries employed were ``federated learning,'' ``medical images,'' and ``survey'' as primary keywords. Since our scope encompasses general medical FL research—of which medical imaging FL is a subset—health-related keywords were deliberately excluded from the query.

\textbf{Inclusion Criteria:} Survey papers and systematic literature reviews that focus specifically on federated learning applied to the detection and classification of the medical image. Studies in which the primary data modality is medical images (e.g., MRI, CT, X-ray, ultrasound). Research that details the implementation of federated learning algorithms or techniques, including differential privacy or secure aggregation.

\textbf{Exclusion Criteria:} Reviews that do not specifically address federated learning for medical image analysis. Studies in which the primary data type is non-medical images. Reviews that do not employ federated learning methods for medical images. Research focused on federated learning applied to non-image data types, such as electronic health records. Studies that do not compare federated learning with other baselines for ML or DL.

\textbf{Research Questions:} Based on the existing literature, the following research questions were defined: RQ1. What federated learning applications exist for classifying medical images and what types of medical imaging datasets are typically used? RQ2. What ML and DL approaches are most commonly used for training local models in FL systems? RQ3. What advanced techniques are used to train the global model? RQ4. How do data heterogeneity and non-IID distributions affect the performance of FL models for medical image analysis? RQ5. What security and privacy challenges are associated with FL in medical imaging? RQ6. How does federated learning compare in effectiveness to conventional centralized ML/DL approaches?

\subsection{Study Selection}

Initial searches of the database returned 186 results, with three additional citations identified from other sources, yielding a total of 189 publications. After removing duplicates, 116 unique articles remained. These were screened by title and abstract, eliminating 94 records, leaving 22 for full-text review. After full-text screening, 11 other studies were excluded, resulting in 11 final studies (including one review from an additional source) with sufficient data for both qualitative and quantitative analysis. The PRISMA-based selection flow is illustrated in Figure~\ref{fig:prisma}.

\begin{figure}[htbp]
\centering
 \includegraphics[width=\columnwidth]{prisma.png}
\caption{PRISMA flow diagram showing the literature selection process. The initial search returned 189 records, which were screened and filtered according to the inclusion and exclusion criteria, resulting in 11 studies used for the final analysis.}
\label{fig:prisma}
\end{figure}

%-----------------------------------------------------------------------
\section{Analysis and Results}

The study \cite{ref19} provided a comprehensive evaluation of deep learning-based medical image analysis using FL; however, its scope was limited by subscription-based access limitations, which prevented the retrieval of several relevant articles and may have introduced selection bias. Additionally, that review focused primarily on deep learning methods for brain cancer detection, potentially overlooking the broader landscape of FL applications in medical imaging. Similarly, the work in \cite{ref20} focuses mainly on the detection of COVID-19, which—while significant—may not capture the wider utility and challenges of FL in different medical imaging tasks. Although \cite{ref29} provides an in-depth treatment of FL, classifying techniques according to client-server dynamics and examining available platforms and datasets, a unified taxonomy that spans all prior reviews has yet to be established.

Nevertheless, federated learning has demonstrated substantial effectiveness in the classification of medical images. Its application has been validated for COVID-19 detection using chest X-rays and CT scans \cite{ref20, ref24, ref21, ref23, ref28}, and it has also shown promise in detecting cancers of the lung, breast, prostate, and brain. FL leverages distributed datasets to enable early-stage disease detection, with reported applications spanning skin cancer diagnosis, pancreatic differentiation, cardiac diseases, brain tumor detection \cite{ref22, ref28, ref27, ref23, ref25, ref26, ref29}, and screening for diabetic retinopathy.

