RAxSS: Retrieval-Augmented Sparse Sampling for Explainable Variable-Length Medical Time Series Classification

TL;DR

The paper tackles variable-length medical time series classification under data sparsity and noise by enhancing the Stochastic Sparse Sampling (SSS) framework with retrieval-informed weighting, producing explainable, probability-space aggregations. It introduces RAxSS, which computes within-series similarities to reweight window predictions, yielding a convex combination of window posteriors $\hat{p}^{(i)}=\sum_{k\in K_i} \alpha_k p_k$ with $\alpha_k$ defined via a temperature-controlled softmax over mean neighbor similarities $\bar{s}_k$. This approach provides an explicit evidence trail for each influential window through a ranked neighbor leaderboard, linking locality to overall predictions while preserving efficiency and backbone compatibility. Evaluated on multicenter iEEG SOZ localization data, RAxSS achieves competitive performance (e.g., AUC ~0.80, F1 ~0.73) relative to strong baselines and delivers enhanced transparency, enabling principled drill-down from series-level decisions to segment-level explanations. The framework is privacy-conscious and model-agnostic, with clear avenues for future extensions such as cross-subject retrieval and pattern-level retrieval to further strengthen evidence and applicability in clinical settings.

Abstract

Medical time series analysis is challenging due to data sparsity, noise, and highly variable recording lengths. Prior work has shown that stochastic sparse sampling effectively handles variable-length signals, while retrieval-augmented approaches improve explainability and robustness to noise and weak temporal correlations. In this study, we generalize the stochastic sparse sampling framework for retrieval-informed classification. Specifically, we weight window predictions by within-channel similarity and aggregate them in probability space, yielding convex series-level scores and an explicit evidence trail for explainability. Our method achieves competitive iEEG classification performance and provides practitioners with greater transparency and explainability. We evaluate our method in iEEG recordings collected in four medical centers, demonstrating its potential for reliable and explainable clinical variable-length time series classification.

Figures (1)

• Figure 1: RAxSS workflow: (a) end-to-end pipeline and (b) retrieval-weighted explainable module.