CoChat
Sign Up
Cited by 114Privacy-Preserving Technologies in DataArtificial Intelligence in Healthcare and EducationDomain Adaptation and Few-Shot Learning

A systematic review on federated learning in medical image analysis

Open access
Sohan, Md Fahimuzzman, Basalamah, Anas·2023·IEEE Access
DOI 10.1109/access.2023.3260027
Open this paper in CoChat — free accountOpen PDF

Abstract

Federated Learning (FL) obtained a lot of attention to the academic and industrial stakeholders from the beginning of its invention. The eye-catching feature of FL is handling data in a decentralized manner which creates a privacy preserving environment in Artificial Intelligence (AI) applications. As we know medical data includes marginal private information of patients which demands excessive data protection from disclosure to unexpected destinations. In this paper, we performed a Systematic Literature Review (SLR) of published research articles on FL based medical image analysis. Firstly, we have collected articles from different databases followed by PRISMA guidelines, then synthesized data from the selected articles, and finally we provided a comprehensive overview on the topic. In order to do that we extracted core information associated with the implementation of FL in medical imaging from the articles. In our findings we briefly presented characteristics of federated data and models, performance achieved by the models and exclusively results comparison with traditional ML models. In addition, we discussed the open issues and challenges of implementing FL and mentioned our recommendations for future direction of this particular research field. We believe this SLR has successfully summarized the state-of-the-art FL methods for medical image analysis using deep learning.

AI snapshot

Written from the paper’s full text.

TL;DR

This PRISMA-based systematic literature review synthesizes 17 studies on federated learning (FL) for medical image analysis, covering applications such as COVID-19, cancer, diabetic retinopathy, pneumonia, autism, and cardiac disease. It finds FL models often achieve high accuracy but consistently underperform comparable traditional ML models in the reviewed comparisons.

Key findings

  • From 161 initially collected articles, 17 met inclusion criteria; COVID-19 was the most common application (6/17), cancer accounted for 5/17 (29.4%), and dataset sizes ranged from 153 to 223,414 images.
  • Eight of 17 FL studies reported ≥90% accuracy; the highest reported FL accuracy was 94.4% (FL+VGG16 with data augmentation, Feki et al.), followed by 93.57% (FL+VGG16). Medium-range examples included 89.82% (VGG-based FL) and 85%; low multi-class examples included AUC 80%, accuracy 70%, and 76%.
  • In every comparison in Table 7, traditional ML models outperformed the corresponding FL models; e.g., 89% FL vs 92% ML, 94.4% FL vs 99.96%/98% ML, and AUC 80% FL vs 86%/87% ML.
  • Only four papers considered non-IID data distribution (with insufficient explanation from [26] and [21]); the remaining 13 did not discuss it. Additional privacy methods were rare: differential privacy in Li et al., Kaissis et al., and Adnan et al., and encryption/blockchain in Połap et al.

Methods

  • Design: PRISMA systematic literature review; search period 1 January 2017 to 30 June 2022; databases included Science Direct, IEEE Xplore, Springer Link, Wiley Online Library, SPIE, ACM, MDPI, Nature Portfolio, Taylor & Francis, and Google Scholar; Boolean “AND”/“OR” searches over titles, abstracts, and keywords.
  • Sample: 161 initial records; after duplicate removal, 138 remained; title/abstract screening removed 96; full-text review of 42 removed 25; final set was 17 articles. Inclusion required medical image datasets, an ML model developed in an FL environment, and FL as the main focus; exclusions included private datasets, non-FL/medical-image focus, hybrid FL, abstracts/short articles/preprints/books, and no clear ML performance measures.
  • Analysis: extracted document metadata, datasets and federated settings, privacy protocols, training algorithms, and performance; organized findings around 12 research questions and a 12-question quality analysis.

Limitations

  • Paper-stated: subscription paywalls prevented retrieval of some full texts despite emailing corresponding authors, with failures for [94] and [95]; preprints and some databases/articles may have been missed; the authors did not re-run the models under their supervision; limited time and resources, plus FL’s recent introduction, made strong tested historical conclusions difficult.
  • Evident from the text: the quality analysis found most reviewed articles failed to meet quality requirements, with the highest score 9/12 ([27]) and next 6/12 ([20], [28]); many lacked non-IID consideration, performance comparison, or clear federated data handling. The paper also states there is no benchmark dataset or benchmark FL model in this field.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/access.2023.3260027

Cites

87 of 88 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

0 of 152 cited sources checked · 0 supported · 0 flagged

  • Findingnone given

    The authors did not find any article that worked particularly on medical images.

