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

The DOI matches its Crossref record; 95 of the 95 references it cites were found in a registry or an index; 61 of the 76 sources its claims cite were checked for support.

Orchestrating explainable artificial intelligence for multimodal and longitudinal data in medical imaging

Open access
Pahud de Mortanges, Aurélie, Luo, Haozhe, Shu, Shelley Zixin, Kamath, Amith, Suter, Yannick, Shelan, Mohamed, Pöllinger, Alexander, Reyes, Mauricio · 2024 · npj Digital Medicine

Abstract

Explainable artificial intelligence (XAI) has experienced a vast increase in recognition over the last few years. While the technical developments are manifold, less focus has been placed on the clinical applicability and usability of systems. Moreover, not much attention has been given to XAI systems that can handle multimodal and longitudinal data, which we postulate are important features in many clinical workflows. In this study, we review, from a clinical perspective, the current state of XAI for multimodal and longitudinal datasets and highlight the challenges thereof. Additionally, we propose the XAI orchestrator, an instance that aims to help clinicians with the synopsis of multimodal and longitudinal data, the resulting AI predictions, and the corresponding explainability output. We propose several desirable properties of the XAI orchestrator, such as being adaptive, hierarchical, interactive, and uncertainty-aware.

Overview

CoChat Summary

## TL;DR

This narrative review argues that explainable AI for medical imaging needs to better handle multimodal and longitudinal data, because clinical decisions are usually made from the synopsis of many information sources over time. The paper does not present a new experimental study; instead, it proposes an “XAI orchestrator” concept and a set of desirable properties for such a system.

## Key findings

  • The authors state that there is a lack of studies that enrich radiological XAI systems with other clinical data types or use longitudinal datasets, and that merging these data types into a meaningful overall explanation has received little attention.
  • In a cited multimodal XAI example, Soenksen et al.’s HAIM framework tested “over 14’000 different prediction models,” and multimodal predictions outperformed unimodal comparators by “6-30%.”
  • For the XAI orchestrator, the paper proposes properties including being adaptive, hierarchical, uncertainty-aware, interactive, time effective, causality- and co-dependency aware, modular, privacy preserving, resilient to data drift, and up to date.
  • The authors suggest future implementations could combine modality-specific encoders, a central LLM-based decoder, and retrieval-augmented generation to produce user-friendly explanations and answer follow-up questions.

## Methods

  • Narrative review from a clinical perspective of the current state of XAI for multimodal and longitudinal datasets in medical imaging.
  • Synthesis of prior work on multimodal XAI, longitudinal XAI, XAI toolkits, and LLM-based medical applications; the paper also discusses illustrative clinical case examples in the supplementary materials.
  • Conceptual proposal of the “XAI orchestrator,” including desirable properties, functionalities, and metrics, but no original dataset analysis or model evaluation is reported.

## Limitations

  • The paper explicitly states: “we do not provide a full implementation and results of the XAI orchestrator.”
  • It also notes that the review is limited by the current state of the literature, where many methods remain at a prototype level and have not yet been extensively applied to real-world clinical questions.
  • Evident from the text: the article is conceptual and does not validate the proposed orchestrator on patient data, so its effectiveness, reliability, and clinical utility are not empirically tested in this paper.