\begin{table*}[t]
\centering
\caption{An Accessible SLR on FL-Driven Medical Image Analysis}
\label{tab:slr}
\renewcommand{\arraystretch}{1.4}
\begin{tabular}{|p{4.2cm}|p{0.8cm}|p{2.5cm}|p{3.2cm}|p{2.5cm}|p{2.8cm}|}
\hline
\textbf{Article Title} & \textbf{Ref.} & \textbf{Journal/Conference} & \textbf{Dataset} & \textbf{Disease Detected} & \textbf{ML/DL Method} \\
\hline
A Systematic Review on Federated Learning in Medical Image Analysis & \cite{ref19} & IEEE Access & Lung X-ray, CT, Skin (HAM10000), Retina, MRI (BraTS) & Diabetic Retinopathy, Skin Cancer, Brain Tumor, COVID-19 & CNN, VGG16, ResNet-50/18, AlexNet, MLP, VGG19 \\
\hline
A Comprehensive Review of Federated Learning for COVID-19 Detection & \cite{ref20} & Int. J. Intelligent Systems & CXR and/or CT images & COVID-19 detection & CNN, Transfer Learning, Deep Learning, ANN \\
\hline
Federated Learning for Medical Image Analysis with Deep Neural Networks & \cite{ref21} & Diagnostics & BreakHis, BraTS, COVID-19 CXR, Pneumonia, TCGA, DR dataset, HAM10000 & COVID-19, Brain Tumour, Skin Cancer, Lung Cancer, DR, Pneumonia & Deep Learning, CNN, ResNet-18, Transformers, VGG \\
\hline
A Comprehensive Analysis of Deep and Federated-Learning for Brain Tumor Diagnosis & \cite{ref22} & J. Personalized Medicine & BraTS, multiple brain tumor datasets & Brain tumor detection and classification & CNN, Deep Learning \\
\hline
Federated Learning for Medical Applications: A Taxonomy & \cite{ref23} & arXiv preprint & COVID-19 CT/X-ray, MRI BraTS, TCGA, GLOBOCAN & COVID-19, Breast/Lung/Prostate/Brain/Liver Cancer & CNN, Multi-modal, ML, DNN \\
\hline
Applications and Challenges of FL in Big Data Era (COVID-19) & \cite{ref24} & Big Data \& Cognitive Computing & Private CT, X-ray, Chest X-ray & COVID-19 detection, classification, segmentation & Deep CNN, ResNet-34/50, DenseNet-201, ANN, 3D UNet \\
\hline
Reviewing Federated Machine Learning for Disease Prediction & \cite{ref25} & Sensors & BraTS, ISIC 2018, FGADR, private datasets & Brain tumors, breast/skin/lung/prostate cancer, CVD, diabetes & CNN, Deep Learning, Transformers \\
\hline
Medical Image Analysis Using FL Frameworks: Technical Review & \cite{ref26} & IEEE/Conference & BraTS, MNIST, CIFAR-10 & Brain Tumor Analysis & Bayesian FL ConvNet, ResNet, Transfer Learning \\
\hline
A Review of Medical FL in Oncology and Cancer Research & \cite{ref27} & BrainLes 2021 Workshop & BraTS, HAM10000, WBCD, TCGA, ISIC 2018 & Brain MRI, Skin/Breast/Prostate/Lung/Thyroid Cancer & Deep Learning, CNN, Transfer Learning \\
\hline
A Comprehensive Survey on FL for Healthcare Informatics & \cite{ref28} & Comp. Intelligence \& Neuroscience & MRI, BraTS, X-Ray, Ultrasound & MRI reconstruction, FMI classification, COVID-19, Skin disease & Deep Learning, CNN \\
\hline
Federated Learning for Medical Image Analysis: A Survey & \cite{ref29} & arXiv preprint & MRI, BraTS, ABIDE, HAM10000, BreakHis, COVID-19 CXR & Prostate/Breast cancer, COVID-19, Brain tumor, Cardiac, Skin lesions & GCN, CNN, VGG, ResNet, UNet, Transformers, GAN \\
\hline
\end{tabular}
\end{table*}

The meta-analysis indicates that federated learning is widely employed in medical imaging for classification and detection tasks. Among the most frequently referenced datasets is BraTS, used extensively for brain tumor research. For COVID-19-related tasks, COVID-19 CT and X-ray datasets were predominantly used. Other commonly cited datasets include HAM10000 for skin lesion analysis \cite{ref19, ref21, ref27, ref29}, retinal imaging datasets for diabetic retinopathy, and private CT and X-ray collections for COVID-19 studies. The BreakHis dataset was used for breast cancer histopathology \cite{ref21, ref29}, while ISIC 2018 served as a benchmark for skin cancer classification and the Cancer Genome Atlas (TCGA) was utilized for multi-cancer studies. X-ray datasets, which are essential for thoracic pathology, appeared consistently across studies addressing COVID-19 and pneumonia \cite{ref22, ref23, ref25, ref26, ref27, ref28, ref29}.