    “We did not find any article which has worked particularly on medical images.”
    Quote in the paperSubstantial claim
  • Findingcitation

    Zhang et al. [24] and Yan et al. [29] used X-ray and CT image data for different CNN architectures in FL settings.

    “Similarly, Zhang et al. [24] and Yan et al. [29] used X-ray and CT image data for different Convolutional Neural Network (CNN) architectures in FL settings.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [24] source

      Not checked in the free run.

    • cites [29] source

      Not checked in the free run.

  • Findingcitation

    Hashmani et al. [30] applied FL on dermoscopy images to classify nine different skin diseases.

    “Moreover, Hashmani et al. [30] applied FL on a series of dermoscopy images to classify nine different skin diseases.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [30] source

      Not checked in the free run.

  • Findingown data

    Five out of 17 reviewed articles (29.4%) were formed on cancer detection.

    “Fig. 4 shows 29.4% articles (five out of 17) of this review were formed on cancer detection.”
    Quote in the paperNot a substantial claim
  • Findingcitation

    Linardos et al. [26] considered FL for diagnosing hypertrophic cardiomyopathy, whether subjects are suffering from HCM or normal.

    “Linardos et al. [26] considered FL for Diagnosing Hypertrophic Cardiomyopathy (HCM), whether the subjects are suffering from HMC or normal.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] source

      Not checked in the free run.

  • Classified as probably not a substantial claim. 74
  • May combine two claims. 46
  • The quote only partly states this claim. 5

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    Adnan et al.'s model achieved acceptable performance while decentralized data configuration was applied.

    “They demonstrated that their model achieved acceptable performance while decentralized data configuration applied.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [28] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Feki et al. [18] proposed a collaborative FL for COVID-19 screening from chest X-ray images, cooperating with multiple medical institutions without sharing their data.

    “Feki et al. [18] proposed a collaborative FL for COVID-19 screening from chest X-ray images; they cooperated with multiple medical institutions without sharing their data.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    The review identified eight different types of medical images collected from various datasets used in the research field.

    “Fig. 5 displays eight different types of medical images collected from various datasets used in the research field.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingnone given

    Polap and their team published three research papers [17], [19], [22], all focused on skin cancer detection with FL.

    “Researcher Polap and their team have published three research papers [17], [19], [22], all of them focused on skin cancer detection with the FL environment.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [17] source

      Not checked in the free run.

    • cites [19] source

      Not checked in the free run.

    • cites [22] source

      Not checked in the free run.

  • May combine two claims.

    Findingnone given

    The X-ray datasets considered in the reviewed articles are Cohen JP, TB x-ray, CheXpert, Mendeley data, COVIDx, Chest X-ray (CXR), and COVID 2019 dataset.

    “The X-ray datasets considered in the articles are Cohen JP, TB x-ray, CheXpert, Mendeley data, COVIDx, Chest X-ray (CXR), and COVID 2019 dataset.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The review found FL applications in different research fields including diabetic retinopathy, MRI classification, cancer, pneumonia, and COVID-19 detection.

    “We found the application of FL in different research fields, such as, Diabetic Retinopathy (DR), MRI classification, cancer, pneumonia, COVID-19 detection, and few more.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Lo et al. [16] collected a total of 153 data samples from four different sources.

    “In another article, Lo et al. [16] collected a total of 153 data samples from four different sources.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Adnan et al. [28] used lung tissue images to classify cancer in a privacy-guaranteed ML model.

    “Adnan et al. [28] used tissue image data, more specifically they proposed a privacy guaranteed ML model where lung tissue images were considered to classify cancer.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [28] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

    Findingcitation

    Połap and the groups used the MNIST: HAM10000 dataset in a series of articles [17], [19], [22] with FL environment.

    “Połap and the groups used the dataset in a series of articles [17], [19], [22] with FL environment.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [17] source

      Not checked in the free run.

    • cites [19] source

      Not checked in the free run.

    • cites [22] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Zhang et al. [24] proposed a FL-oriented COVID-19 detection model using chest X-ray and CT images from three datasets: Qatar-Dhaka data, COVID-CT, and Figure 1 dataset.

    “Moreover, Zhang et al. [24] proposed a FL oriented COVID-19 detection model where chest X-ray and CT images were considered from three datasets, Qatar-Dhaka data, COVID-CT, and Figure 1 dataset.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [24] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Zhou et al. [31] used a DR dataset consisting of 3,662 images.

