CoChat

Distributed deep learning networks among institutions for medical imaging

Open access
Chang, Ken, Balachandar, Niranjan, Lam, Carson, Yi, Darvin, Brown, James, Beers, Andrew, Rosen, Bruce, Rubin, Daniel L, Kalpathy-Cramer, Jayashree · 2018 · Journal of the American Medical Informatics Association

The DOI matches its Crossref record; 21 of the 21 references it cites were found in a registry or an index.

Abstract

Objective: Deep learning has become a promising approach for automated support for clinical diagnosis. When medical data samples are limited, collaboration among multiple institutions is necessary to achieve high algorithm performance. However, sharing patient data often has limitations due to technical, legal, or ethical concerns. In this study, we propose methods of distributing deep learning models as an attractive alternative to sharing patient data. Methods: We simulate the distribution of deep learning models across 4 institutions using various training heuristics and compare the results with a deep learning model trained on centrally hosted patient data. The training heuristics investigated include ensembling single institution models, single weight transfer, and cyclical weight transfer. We evaluated these approaches for image classification in 3 independent image collections (retinal fundus photos, mammography, and ImageNet). Results: We find that cyclical weight transfer resulted in a performance that was comparable to that of centrally hosted patient data. We also found that there is an improvement in the performance of cyclical weight transfer heuristic with a high frequency of weight transfer. Conclusions: We show that distributing deep learning models is an effective alternative to sharing patient data. This finding has implications for any collaborative deep learning study.

Overview

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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

Integrity signals

Registry

Crossref · 10.1093/jamia/ocy017

Cites

21 of 21 references verified0 citation errors · 0 not checkable

Statistics

Figures

Figure 1 (A) ResNet-34 architecture was utilized for the Diabetic Retinopathy dataset. (B) The dataset was randomly divided into 4 institutions along with a validation and testing set. (C) The learning rate decayed to 0.25 of its value when the same input samples were inputted into the network 20 times at a given learning rate without an improvement of the validation loss.
Figure 1 (A) ResNet-34 architecture was utilized for the Diabetic Retinopathy dataset. (B) The dataset was randomly divided into 4 institutions along with a validation and testing set. (C) The learning rate decayed to 0.25 of its value when the same input samples were inputted into the network 20 times at a given learning rate without an improvement of the validation loss.CC BY-NC · PubMed Central
Figure 2 Model training heuristics investigated include (A) centrally hosted, (B) ensemble single institution models, (C) single weight transfer, and (D) cyclical weight transfer.
Figure 2 Model training heuristics investigated include (A) centrally hosted, (B) ensemble single institution models, (C) single weight transfer, and (D) cyclical weight transfer.CC BY-NC · PubMed Central
Figure 3 Performance of a neural network when trained on (A) Institution 1, (B) Institution 2, (C) Institution 3, and (D) Institution 4 for the Diabetic Retinopathy dataset. The training and validation accuracies for a model trained with the centrally hosted training and single weight transfer training heuristics are shown in (E) and (F), respectively.
Figure 3 Performance of a neural network when trained on (A) Institution 1, (B) Institution 2, (C) Institution 3, and (D) Institution 4 for the Diabetic Retinopathy dataset. The training and validation accuracies for a model trained with the centrally hosted training and single weight transfer training heuristics are shown in (E) and (F), respectively.CC BY-NC · PubMed Central
Figure 4 Training and validation accuracies during training on the Diabetic Retinopathy dataset with cyclical weight transfer with weight transfer frequencies of every (A) 20 epochs, (B) 10 epochs, (C) 5 epochs, (D) 4 epochs, (E) 2 epochs, or (F) every epoch.
Figure 4 Training and validation accuracies during training on the Diabetic Retinopathy dataset with cyclical weight transfer with weight transfer frequencies of every (A) 20 epochs, (B) 10 epochs, (C) 5 epochs, (D) 4 epochs, (E) 2 epochs, or (F) every epoch.CC BY-NC · PubMed Central
Figure 5 (A) Testing accuracies of centrally hosted training, ensembling models, single weight transfer, and cyclical weight transfer for our 4 “institution” experiment on the Diabetic Retinopathy dataset. Cyclical weight transfer had the performance that was on par with centrally hosted training. (B) To show distributed computation on a larger scale, we performed a 20 “institution” experiment with n = 300 patients per institution. The plot shown is the testing accuracy as a function of the number of collaborating institutions. All models were trained using the cyclical weight transfer training heuristic with a weight exchange frequency of 1. For reference, testing accuracy expected from random classification (bottom dotted line) and centrally hosted data (n = 6000 patients, top dotted line) are shown.
Figure 5 (A) Testing accuracies of centrally hosted training, ensembling models, single weight transfer, and cyclical weight transfer for our 4 “institution” experiment on the Diabetic Retinopathy dataset. Cyclical weight transfer had the performance that was on par with centrally hosted training. (B) To show distributed computation on a larger scale, we performed a 20 “institution” experiment with n = 300 patients per institution. The plot shown is the testing accuracy as a function of the number of collaborating institutions. All models were trained using the cyclical weight transfer training heuristic with a weight exchange frequency of 1. For reference, testing accuracy expected from random classification (bottom dotted line) and centrally hosted data (n = 6000 patients, top dotted line) are shown.CC BY-NC · PubMed Central