Local model training within federated learning environments is predominantly based on DL and ML approaches. CNNs appear across the majority of reviewed studies and remain foundational for tasks involving detection, classification, and segmentation of biomedical images. Collectively, these ML/DL architectures have demonstrated strong performance as local models in COVID-19 detection, brain MRI analysis, skin cancer classification, and breast cancer detection \cite{ref19, ref20, ref21, ref22, ref23, ref24, ref25, ref26, ref27, ref28, ref29}. Regarding global model aggregation, Federated Averaging (FedAvg) is the most widely adopted strategy reported in the literature \cite{ref19, ref20, ref23, ref25, ref28, ref29}. Secure aggregation has also been explored as a complementary technique for enhancing privacy. One study \cite{ref19} incorporated differential privacy for COVID-19 detection from chest X-rays and examined the integration of FL with IoT and blockchain technologies. Bayesian-based federated CNNs were applied specifically for MRI-based analysis in another work.

Data heterogeneity represents a persistent source of performance degradation in FL systems, arising from differences in image acquisition hardware, scanning protocols, and institutional quality standards. The use of heterogeneous modalities—including CT, MRI, X-ray, and ultrasound—in combination further complicates model generalization due to the divergent feature distributions between imaging types. Non-IID data can introduce statistical biases when certain clients exhibit disproportionate representation of specific classes or features, ultimately degrading the global model's predictive accuracy \cite{ref28}. Although FedAvg attempts to mitigate non-IID effects, it frequently struggles in highly heterogeneous settings. As an alternative, the ``SplitAVG'' method was proposed \cite{ref19} to more effectively address data heterogeneity. While sharing only model parameters—rather than raw data—protects patient confidentiality and enables cooperative learning \cite{ref19, ref20, ref21}, this approach remains vulnerable to gradient inversion and reverse engineering attacks, wherein adversarial clients manipulate shared gradients to infer private information \cite{ref24, ref27}. Countermeasures such as gradient encryption, noise injection, and partial weight sharing have been proposed \cite{ref24}. Blockchain-based approaches \cite{ref26, ref27} and homomorphic encryption offer additional layers of protection for data integrity and privacy in FL deployments \cite{ref24}.

When compared to traditional centralized ML and DL methods, federated learning has shown promise in medical image analysis, though its effectiveness varies with the dataset and clinical application \cite{ref19, ref25}. While centralized ML approaches have occasionally surpassed FL in terms of classification accuracy, FL has consistently excelled in tasks such as brain tumor segmentation and has demonstrated particularly strong performance in prostate and thyroid cancer scenarios \cite{ref25}.

\subsection{Limitations and Future Work}

Federated learning in medical image processing faces several significant challenges. Patient data sensitivity demands that privacy and security be maintained throughout the model training lifecycle \cite{ref20}. Managing the variability inherent in medical imaging datasets—encompassing differences in modality, resolution, and contrast—adds further complexity \cite{ref21}. Designing FL models for specific diseases, such as COVID-19 or various cancer types, requires careful consideration of disease-specific image characteristics and patterns. Non-IID data distributions across client nodes make it difficult to train globally representative models \cite{ref29}. Communication bottlenecks—especially bandwidth limitations when transmitting large model updates—can impair FL system performance \cite{ref25}. Security vulnerabilities also threaten the integrity of communication between server and client nodes.

Future research directions in this area should prioritize the development of more robust privacy-preserving mechanisms tailored to sensitive medical data. Addressing dataset heterogeneity across modalities and resolutions is critical. Specialized architectures designed for disease-specific imaging characteristics are needed, as are effective methods for managing non-IID data distributions. Efficient communication protocols—potentially leveraging model compression or gradient sparsification—should be explored for large-scale FL deployments. Security mechanisms for parameter exchange must be strengthened to prevent data breaches. In summary, future efforts should seek to optimize FL algorithms for medical imaging with an emphasis on security, communication efficiency, statistical fairness, and patient privacy protection.

Overall, the reviewed studies demonstrate that federated learning has strong potential for collaborative medical image analysis, particularly when data sharing between institutions is restricted.
%-----------------------------------------------------------------------
\section{Conclusion}

Machine learning and deep learning have become important tools for analyzing medical images. However, training these models is challenging because medical data is sensitive and strict privacy regulations limit data sharing. Federated learning provides a decentralized training paradigm that enables the construction of ML and DL models using data distributed across multiple external sources without centralizing raw information. This study evaluated federated learning for medical image detection and classification across multiple organ-specific modalities, including breast, lung, and liver imaging. The survey examined a range of classification techniques, highlighting the advantages, limitations, and influence of non-IID data on federated learning performance. Data heterogeneity, privacy concerns, and the comparative performance of federated versus conventional centralized learning methods were all addressed. With a focus on model aggregation, collaborative learning, personalized medicine, and the potential for FL to enhance medical image analysis while safeguarding patient privacy, the findings offer meaningful guidance for future research in this rapidly evolving field.