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

Integrity signals

Registry

Crossref · 10.1038/s41746-024-01190-w

Cites

95 of 95 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig. 1 Improving over current saliency maps for longitudinal scenarios. The delta saliency map. In this example case of interstitial pulmonary fibrosis, the left image (a) was taken around two years prior to the middle image (b). During the two years, the disease progressed heavily. The delta saliency map (c) shows this disease progression through the yellow, orange, and red color overlays. The frontal and dorsal areas of the lungs, which are heavily affected, as well as the subpleural areas, are expected to contribute most to the classification and are therefore overlayed with the highest opacity of color, whereas the extrapulmonary areas are only lightly overlayed as they are expected to contribute only marginally.
Fig. 1 Improving over current saliency maps for longitudinal scenarios. The delta saliency map. In this example case of interstitial pulmonary fibrosis, the left image (a) was taken around two years prior to the middle image (b). During the two years, the disease progressed heavily. The delta saliency map (c) shows this disease progression through the yellow, orange, and red color overlays. The frontal and dorsal areas of the lungs, which are heavily affected, as well as the subpleural areas, are expected to contribute most to the classification and are therefore overlayed with the highest opacity of color, whereas the extrapulmonary areas are only lightly overlayed as they are expected to contribute only marginally.CC BY · PubMed Central
Fig. 2 Conceptual description of the XAI orchestrator. Clinical guidelines and recent research, constitute the knowledge base of the XAI Orchestrator. Additionally, multimodal patient-specific data is collected. After outcome prediction, XAI methods are applied to generate modality-specific or time-specific explanations. The superordinate XAI orchestrator aggregates all information and generates a comprehensive overall explanation while enabling further inquiries by an expert. Figure created with BioRender.com.
Fig. 2 Conceptual description of the XAI orchestrator. Clinical guidelines and recent research, constitute the knowledge base of the XAI Orchestrator. Additionally, multimodal patient-specific data is collected. After outcome prediction, XAI methods are applied to generate modality-specific or time-specific explanations. The superordinate XAI orchestrator aggregates all information and generates a comprehensive overall explanation while enabling further inquiries by an expert. Figure created with BioRender.com.CC BY · PubMed Central
Fig. 3 Potential implementation of the XAI orchestrator. Multimodal encodings of patient data, combined with retrieved context information and user prompts feed the decoder, which produces explanations for the user. The user’s question, encoded as a prompt, together with the prior medical knowledge, retrieved e.g. from scientific literature data bases like PubMed are sent to the decoder through retrieval augmented generation (RAG). The central XAI orchestrator decoder is constructed with multiple transformer decoder layers, which generate a textual response to the input question.
Fig. 3 Potential implementation of the XAI orchestrator. Multimodal encodings of patient data, combined with retrieved context information and user prompts feed the decoder, which produces explanations for the user. The user’s question, encoded as a prompt, together with the prior medical knowledge, retrieved e.g. from scientific literature data bases like PubMed are sent to the decoder through retrieval augmented generation (RAG). The central XAI orchestrator decoder is constructed with multiple transformer decoder layers, which generate a textual response to the input question.CC BY · PubMed Central

Verification

Claims the paper makes

35 Valid · 12 Not valid · 3 Not checked · 21 Own results and interpretations

15 more claims · 9 with their cited sources checked

Create a free account to see all 71 claims and every source verdict.

See all claims — free account
  • It may state two separate claims as one. 24
  • Probably not a substantial claim. 42

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 89 references

Research works printed but not linked: 5

  1. Klaise, Janis, J., Van Looveren, A., Vacanti, G. & Coca, A. Alibi explain: algorithms for explaining machine learning models. JMLR. 22, 1–7 (2021).
  2. Alber, M. et al. iNNvestigate neural networks! JMLR 20, 1–8 (2019).
  3. Ribeiro, F. D. S., Xia, T., Monteiro, M., Pawlowski, N. & Glocker, B. High fidelity image counterfactuals with probabilistic causal models. Proceedings of the 40th International Conference on Machine Learning. PMLR202. (2023).
  4. Baniecki, H., Kretowicz, W., Piatyszek, P., Wisniewski, J. & Biecek, P. dalex: responsible machine learning with interactive explainability and fairness in Python. JMLR 22, 1–7 (2021).
  5. Li, X. et al. InterpretDL: explaining deep models in PaddlePaddle. JMLR 23, 1–6 (2022).

Cited by

See all 93

Metadata

Authors
Pahud de Mortanges, A., Luo, H., Shu, S. Z., Kamath, A., Suter, Y., Shelan, M., Pöllinger, A., & Reyes, M.
Affiliations
  • University of Bern (CH)
  • University Hospital of Bern (CH)
Venue
npj Digital Medicine
Published
2024-07-22
Volume
7
Issue
1
Pages
195-195
ISSN
2398-6352 (ISSN-L)
OpenAlex
W4400879250
Cited by
93
References
89
Citations by year
2026: 52 · 2025: 37 · 2024: 3
FWCI
3.15
Type
journal article
Language
en
Open access
Licence
Locations
  • npj Digital Medicine (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Bern Open Repository and Information System (University of Bern) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • Open Access CRIS of the University of Bern
Funders
Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (205320_212939)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Explainable Artificial Intelligence (XAI)
  • Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
  • Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning in Healthcare
Keywords
explainable AI, multimodal data, longitudinal data, diagnostic imaging, clinical usability, clinical decision support

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 2026-10-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-09 (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: Orchestrating explainable artificial intelligence for multimodal and longitudinal data in medical imaging. https://app.cochat.ai/papers/10.1038/s41746-024-01190-w
The paper
Pahud de Mortanges, A., Luo, H., Shu, S. Z., Kamath, A., Suter, Y., Shelan, M., Pöllinger, A., & Reyes, M. (2024). Orchestrating explainable artificial intelligence for multimodal and longitudinal data in medical imaging. npj Digital Medicine. https://doi.org/10.1038/s41746-024-01190-w

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.