CoChat
Sign Up
Cited by 139Privacy-Preserving Technologies in DataArtificial Intelligence in Healthcare and EducationDomain Adaptation and Few-Shot Learning

The DOI matches its Crossref record; 56 of the 56 references it cites were found in a registry or an index.

Federated learning for multi-center imaging diagnostics: a simulation study in cardiovascular disease

Open access
Linardos, Akis, Kushibar, Kaisar, Walsh, Sean, Gkontra, Polyxeni, Lekadir, Karim·2022·Scientific Reports
DOI 10.1038/s41598-022-07186-4PMC PMC8894335
Open this paper in CoChat — free accountOpen PDF

Abstract

Deep learning models can enable accurate and efficient disease diagnosis, but have thus far been hampered by the data scarcity present in the medical world. Automated diagnosis studies have been constrained by underpowered single-center datasets, and although some results have shown promise, their generalizability to other institutions remains questionable as the data heterogeneity between institutions is not taken into account. By allowing models to be trained in a distributed manner that preserves patients’ privacy, federated learning promises to alleviate these issues, by enabling diligent multi-center studies. We present the first simulated federated learning study on the modality of cardiovascular magnetic resonance and use four centers derived from subsets of the M&M and ACDC datasets, focusing on the diagnosis of hypertrophic cardiomyopathy. We adapt a 3D-CNN network pretrained on action recognition and explore two different ways of incorporating shape prior information to the model, and four different data augmentation set-ups, systematically analyzing their impact on the different collaborative learning choices. We show that despite the small size of data (180 subjects derived from four centers), the privacy preserving federated learning achieves promising results that are competitive with traditional centralized learning. We further find that federatively trained models exhibit increased robustness and are more sensitive to domain shift effects.

Overview

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1038/s41598-022-07186-4

Cites

56 of 56 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 A diagram of the two collaborative learning frameworks. For FL only models are transferred, while in the CDS case, the data itself is transferred to the central server and all training happens there.
Figure 1 A diagram of the two collaborative learning frameworks. For FL only models are transferred, while in the CDS case, the data itself is transferred to the central server and all training happens there.CC BY · PubMed Central
Figure 2 An example of the induced priors used in this study: (A) a baseline that is the 1-channel cardiac MRI, (B) the baseline multiplied by the segmentation mask, (C) the baseline split into three channels, one for each part of the heart (right ventricle blood pool, left ventricle blood pool, and myocardium).
Figure 2 An example of the induced priors used in this study: (A) a baseline that is the 1-channel cardiac MRI, (B) the baseline multiplied by the segmentation mask, (C) the baseline split into three channels, one for each part of the heart (right ventricle blood pool, left ventricle blood pool, and myocardium).CC BY · PubMed Central
Figure 3 Illustration of augmentation techniques used in our analysis on a single CMR image slice.
Figure 3 Illustration of augmentation techniques used in our analysis on a single CMR image slice.CC BY · PubMed Central
Figure 4 An illustration of a single iteration for: (A) Collaborative Cross Validation (CCV), where centers coordinate their splitting across 5 folds so that each center provides 20% of its data as test set and the rest as training and validation, (B) Leave Center Out Cross Validation (LCO-CV) which runs for as many iterations as there are centers in the dataset, each time using a different center as test set, and the rest as training and validation.
Figure 4 An illustration of a single iteration for: (A) Collaborative Cross Validation (CCV), where centers coordinate their splitting across 5 folds so that each center provides 20% of its data as test set and the rest as training and validation, (B) Leave Center Out Cross Validation (LCO-CV) which runs for as many iterations as there are centers in the dataset, each time using a different center as test set, and the rest as training and validation.CC BY · PubMed Central
Figure 5 Histograms of different center intensities before and after histogram standardization. The two evaluation techniques imply different adaptation set-ups as for the CCV we are standardizing using an average from all of the centers, while for LCO-CV we consider one center entirely unseen.
Figure 5 Histograms of different center intensities before and after histogram standardization. The two evaluation techniques imply different adaptation set-ups as for the CCV we are standardizing using an average from all of the centers, while for LCO-CV we consider one center entirely unseen.CC BY · PubMed Central
Figure 6 An overview of the pipeline used for the experiments in this study.
Figure 6 An overview of the pipeline used for the experiments in this study.CC BY · PubMed Central
Figure 7 CDS, FL and FL-EV are tested given four data augmentation set-ups and for each set-up the experiment is repeated with 5 different network initializations, both for the CCV and LCO-CV set-ups. AUC is evaluated across all folds and we obtain 5 AUC metrics per set-up whose error bars are displayed here.
Figure 7 CDS, FL and FL-EV are tested given four data augmentation set-ups and for each set-up the experiment is repeated with 5 different network initializations, both for the CCV and LCO-CV set-ups. AUC is evaluated across all folds and we obtain 5 AUC metrics per set-up whose error bars are displayed here.CC BY · PubMed Central

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.

A few free checks a day · papers already verified open instantly · results are public

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?

Self-check

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

The self-check needs the paper’s full text.

Citations

Cites

Research works printed but not linked: 3

  1. McMahan, B., Moore, E., Ramage, D., Hampson, S. & y Arcas, B. A. Communication-efficient learning of deep networks from decentralized data. In Artificial Intelligence and Statistics, 1273–1282 (PMLR, 2017).
  2. Bakas, S. et al. Identifying the best machine learning algorithms for brain tumor segmentation, progression assessment, and overall survival prediction in the brats challenge. arXiv preprint arXiv:1811.02629 (2018). · arXiv
  3. Glorot, X. & Bengio, Y. Understanding the difficulty of training deep feedforward neural networks. In Proceedings of the thirteenth international conference on artificial intelligence and statistics (2010).

Cited by

Sorted by citations

See all 139

Related

See all

Metadata

Authors
Linardos, A., Kushibar, K., Walsh, S., Gkontra, P., & Lekadir, K.
Affiliations
Universitat de Barcelona (ES)
Corresponding author
Linardos, Akis
ORCID
Venue
Scientific Reports
Publisher
Springer Science and Business Media LLC
Published
2022-03-03
Volume
12
Issue
1
Pages
3551-3551
ISSN
2045-2322 (ISSN-L)
DOI
10.1038/s41598-022-07186-4
PMC
PMC8894335
OpenAlex
W4214841549
Cited by
130
References
53
Citations by year
2026: 40 · 2025: 39 · 2024: 25 · 2023: 19 · 2022: 7
FWCI
7.40
Type
journal article
Language
en
Open access
Licence
Locations
  • Scientific Reports (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Dipòsit Digital de la Universitat de Barcelona (Universitat de Barcelona) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
European Union’s Horizon 2020 research and innovation programme (952103)
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, cardiovascular magnetic resonance, hypertrophic cardiomyopathy, data heterogeneity, privacy-preserving machine learning, 3D convolutional neural networks, data augmentation, domain shift, model generalization, ACDC dataset
MeSH terms
Cardiovascular Diseases, Computer Simulation, Confidentiality, Humans, Magnetic Resonance Imaging, Privacy

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.

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