%-----------------------------------------------------------------------
\section*{Acknowledgment}

This work was conducted as part of the academic requirements of the Department of Computer Science and Engineering, International Islamic University Chittagong, Bangladesh.

%-----------------------------------------------------------------------
\begin{thebibliography}{00}

\bibitem{ref1}
A. Qayyum, K. Ahmad, M. A. Ahsan, A. Al-Fuqaha, and J. Qadir, ``Collaborative federated learning for healthcare: Multi-modal covid-19 diagnosis at the edge,'' \textit{IEEE Open Journal of the Computer Society}, vol. 3, pp. 172--184, 2022.

\bibitem{ref2}
M. Islam, M. T. Reza, M. Kaosar, and M. Z. Parvez, ``Effectiveness of federated learning and CNN ensemble architectures for identifying brain tumors using MRI images,'' \textit{Neural Processing Letters}, pp. 1--31, 2022.

\bibitem{ref3}
E. Khan, M. Z. U. Rehman, F. Ahmed, F. A. Alfouzan, N. M. Alzahrani, and J. Ahmad, ``Chest X-ray classification for the detection of COVID-19 using deep learning techniques,'' \textit{Sensors}, vol. 22, no. 3, p. 1211, 2022.

\bibitem{ref4}
H. Lee, Y. J. Chai, H. Joo, K. Lee, J. Y. Hwang, S.-M. Kim, K. Kim, I.-C. Nam, J. Y. Choi, H. W. Yu \textit{et al.}, ``Federated learning for thyroid ultrasound image analysis to protect personal information: Validation study in a real health care environment,'' \textit{JMIR Medical Informatics}, vol. 9, no. 5, p. e25869, 2021.

\bibitem{ref5}
A. Jim\'{e}nez-S\'{a}nchez, M. Tardy, M. A. G. Ballester, D. Mateus, and G. Piella, ``Memory-aware curriculum federated learning for breast cancer classification,'' \textit{Computer Methods and Programs in Biomedicine}, vol. 229, p. 107318, 2023.

\bibitem{ref6}
A. Raza, H. Ayub, J. A. Khan, I. Ahmad, A. S. Salama, Y. I. Daradkeh, D. Javeed, A. Ur Rehman, and H. Hamam, ``A hybrid deep learning-based approach for brain tumor classification,'' \textit{Electronics}, vol. 11, no. 7, p. 1146, 2022.

\bibitem{ref7}
A. Conti, A. Duggento, I. Indovina, M. Guerrisi, and N. Toschi, ``Radiomics in breast cancer classification and prediction,'' in \textit{Seminars in Cancer Biology}, vol. 72, Elsevier, 2021, pp. 238--250.

\bibitem{ref8}
C. Janiesch, P. Zschech, and K. Heinrich, ``Machine learning and deep learning,'' \textit{Electronic Markets}, vol. 31, no. 3, pp. 685--695, 2021.

\bibitem{ref9}
K. Suzuki, ``Overview of deep learning in medical imaging,'' \textit{Radiological Physics and Technology}, vol. 10, no. 3, pp. 257--273, 2017.

\bibitem{ref10}
L. Bradford, M. Aboy, and K. Liddell, ``Covid-19 contact tracing apps: a stress test for privacy, the GDPR, and data protection regimes,'' \textit{Journal of Law and the Biosciences}, vol. 7, no. 1, p. lsaa034, 2020.

\bibitem{ref11}
Y. Cheng, Y. Liu, T. Chen, and Q. Yang, ``Federated learning for privacy-preserving AI,'' \textit{Communications of the ACM}, vol. 63, no. 12, pp. 33--36, 2020.

\bibitem{ref12}
K. Bonawitz, H. Eichner, W. Grieskamp, D. Huba, A. Ingerman, V. Ivanov, C. Kiddon, J. Kone\v{c}n\'{y}, S. Mazzocchi, B. McMahan \textit{et al.}, ``Towards federated learning at scale: System design,'' \textit{Proceedings of Machine Learning and Systems}, vol. 1, pp. 374--388, 2019.

