Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 61 of the 68 references it cites were found in a registry or an index; 10 of the 31 sources its claims cite were checked for support.

A scoping review of robustness concepts for machine learning in healthcare

Open access
Balendran, Alan, Beji, Céline, Bouvier, Florie, Khalifa, Ottavio, Evgeniou, Theodoros, Ravaud, Philippe, Porcher, Raphaël · 2025 · npj Digital Medicine

Abstract

While machine learning-based solutions have demonstrated strong performance, robustness remains ambiguous and often overlooked. Eight general concepts of robustness emerged from 274 included records.

Overview

CoChat Summary

TL;DR

This scoping review mapped how robustness is defined in healthcare machine learning literature and found 8 distinct robustness concepts across 274 eligible records. The concepts differed by data type and model type, suggesting that “robustness” is not a single uniform property in healthcare ML.

Key findings

  • The review included 274 eligible records from 8585 initially retrieved records; 190 (69%) were journal articles and 81 (30%) were from conferences, workshops, or symposiums.
  • Across 526 extracted combinations, the authors identified 8 robustness concepts: input perturbations and alterations, missing data, label noise, imbalanced data, feature extraction and selection, model specification and learning, external data and domain shift, and adversarial attacks.
  • The most frequently addressed concept was input perturbations and alterations (142/526, 27%), while imbalanced data was the least commonly tackled (15/526, 3%).
  • Robustness concepts varied by data and model: feature extraction/selection was mainly emphasized in image-derived data (33%) and omics (22%); adversarial attacks were mainly tackled in image (22%) and physiological signal data (7%); missing data was mostly addressed in clinical data (20%); deep learning applications covered the most concepts (8) and hybrid models the fewest (4).

Methods

  • This was a scoping review conducted and reported per Joanna Briggs Institute methodology and PRISMA-ScR, with a protocol uploaded on Open Science Framework.
  • The authors searched PubMed, IEEE Xplore, and Web of Science (search date: 1 March 2023), plus grey literature sources (Google Scholar, conference/workshop websites) and reference lists; they included English records with abstracts on supervised or semi-supervised ML models for healthcare decision support.
  • Data extraction captured medical application, model, data type, notion of robustness, and robustness metrics; extracted records were categorized by medical specialty, model class, and data type, then robustness notions were grouped into general concepts through iterative reviewer discussion.

Limitations

  • The paper states it was not meant to provide or assess methods for “robustifying” a model; it only maps robustness concepts.
  • The review is mostly restricted to studies using the terms “robust,” “noise,” and “perturbation”; the authors note that using other terms (e.g., stability, resilience, reliability, vulnerability) might have yielded different studies.
  • The review is temporally limited and does not include more recent foundation-model studies published after the search date.
  • Evident from the text: the search was limited to English-language records with abstracts, and the screening/search strategy may therefore have missed relevant non-English or abstractless studies.

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

Integrity signals

Registry

Crossref · 10.1038/s41746-024-01420-1

Cites

61 of 61 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 PRISMA flow diagram. The flow chart indicates the different number of records kept and excluded at each step of the screening process.
Fig. 1 PRISMA flow diagram. The flow chart indicates the different number of records kept and excluded at each step of the screening process.CC BY · PubMed Central
Fig. 2 The eight robustness concepts. The radial dendrogram shows the distribution of the eight robustness concepts based on combinations of predictive models and data types from the included studies. Examples of different sources of variations for each concept are also provided. The arrow indicates the order in which these concepts can intervene during the lifecycle of a machine learning model.
Fig. 2 The eight robustness concepts. The radial dendrogram shows the distribution of the eight robustness concepts based on combinations of predictive models and data types from the included studies. Examples of different sources of variations for each concept are also provided. The arrow indicates the order in which these concepts can intervene during the lifecycle of a machine learning model.CC BY · PubMed Central
Fig. 3 Stratified view of the eight concepts of robustness. The eight robustness concepts are stratified by data type a and predictive model type b. The emphasis on specific robustness concepts varies depending on the type of data or predictive model used. For instance, panel a shows that the feature extraction and selection concept is more commonly emphasized for high-dimensional tabular data, such as omics or image-derived features. Similarly, panel b shows that adversarial attacks are primarily addressed in deep learning-based applications.
Fig. 3 Stratified view of the eight concepts of robustness. The eight robustness concepts are stratified by data type a and predictive model type b. The emphasis on specific robustness concepts varies depending on the type of data or predictive model used. For instance, panel a shows that the feature extraction and selection concept is more commonly emphasized for high-dimensional tabular data, such as omics or image-derived features. Similarly, panel b shows that adversarial attacks are primarily addressed in deep learning-based applications.CC BY · PubMed Central

Verification

Claims the paper makes

8 Valid · 0 Not valid · 10 Not checked · 43 Own results and interpretations

29 more claims

Create a free account to see all 61 claims and to verify the 8 references still open.

See all claims — free account
  • It may state two separate claims as one. 19
  • Probably not a substantial claim. 14

Create a free account to see these claims.

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?

Citations

Cites

Show all 68 references

Research works printed but not linked: 1

  1. Saab, K. et al. Reducing Reliance on Spurious Features in Medical Image Classification with Spatial Specificity. in Proceedings of the 7th Machine Learning for Healthcare Conference 760–784 (PMLR, 2022).

Cited by

See all 43

Metadata

Authors
Balendran, A., Beji, C., Bouvier, F., Khalifa, O., Evgeniou, T., Ravaud, P., & Porcher, R.
Affiliations
  • Inserm (FR)
  • INSEAD (FR)
Corresponding author
Balendran, Alan
Venue
npj Digital Medicine
Published
2025-01-17
Volume
8
Issue
1
Pages
38-38
ISSN
2398-6352 (ISSN-L)
OpenAlex
W4406494794
Cited by
43
References
68
Citations by year
2026: 25 · 2025: 18
FWCI
10.28
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Digital Medicine (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • HAL (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • Bpifrance
  • Agence Nationale de la Recherche (ANR-23-IACL-0008, ANR-19-P3IA-0001)
Topics
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
  • Physical Sciences › Computer Science › Artificial Intelligence › Adversarial Robustness in Machine Learning
Keywords
robustness, machine learning, artificial intelligence, health care, model stability, generalizability, model validation, model deployment, interpretability

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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: A scoping review of robustness concepts for machine learning in healthcare. https://app.cochat.ai/papers/10.1038/s41746-024-01420-1
The paper
Balendran, A., Beji, C., Bouvier, F., Khalifa, O., Evgeniou, T., Ravaud, P., & Porcher, R. (2025). A scoping review of robustness concepts for machine learning in healthcare. npj Digital Medicine. https://doi.org/10.1038/s41746-024-01420-1

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.