    “In [31], Zhou et al. used a DR dataset consisting of 3,662 images.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [31] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Similar data has been used in [30], which was released under ISIC 2019 and contains 25,331 dermoscopy images.

    “Similar data has been used in [30], which was released under a dataset challenge competition called ISIC 2019 and contains 25,331 dermoscopy images.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [30] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    Li et al. [20] and Linardos et al. [26] both used MRI images for their models, brain and heart MRI data respectively.

    “In addition, Li et al. [20] and Linardos et al. [26] both used MRI images for their models, brain and heart MRI data consequently.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [20] source

      Not checked in the free run.

    • cites [26] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    The MNIST: HAM10000 repository contains 10,015 dermatoscopic images divided into seven different classes.

    “This repository contains 10,015 dermatoscopic images divided into seven different classes.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [36] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    CheXpert, the largest dataset found under the investigation (overall 223,414 CXR images), was considered by Chakravarty et al. [23].

    “CheXpert, the largest dataset (overall 223,414 CXR image) found under our investigation was considered by Chakravarty et al. [23].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [23] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    VGG19 was used by Lo et al. [16].

    “Likewise, VGG19 is a 19 layers CNN model and used by Lo et al. [16].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    Inception was used in [17], [19], and [22]; AlexNet in [17] and [19].

    “Other pre-trained CNN models are Inception ( [17], [19], [22]), AlexNet ( [17], [19]).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [17] source

      Not checked in the free run.

    • cites [19] source

      Not checked in the free run.

    • cites [22] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    In [18], Feki et al. divided the collected dataset into four parts for client data; for IID they used an equal number of images from both sides, client and class.

    “In [18], Feki et al. divided the collected dataset into four parts for clients data, for IID, they used an equal number of images from both sides, client and class.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    Only four of the 17 reviewed papers considered non-IID data along with IID data, and the remaining 13 did not discuss it.

    “Our results show only four papers (we did not find sufficient explanation from [26] and [21]) considered non-IID type along with IID data and the rest 13 did not talk about the content.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [26] source

      Not checked in the free run.

    • cites [21] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    ResNet18 was used in [23], [26], [27], and [29]; ResNet50 in [18] and [24]; ResNet101 in [24].

    “Residual Network (ResNet) is also a commonly used algorithm that can be constructed by different numbers of layers, e.g., ResNet18 ( [23], [26], [27], [29]), ResNet50 ( [18], [24]), ResNet101 ( [24]).”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [23] source

      Not checked in the free run.

    • cites [26] source

      Not checked in the free run.

    • cites [27] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Adnan et al. [28] performed FL with IID and non-IID data individually, with different numbers of samples in each client under non-IID.

    “Likewise, Adnan et al. [28] performed FL with IID and non-IID data individually where number of samples were different in each client under non-IID scenario.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [28] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Li et al. [20] used a multi-layer perceptron (MLP) classifier, a deep neural network with one input, hidden, and output layer.

    “Li et al. [20] have used multi-layer perceptron (MLP) classifier which was a deep neural network constructed by one input, hidden, and output layers.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [20] source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    Among the three FL categories, Federated Transfer Learning and Vertical FL are rarely considered in medical research, while Horizontal FL was used widely.

    “Among the three types, Federated Transfer Learning (FTL) and Vertical FL (VFL) are rarely considered in medical research; another one, Horizontal FL (HFL) was used widely.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Federated mechanisms in the reviewed articles were executed in two ways: some driven by formerly proposed built-in FL algorithms and others with basic concepts for aggregation.

    “We observed federated mechanisms are executed in two ways, some articles were driven by formerly proposed build-in FL algorithms and others with basic concepts for aggregation.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingnone given

    [20] and [27] used two different federated algorithms named Fed and secure aggregation (SecAgg), respectively.

    “Likewise, [20] and [27] used two different federated algorithms named Fed, secure aggregation (SecAgg) respectively.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [20] source

      Not checked in the free run.

    • cites [27] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Li et al. [20] used two DP mechanisms, Gaussian and Laplace, with noise level α varying from 0.001 to 1.

    “Li et al. [20] used two different mechanisms of DP, Gaussian and Laplace.They defined the noise level α which varied from 0.001 to 1.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [20] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    [24] used three datasets (two X-ray and one CT) and combined both image types to train and test models.

    “[24] have taken three datasets, two X-ray and one CT image based.In the article the authors combined the both types of images and used them to train and test models.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [24] source

      Not checked in the free run.

