Group-Sparse Manifold-Aware Integrated Gradients for Multimodal Transformers on EHR Trajectories

Ali Amirahmadi, Farzaneh Etminani, Mattias Ohlsson

Keywords: Integrated Gradients, Explainability, Multimodal Transformers, Group Sparsity, Manifold-aware, Electronic Health Records (EHR), Patient trajectories

TL;DR: We introduce a manifold-aware baseline and a group-sparse IG schedule that make explanations for multimodal EHR transformers faithful, sparse, concise, and practical.

Track: Proceedings

Abstract: Integrated Gradients (IG) is a popular method for explaining clinical deep models—including widely used multimodal, pretrained Transformers—but its utility on EHR code sequences is hampered by (i) the lack of principled baselines for sequence of discrete tokens and (ii) dense, hard-to-interpret generated attributions. To address both, first, we introduce a manifold-aware baseline: the expected value under the empirical dist—implemented as the position-wise empirical mean of pre-Transformer token embeddings on held-out validation data, which keeps IG interpolants near the data manifold. Second, we introduce {GS-IG}, which preserves the straight path geometry but re-parameterizes the schedule \(\alpha(t)=t^{\theta}\) and selects \(\theta\) per input by minimizing a token-level \(\ell_{2,1}\) (group-sparsity) objective, producing concise, practitioner-friendly explanations. On MIMIC-IV (incident heart failure) and MDC (early mortality), the manifold-aware baseline improves faithfulness (higher Comprehensiveness, lower Sufficiency), and GS-IG reduces token-level \(\ell_{2,1}\) by 9–18\% with negligible change in those metrics on the manifold-aware baseline. The method is lightweight and yields faithful, sparse, and actionable.

General Area: Models and Methods

Specific Subject Areas: Explainability & Interpretability, Time Series

Data And Code Availability: Yes

Ethics Board Approval: Yes

Entered Conflicts: I confirm the above

Anonymity: I confirm the above

Submission Number: 74