Verification

Claims the paper makes

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

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

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

Show all 21 references

Research works printed but not linked: 4

  1. Graham, 2015, Kaggle Diabetic Retinopathy Detection Competition Report
  2. Chollet, 2015, Keras: Deep Learning library for Theano and TensorFlow, GitHub Repos, 1
  3. Theano Development Team, 2016, Theano: a Python framework for fast computation of mathematical expressions, arXiv e-prints, 19
  4. USF Digital Mammography, DDSM: Digital Database for Screening Mammography

Cited by

See all 403

Metadata

Authors
Chang, K., Balachandar, N., Lam, C., Yi, D., Brown, J., Beers, A., Rosen, B., Rubin, D. L., & Kalpathy-Cramer, J.
Affiliations
  • Massachusetts General Hospital (US)
  • Stanford University (US)
  • Athinoula A. Martinos Center for Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Charlestown, MA, 02129, USA
Corresponding author
Rubin, Daniel L
Venue
Journal of the American Medical Informatics Association
Published
2018-03-29
Volume
25
Issue
8
Pages
945-954
ISSN
  • 1067-5027 (ISSN-L)
  • 1527-974X
OpenAlex
W2795354529
Cited by
340
References
21
Citations by year
2026: 21 · 2025: 30 · 2024: 44 · 2023: 45 · 2022: 64 · 2021: 53 · 2020: 47 · 2019: 24 · 2018: 12
FWCI
19.16
Type
journal article
Language
en
Open access
Locations
  • Journal of the American Medical Informatics Association (publishedVersion, cc-by-nc)
  • Europe PMC (PubMed Central) (publishedVersion, cc-by-nc)
  • PubMed Central (publishedVersion, cc-by-nc)
  • PubMed (publishedVersion)
  • Lincoln Repository (University of Lincoln) (acceptedVersion)
  • Figshare (submittedVersion, cc-by-nc)
Funders
  • National Institutes of Health Blueprint for Neuroscience Research (T90DA022759/R90DA023427)
  • NIH
  • National Institute of Biomedical Imaging and Bioengineering
  • National Institutes of Health (U01CA154601, U24CA180927, U24CA180918, U01CA190214, U01CA187947)
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
distributed deep learning, multi-institutional collaboration, model ensembling, limited medical data, retinal fundus image classification, mammography classification, ImageNet classification, federated learning
MeSH terms
Deep Learning, Computer Communication Networks, Diagnostic Imaging, Humans, Medical Record Linkage, Neural Networks, Computer

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Distributed deep learning networks among institutions for medical imaging. https://app.cochat.ai/papers/10.1093/jamia/ocy017
The paper
Chang, K., Balachandar, N., Lam, C., Yi, D., Brown, J., Beers, A., Rosen, B., Rubin, D. L., & Kalpathy-Cramer, J. (2018). Distributed deep learning networks among institutions for medical imaging. Journal of the American Medical Informatics Association. https://doi.org/10.1093/jamia/ocy017

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.