  • The quote only partly states this claim.

    Findingown data

    Only two reviewed articles were defined as VFL: [24] and [25].

    “Only two articles we defined as VFL; [24] have taken three datasets, two X-ray and one CT image based.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [24] source

      Not checked in the free run.

    • cites [25] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Połap and Woźniak [19] proposed a meta-heuristic search based federated model that calculated average loss of all local models and selected only models scoring higher than average loss for server aggregation.

    “Połap and Woźniak [19] proposed a meta-heuristic search based federated model, first they calculated average loss of all local models and then selected only models that have scored higher than the average loss for aggregation in server.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [19] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Adnan et al. [28] performed image segmentation using Multiple-Instance Learning (MIL) to train local models.

    “Adnan et al. [28] performed image segmentation using a supervised learning approach called Multiple-Instance Learning (MIL) to train the local models.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [28] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Kaissis et al. [27] applied both Gaussian and Laplace DP techniques, while Adnan et al. [28] used only Gaussian noise.

    “Similarly, Kaissis et al. [27] have applied both techniques and Adnan et al. [28] have used only Gaussian noise in their experiments.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [27] source

      Not checked in the free run.

    • cites [28] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Feki et al. [18] utilized HFL with a chest X-ray dataset where features were the same for all clients but samples differed.

    “For example, Feki et al. [18] utilized HFL, they used a chest X-ray image dataset where features are same for all clients but samples are different.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Kaissis et al. [27] used two different datasets for training and testing their FL models, both containing X-ray images (same features) but different data.

    “Interestingly, Kaissis et al. [27] used two different datasets for training and testing their FL models, the fact is both datasets contain X-ray images (same features) and different data.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [27] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    Polap et al. [17] used encryption and blockchain techniques to make their FL model more secure, proposing three learning agents with blockchain in the Data Management Agent (DMA).

    “In addition, Polap et al. [17] used encryption and blockchain techniques to make their FL model more secure.They proposed three different learning agents where blockchain technique was applied in Data Management Agent (DMA).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [17] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Feki et al. [18] performed binary classification with the highest accuracy scores: 94.4% for FL+VGG16 with data augmentation and 93.57% for FL+VGG16.

    “Feki et al. [18] performed binary classification, their accuracy score is highest, for FL+VGG16 with data augmentation model 94.4% and for FL+VGG16 93.57%.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    In [26], Linardos et al. considered AUC, and the highest score achieved by the FL model was 89%.

    “In [26], Linardos et al. considered AUC, the highest score achieved by the FL model was 89%.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    The remaining three articles performed multi-class classification: Chakravarty et al. [23] used 14 classes, Połap et al. [17] seven classes, and Li et al. [20] four classes; their performances were AUC 80%, accuracy 70%, and 76%, respectively.

    “Low: The rest three articles performed multi-class classification, where Chakravarty et al. [23] 14, Połap et al. [17] seven, and Li et al. [20] have considered four classes, their acquired performances are AUC 80%, accuracy 70% and 76% respectively.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [23] source

      Not checked in the free run.

    • cites [17] source

      Not checked in the free run.

    • cites [20] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    The accuracy scores of [30], [25], and [24] were not clear, but they discussed accuracy between 90-95%.

    “Score of [30], [25] and [24] is not clear, they discussed the accuracy between 90-95%.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [30] source

      Not checked in the free run.

    • cites [25] source

      Not checked in the free run.

    • cites [24] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Area Under the Curve (AUC) score was used by three articles.

    “Area Under the Curve (AUC) score three and precision were used two times.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingcitation

    Połap and Woźniak [19] used the inception91 classifier for the FL model and obtained an accuracy of 91%.

    “Połap and Woźniak [19] used the inception91 classifier for the FL model and obtained an accuracy of 91%.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [19] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Lo et al. [16] performed classification and segmentation on different datasets; for the SFU dataset, classification and segmentation accuracy were 88% and 85%, respectively, and classification accuracy for the OHSU dataset was 89%.

    “Lo et al. [16] performed classification and segmentation both tasks on different datasets, the classification and segmentation accuracy for SFU dataset were 88% and 85% respectively, classification accuracy of OHSU dataset was 89%.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Precision was used by two articles.

    “Area Under the Curve (AUC) score three and precision were used two times.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingcitation

    Article [32] achieved an accuracy of 90.61% and [27] achieved 90%.

