Publication:

Improved interpretability and feature attribution for hybrid-AI ECG models by controlling metadata leakage and feature interactions

 
cris.virtual.department#PLACEHOLDER_PARENT_METADATA_VALUE#
cris.virtual.department#PLACEHOLDER_PARENT_METADATA_VALUE#
cris.virtual.department#PLACEHOLDER_PARENT_METADATA_VALUE#
cris.virtual.department#PLACEHOLDER_PARENT_METADATA_VALUE#
cris.virtual.orcid0000-0001-8223-3454
cris.virtual.orcid0000-0001-6600-1792
cris.virtual.orcid0000-0003-2899-4636
cris.virtual.orcid0000-0003-3513-0174
cris.virtualsource.department76ef155c-d882-4fa9-9c29-e622de615572
cris.virtualsource.department3d0467d5-8f2f-463b-9a89-cd2e89911f08
cris.virtualsource.departmente8043942-f5dc-4e9f-b5ef-85780b08f47a
cris.virtualsource.department2c235e20-6a18-447c-84db-1c57bafdd89c
cris.virtualsource.orcid76ef155c-d882-4fa9-9c29-e622de615572
cris.virtualsource.orcid3d0467d5-8f2f-463b-9a89-cd2e89911f08
cris.virtualsource.orcide8043942-f5dc-4e9f-b5ef-85780b08f47a
cris.virtualsource.orcid2c235e20-6a18-447c-84db-1c57bafdd89c
dc.contributor.authorBontinck, Lennert
dc.contributor.authorFonteyn, Karel
dc.contributor.authorRibeiro, Antonio H.
dc.contributor.authorDhaene, Tom
dc.contributor.authorDeschrijver, Dirk
dc.date.accessioned2026-11-03T08:57:17Z
dc.date.available2026-11-03T08:57:17Z
dc.date.createdwos2026
dc.date.issued2026
dc.description.abstractDeep learning has shown strong potential for predicting atrial fibrillation (AFib) risk from electrocardiograms (ECGs). However, in hybrid-AI settings that combine an ECG latent representation with patient metadata (age and sex in this study), it can remain unclear to what extent predictions are driven by ECG latent features or by metadata. This ambiguity arises because the ECG latent representation may already encode metadata, causing attribution quantification methods to underestimate the attribution of explicitly provided metadata. A unified framework is proposed to improve interpretability by controlling metadata leakage, modelling ECG-metadata interactions explicitly, and enabling modality-aware attribution quantification. First, a Gradient-Reversed encoder (GRe) learns an ECG latent representation that is invariant to prespecified metadata, reducing metadata leakage. Second, a Multiplicative Feature-Interaction Dense (MFID) layer reintroduces metadata through gated multiplicative interactions with the ECG latent representation. Third, Feature-Interaction Gradient Interpretation (FIGI) combines ECG-level saliency visualisation using Grad-CAM++ with integrated gradients on a latent proxy model to quantify, per recording, the relative attribution of the ECG latent representation versus metadata in the final prediction. Across in-distribution and out-of-distribution evaluation, the proposed framework yields more clinically plausible attribution patterns and demonstrates that metadata can strongly influence AFib prediction, an effect that may be obscured by naive hybridisation. In the evaluated experiments, the proposed framework improved out-of-distribution (OOD) AFib prediction performance, although the magnitude of the benefit was task- and dataset-dependent.
dc.description.wosFundingTextThis work was conducted as part of the VASCULAI project, supported by the European Union through the Interreg France-Wallonie-Vlaanderen (FWVL) program (ref. 0100224) and the province Oost-Vlaanderen. The research was further supported by the Flemish Government via the AI Research Program.
dc.identifier.doi10.1016/j.eswa.2026.134714
dc.identifier.issn0957-4174
dc.identifier.urihttps://imec-publications.be/handle/20.500.12860/60703
dc.language.isoeng
dc.provenance.editstepusergreet.vanhoof@imec.be
dc.publisherPERGAMON-ELSEVIER SCIENCE LTD
dc.source.beginpage134714
dc.source.journalEXPERT SYSTEMS WITH APPLICATIONS
dc.source.numberofpages13
dc.source.volume334
dc.title

Improved interpretability and feature attribution for hybrid-AI ECG models by controlling metadata leakage and feature interactions

dc.typeJournal article
dspace.entity.typePublication
imec.internal.crawledAt2026-11-02
imec.internal.sourcecrawler
imec.internal.wosCreatedAt2026-11-02
Files
Publication available in collections: