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

2025-10-03

AI Summary

Medical time-series classification faces challenges including data sparsity, noise contamination, and variable-length inputs, while existing methods suffer from limited robustness and interpretability. This paper proposes a retrieval-augmented sparse sampling framework: it performs adaptive feature selection via intra-channel similarity-weighted sampling, and generates explicit evidence paths through probabilistic-space aggregation and convex sequence scoring—enhancing robustness against weak temporal dependencies and noise while ensuring decision transparency. The method requires no fixed input length and supports end-to-end interpretable prediction. Evaluated on intracranial electroencephalography (iEEG) data from four clinical centers, it achieves state-of-the-art classification performance using significantly fewer labeled samples. Our approach establishes a new paradigm for clinical time-series analysis that jointly delivers high reliability and strong interpretability.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Computer Vision: Interpretability, Explainability, and Transparency

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

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.

Problem

• Classifying variable-length medical time series data
• Addressing data sparsity and noise in medical signals
• Improving explainability for clinical time series analysis

Innovation

• Retrieval-augmented sparse sampling for medical classification
• Weighting window predictions by within-channel similarity
• Aggregating predictions in probability space for explainability