    “Article [32] and [27] achieved an accuracy of 90.61% and 90% respectively.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [32] source

      Not checked in the free run.

    • cites [27] source

      Not checked in the free run.

  • May combine two claims.

    Findingnone given

    The federated models presented in the reviewed articles are mostly theoretical and less practically implemented; few articles included their open source code.

    “We observed that the federated models presented in the reviewed articles are mostly theoretical and less practically implemented, few articles included their open source code with their articles.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Chakravarty et al. [23] had an AUC score of 80% with FL, while with the same dataset and ML algorithm articles [67] and [68] had 86% and 87% AUC, respectively.

    “Then as well Chakravarty et al. [23] has an AUC score of 80% with FL environment, but with same dataset and ML algorithm article [67] and [68] have 86% and 87% AUC respectively.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [23] source

      Not checked in the free run.

    • cites [67] source

      Not checked in the free run.

    • cites [68] source

      Not checked in the free run.

  • May combine two claims.

    Findingcitation

    Połap et al. [17] achieved 70% accuracy with federated VGG16 and 67% with Inception, while Jain et al. [56] achieved 79.23% with Inception and Liu et al. [57] 87% with ResNet50 on the MNIST: HAM10000 dataset.

    “For instance, Połap et al. [17] have achieved accuracy with federated VGG16 70% and Inception 67%, however in ML part, Jain et al. [56] achieved 79.23% accuracy with Inception and Liu et al. [57] 87% with ResNet50; all of three have considered the MNIST: HAM10000 dataset.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [17] source

      Not checked in the free run.

    • cites [56] source

      Not checked in the free run.

    • cites [57] source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    Data heterogeneity in the reviewed studies could occur in two ways: different numbers of samples (non-IID data) and different data features (VFL) among clients.

    “Our investigation shows data heterogeneity could occur in two ways: number of samples are different (non-IID data) and data features are different (VFL) among the clients.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    All ML models in the comparison had improved accuracy compared to their respective FL models in existing literature; better ML results were found against every FL article.

    “Our investigation shows in Table 7 that all of the ML models have improved accuracy compared to their respective FL models in existing literature, more specifically we found better ML results against every FL article.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: Maintaining conditions of similar datasets, algorithms, and performance measures ensures an accurate comparison between FL and ML papers.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In Lo et al. [16], clear discussion over the experiment procedure is not considered.

    “clear dis-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “cussion over experiment proce-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “dure are not considered.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • May combine two claims.

    Findingown data

    The highest quality score was 9 out of 12, gained by [27], followed by six for [20] and [28] individually.

    “Highest score is 9 out of 12 gained by [27], followed by six for [20] and [28] both articles individually.”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [27] source

      Not checked in the free run.

    • cites [20] source

      Not checked in the free run.

    • cites [28] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing). The quote only partly states this claim.

    Findingcitation

    In Li et al. [20], all segments of the experiment are well presented.

    “All segments of the experiment are well”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · boosted
      “presented”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [20] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Lo et al. [16] provided source code online.

    “provided source code online.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In Lo et al. [16], the number of data samples is very small.

    “Number of data samples is very”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “small”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    In Feki et al. [18], non-IID data is considered.

    “non-IID data considered”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In Połap and Woźniak [19], non-IID data is not considered.

    “Non-IID data is not considered.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [19] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    In Feki et al. [18], the number of data samples is very small.

    “Number of data samples is very”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “small.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    In Połap and Woźniak [19], the proposed FL framework is well described.

    “The proposed FL framework is well de-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · boosted
      “scribed”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [19] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Lo et al. [16] performed both classification and segmentation tasks using a federated learning framework.

    “performed classification and”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “segmentation both tasks using FL frame-”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In Połap and Woźniak [19], result comparison with state-of-the-art is given.

    “result comparison with state-of-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “the-art is given.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [19] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Findingcitation

    In Feki et al. [18], the proposed FL framework is well described.

    “The proposed FL framework is well de-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · boosted
      “scribed”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    In Feki et al. [18], results comparison is given.

    “comparison is given.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Połap et al. [17] used blockchain-based additional privacy for federated models.

    “Blockchain based additional privacy used”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “for federated models.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [17] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    In Połap et al. [17], the accuracy is low.

    “The accuracy is low”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [17] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In Li et al. [20], the number of data samples is very small.

    “Number of data samples is very”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “small”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [20] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    In Li et al. [20], an additional privacy method is used.

    “additional privacy method is used.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [20] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Połap [22] proposed a fuzzy consensus with the FL method.