\bibitem{ref13}
R. Sarkis-Onofre, F. Catal\'{a}-L\'{o}pez, E. Aromataris, and C. Lockwood, ``How to properly use the PRISMA statement,'' \textit{Systematic Reviews}, vol. 10, no. 1, pp. 1--3, 2021.

\bibitem{ref14}
A. Guzzo, G. Fortino, G. Greco, and M. Maggiolini, ``Data and model aggregation for radiomics applications: Emerging trend and open challenges,'' \textit{Information Fusion}, p. 101923, 2023.

\bibitem{ref15}
P. Qi, D. Chiaro, A. Guzzo, M. Ianni, G. Fortino, and F. Piccialli, ``Model aggregation techniques in federated learning: A comprehensive survey,'' \textit{Future Generation Computer Systems}, 2023.

\bibitem{ref16}
D. Chiaro, E. Prezioso, M. Ianni, and F. Giampaolo, ``FL-enhance: A federated learning framework for balancing non-IID data with augmented and shared compressed samples,'' \textit{Information Fusion}, vol. 98, p. 101836, 2023.

\bibitem{ref17}
M. Ianni and E. Masciari, ``Scout: Security by computing outliers on activity logs,'' \textit{Computers \& Security}, p. 103355, 2023.

\bibitem{ref18}
F. Sun, W. Zang, R. Gravina, G. Fortino, and Y. Li, ``Gait-based identification for elderly users in wearable healthcare systems,'' \textit{Information Fusion}, vol. 53, pp. 134--144, 2020.

\bibitem{ref19}
M. F. Sohan and A. Basalamah, ``A systematic review on federated learning in medical image analysis,'' \textit{IEEE Access}, 2023.

\bibitem{ref20}
S. Naz, K. T. Phan, and Y.-P. P. Chen, ``A comprehensive review of federated learning for COVID-19 detection,'' \textit{International Journal of Intelligent Systems}, vol. 37, no. 3, pp. 2371--2392, 2022.

\bibitem{ref21}
S. Nazir and M. Kaleem, ``Federated learning for medical image analysis with deep neural networks,'' \textit{Diagnostics}, vol. 13, no. 9, p. 1532, 2023.

\bibitem{ref22}
A. Naeem, T. Anees, R. A. Naqvi, and W.-K. Loh, ``A comprehensive analysis of recent deep and federated-learning-based methodologies for brain tumor diagnosis,'' \textit{Journal of Personalized Medicine}, vol. 12, no. 2, p. 275, 2022.

\bibitem{ref23}
A. Rauniyar, D. H. Hagos, D. Jha, J. E. H\r{a}keg\r{a}rd, U. Bagci, D. B. Rawat, and V. Vlassov, ``Federated learning for medical applications: A taxonomy, current trends, challenges, and future research directions,'' \textit{arXiv preprint arXiv:2208.03392}, 2022.

\bibitem{ref24}
A. Majeed, X. Zhang, and S. O. Hwang, ``Applications and challenges of federated learning paradigm in the big data era with special emphasis on COVID-19,'' \textit{Big Data and Cognitive Computing}, vol. 6, no. 4, p. 127, 2022.

\bibitem{ref25}
M. Moshawrab, M. Adda, A. Bouzouane, H. Ibrahim, and A. Raad, ``Reviewing federated machine learning and its use in diseases prediction,'' \textit{Sensors}, vol. 23, no. 4, p. 2112, 2023.

\bibitem{ref26}
V. Kamble and A. Phophalia, ``Medical image analysis using federated learning frameworks: Technical review,'' in \textit{Proc. IEEE 10th Region 10 Humanitarian Technology Conference (R10-HTC)}, IEEE, 2022, pp. 44--48.

\bibitem{ref27}
A. Chowdhury, H. Kassem, N. Padoy, R. Umeton, and A. Karargyris, ``A review of medical federated learning: Applications in oncology and cancer research,'' in \textit{International MICCAI Brainlesion Workshop}, Springer, 2021, pp. 3--24.

\bibitem{ref28}
K. Dasaradharami Reddy, T. R. Gadekallu \textit{et al.}, ``A comprehensive survey on federated learning techniques for healthcare informatics,'' \textit{Computational Intelligence and Neuroscience}, vol. 2023, 2023.

\bibitem{ref29}
H. Guan and M. Liu, ``Federated learning for medical image analysis: A survey,'' \textit{arXiv preprint arXiv:2306.05980}, 2023.

\end{thebibliography}

\end{document}