    “A fuzzy consensus with the FL method”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “proposed”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [22] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In Li et al. [20], non-IID data is not considered.

    “non-IID data is not consid-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “ered.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [20] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingcitation

    In Zhang et al. [24], non-IID data is not considered.

    “Non-IID data is not considered.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [24] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingnone given

    In Połap [22], all of the experiment procedures are presented properly.

    “all of the experiment proce-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “dures are presented properly.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [22] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Zhang et al. [24] proposed a dynamic fusion based vertical FL for COVID-19 detection.

    “A dynamic fusion based vertical FL pro-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “posed for COVID-19 detection.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [24] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    In Yang et al. [21], there is a need to compare results with state-of-the-art methods.

    “Need to compare results with”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “state-of-the-art methods.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [21] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingown data

    Yang et al. [21] presented a federated semi-supervised learning method with sufficient experimental details.

    “A federated semi-supervised learning”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “method is presented with sufficient exper-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “imental details.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [21] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    Li et al. [20] provided source code online.

    “provided source code online”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [20] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Chakravarty et al. [23] applied FL to chest radiograph screening using deep learning methods.

    “FL applied to chest radiograph screening”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “using deep learning methods.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [23] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In Chakravarty et al. [23], performance comparison with state-of-the-art methods is not given.

    “Performance comparison with”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “state-of-the-art methods is not”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “given”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [23] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    In Połap [22], non-IID data is not considered.

    “Non-IID data is not considered.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [22] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).

    Findingcitation

    In Chakravarty et al. [23], non-IID data is not considered.

    “non-IID data is not consid-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “ered.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [23] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).

    Findingnone given

    In Linardos et al. [26], the number of data samples is very small.

    “Number of data samples is very”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “small.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    Qayyum et al. [25] presented a vertical FL based multi-modal COVID-19 diagnosis approach with sufficient experimental details.

    “A vertical FL based multi-modal COVID-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “19 diagnosis approach is presented with”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “sufficient experimental details.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [25] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Adnan et al. [28] proposed a collaborative ML based approach for tissue image data.

    “A collaborative ML based approach is”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “proposed for tissue image data”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [28] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingcitation

    Linardos et al. [26] developed a deep learning based FL model for cardiovascular disease detection.

    “Deep learning based FL model for car-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “diovascular disease detection”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingcitation

    In Qayyum et al. [25], non-IID data is not considered.

    “Non-IID data is not considered.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [25] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).

    Findingcitation

    Linardos et al. [26] provided source code online.

    “provided source code online.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [26] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).

    Findingcitation

    In Adnan et al. [28], performance comparison with state-of-the-art methods is not given.

    “Performance comparison with”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “state-of-the-art methods is not”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “given.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [28] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In Kaissis et al. [27], all required sections and experimental details are presented.

    “All required sections and experimental de-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “tails are presented”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [27] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based). The quote only partly states this claim.

    Findingcitation

    Kaissis et al. [27] provided source code online.

    “provided source code”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “online”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [27] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Findingown data

    In Adnan et al. [28], all necessary sections are included.

    “all neces-”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “sary sections are included.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [28] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Findingcitation

    In Kaissis et al. [27], non-IID data is not considered.

    “Non-IID data is not considered.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [27] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). The quote only partly states this claim.

    Findingcitation

    Yan et al. [29] used FL for a COVID-19 detection model using chest x-ray images.

    “FL for COVID-19 detection model using”
    • Also stated in: § APPENDIX A QUALITY ANALYSIS · plain
      “chest x-ray images.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [29] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S1 (asserted), S3 (load-bearing), S4 (based).

    Findingcitation

    In Kaissis et al. [27], an additional privacy method is used.

    “additional privacy method is used.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [27] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    Połap et al. [17] used AlexNet, VGG16, and Inception, and Basic FL as the federated algorithm.

    “[17] | AlexNet, VGG16, and Inception | Basic FL”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [17] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    Feki et al. [18] used VGG16 and ResNet50, and FedAvg as the federated algorithm.

    “[18] | VGG16 and ResNet50 | FedAvg”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [18] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Findingcitation

    Lo et al. [16] used VGG19 for classification and Basic FL as the federated algorithm.

    “[16] | VGG19 (classification) | Basic FL”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [16] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Naeem et al. [10] worked particularly on brain tumor diagnosis using MRI images.

    “Naeem et al. [10] worked particularly on brain tumor diagnosis using MRI images.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [10] source

      Not checked in the free run.

  • May combine two claims.

    Premisecitation

    Data encryption-decryption could be a potential solution to exchange information between participants, but the process could be complex, time consuming and not sustainable.

    “For such that scenario, data encryption-decryption could be a potential solution to exchange information between participants; however the process could be complex, time consuming and not sustainable [3].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [3] Rauniyar, 2022, Federated learning for medical applications: A taxonomy, current trends, challenges, and future research directions, arXiv:2208.03392

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    A well-known article [12] discussed prime factors related to FL in digital health with challenges and solutions.

    “A well-known article in this field [12], where the authors discussed the prime factors related to FL in digital health with challenges and solutions.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [12] source

      Not checked in the free run.

  • Classified as probably not a substantial claim.

    Study design and conductreasoning

    The search process was limited to 1 January 2017 to 30 June 2022.

    “Since FL was first presented in 2016, the search process of the review was limited over the time period from 1 January 2017 to 30 June 2022.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Xu et al. [11] surveyed papers focusing on FL in the biomedical area to provide a review.

    “Xu et al. [11] surveyed the papers that focus FL in the biomedical area to provide a review.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [11] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation

    FL was first introduced in 2016 and gained attraction in the healthcare domain within the last couple of years.

    “It was first introduced in 2016 [4] and gained a lot of attraction within last couple of years for the healthcare domain.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [4] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning

    153 data samples are not technically sound.

    “Since larger dataset belongs to better potentiality of inside analysis, literally 153 data samples are not technically sound.”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: More training data leads to better prediction and smaller datasets increase overfitting risk.

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning

    X-ray with CT or ultrasound images are technically different, so their features differ, and using various data features in different clients makes a VFL scenario.

    “Xray with CT or ultrasound images are technically different, thus their features will be also different and they used various data features in different clients which makes a VFL scenario.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    The review could not retrieve some papers ([94] and [95]) due to subscription download policies, limiting the range of the survey.

    “However, still we failed to reach some of them ( [94] and [95]) which is limiting the range of this survey.”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [94] source

      Not checked in the free run.

    • cites [95] source

      Not checked in the free run.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductreasoning

    It is difficult to conclude this study with strong and tested historical evidence because the review was on very limited time and with insufficient resources since FL was recently introduced.

    “Overall, it is difficult to conclude this study with strong and tested historical evidence, because our review was on very limited time and with insufficient resources since FL was recently introduced.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Every privacy mechanism comes with a significant computational cost on the federation.

    “But the reality is every privacy mechanism comes with a significant computational cost on the federation.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationreasoning

    Practically, features of federated datasets are not the same in many cases, for instance X-ray and CT images can be used in two different clients, which makes trouble during central aggregation of model parameters in FL.

    “Furthermore, practically features of federated datasets are not the same in many cases, for instance X-ray and CT images data can be used in two different clients which makes trouble during aggregate the models parameters centrally in a FL setting.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    The inclusion and exclusion process removed articles from the initial fleet, preprint articles were not included, and not all searching databases were explored, so relevant articles may have been missed.

    “In addition, our inclusion and exclusion process removed articles from the initial fleet and preprint articles were not included there, besides we could not explore all of the searching databases so it could be possible that we missed to include any relevant article(s) on the topic.”
    Quote in the paperNot a substantial claim
106 of 106

  • The paper relies on this without support.

    Medical images and patient data are sensitive and require protection from disclosure.

    “Since medical images are sensitive data, it needs to be protected and preserves the rights of users' personal information.”

  • The paper relies on this without support.

    FL is privacy-preserving and addresses data protection concerns.

    “It addresses the privacy and data protection concern, which is currently an important problem in developing medical AI.”
    “Hence, FL is a privacy preserving way of training AI algorithms, allows to move the model to the data rather than moving the data to the model and this makes it very useful in cases where sensitive data cannot be shared.”

  • The paper relies on this without support.

    More training data leads to better prediction and smaller datasets increase overfitting risk.

    “In ML research, it is very established that the more data we have for training purposes the better prediction we will get from the models.Also, chances of model overfitting will increase when we have a smaller dataset; so, it is always advisable to use a larger dataset.”

  • The paper relies on this without support.

    Many articles have been published on FL-oriented medical image analysis.

    “Already many articles have been published on FL oriented medical image analysis and they successfully applied this unique data management technique in their research articles.”

  • The paper relies on this without support.

    Maintaining conditions of similar datasets, algorithms, and performance measures ensures an accurate comparison between FL and ML papers.

    “We expect maintaining this condition will ensure an accurate comparison among the two parties.”

  • The paper relies on this without support.

    FL mainly preserves privacy because trained models of stakeholders are shared instead of sharing data directly.

    “FL mainly preserves the privacy concern since trained models of stakeholders are shared instead of sharing data directly.”

  • The paper relies on this without support.

    Medical image analysis or computer vision tasks are mostly conducted by CNN-oriented deep learning models.

    “Since our review is based on medical image data and this image analysis or computer vision task is mostly conducted by CNN oriented deep learning models.”

  • The paper relies on this without support.

    The 17 reviewed articles are representative of FL research in medical image analysis.

    “In this study, we systematically reviewed the articles that considered FL in their ML based medical image research.”
8 of 8

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • Some IoT devices and local systems in healthcare lack the computational power to handle resource-intensive model training, making it difficult to implement advanced AI techniques like deep learning in low-resource settings.

    “Some IoT devices and local systems in healthcare lack the computational power to handle resource-intensive model training, making it difficult to implement advanced AI techniques like DL in low-resource settings [40].”

    From Federated Learning in Smart Healthcare: A Comprehensive Review on Privacy, Security, and Predictive Analytics with IoT Integration (2024) · cites this paper as [40]

    The Introduction describes training models locally at hospitals, and RQ6 says medical-image analysis commonly uses CNN-based deep learning. Neither section says that healthcare IoT devices or local systems lack the computational power for training. Searches for computational resources, device processing power, and low-resource training did not find a passage addressing that limitation. (LLM, conf=0.72)

    Experimental — accuracy not yet measured

  • A recent survey also reviews the application of federated learning on medical image analysis.

    “A recent survey also reviews the application of federated learning on medical image analysis [27].”

    From Federated learning for medical image analysis: A survey (2024) · cites this paper as [27]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • A private-dataset study was included despite the exclusion criterion

    Major

    The review says studies using private datasets were excluded, but its dataset listing identifies the included study [21] as using private data.

  • Study [32] has two different reported accuracies

    Minor

    The performance discussion reports 90.61% for [32], but Table 7 reports 91.61%.

  • Study [32] has conflicting publication years

    Minor

    The results listing dates Salam et al. [32] to 2022, while its reference gives 2021.

  • An 80% result is placed in the below-80% category

    Minor

    The paper defines its low-performance category as below 80%, then places study [23], reported as having an AUC of 80%, in that category.

  • Study [25] has conflicting publication years

    Minor

    Table 4 dates Qayyum et al. [25] to 2021, while its reference gives 2022.

  • Study [29]'s sensitivity is presented in an accuracy comparison

    Minor

    The discussion identifies 91.26% as a sensitivity score for [29], but Table 7 labels the corresponding FL-ML comparison as an accuracy comparison.

6 of 6

Cites

1 retracted

Flagged first, then by citations

Research works printed but not linked: 9

  1. McMahan, 2016, Communication-efficient learning of deep networks from decentralized data, arXiv:1602.05629 · arXiv
  2. 10.1371/joumal.pmed1000097 · DOI
  3. Irvin, CheXpert: A large chest radiograph dataset with uncertainty labels and expert comparison, 590
  4. 2022, Agchung/Figure1-COVID-Chestxray-Dataset: Figure 1 COVID-19
  5. 2022, Project-Monai/Monai: AI Toolkit for Healthcare Imaging
  6. 2022, Medical Segmentation Decathlon
  7. 2022, ISIC Challenge
  8. Patel, 2020, Chest X-ray (COVID-19& Pneumonia). Kaggle
  9. 2021, Novel Corona Virus 2019 Dataset

Cited by

Sorted by citations

Related

Authors
Sohan, M. F., & Basalamah, A.
Affiliations
  • Daffodil International University (BD)
  • Umm al-Qura University (SA)
ORCID
Venue
IEEE Access
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Published
2023
Volume
11
Pages
28628-28644
ISSN
2169-3536 (ISSN-L)
DOI
10.1109/access.2023.3260027
OpenAlex
W4352977741
Cited by
114
References
88
Citations by year
2026: 31 · 2025: 51 · 2024: 22 · 2023: 10
FWCI
2.38
Type
journal article
Language
en
Open access
Licence
Locations
  • IEEE Access (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Greater South Information System
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
Keywords
federated learning, medical image analysis, privacy preservation, decentralized data processing, medical data protection, deep learning, federated model

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-09.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-09.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review