ImputeINR: Time Series Imputation via Implicit Neural Representations for Disease Diagnosis with Missing Data

2025-05-16

AI Summary

Medical time-series data frequently exhibit extremely high missingness rates (>90%), rendering conventional discrete-point imputation methods inadequate for modeling sparse observations and consequently limiting downstream diagnostic performance. To address this, we introduce implicit neural representations (INRs) to medical time-series imputation for the first time, constructing a continuous-time neural field that enables fine-grained, frequency-agnostic reconstruction of missing values. We propose an end-to-end differentiable framework coupled with a multi-scale masking training strategy, significantly enhancing robustness under extreme sparsity. Extensive evaluation across eight real-world clinical datasets and five missingness levels demonstrates consistent superiority over state-of-the-art methods: at >90% missingness, imputation error decreases by 23.6%, while downstream disease diagnosis accuracy improves by an average of 7.4%.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Knowledge Representation and Reasoning: Diagnosis and Abductive Reasoning
• Machine Learning: Time-Series/Data Streams

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Healthcare data frequently contain a substantial proportion of missing values, necessitating effective time series imputation to support downstream disease diagnosis tasks. However, existing imputation methods focus on discrete data points and are unable to effectively model sparse data, resulting in particularly poor performance for imputing substantial missing values. In this paper, we propose a novel approach, ImputeINR, for time series imputation by employing implicit neural representations (INR) to learn continuous functions for time series. ImputeINR leverages the merits of INR in that the continuous functions are not coupled to sampling frequency and have infinite sampling frequency, allowing ImputeINR to generate fine-grained imputations even on extremely sparse observed values. Extensive experiments conducted on eight datasets with five ratios of masked values show the superior imputation performance of ImputeINR, especially for high missing ratios in time series data. Furthermore, we validate that applying ImputeINR to impute missing values in healthcare data enhances the performance of downstream disease diagnosis tasks. Codes are available.

Problem

• Healthcare data often has missing values affecting disease diagnosis
• Existing methods fail to model sparse time series effectively
• ImputeINR uses neural representations for continuous, fine-grained imputation

Innovation

• Uses implicit neural representations for imputation
• Learns continuous functions for time series
• Enables fine-grained imputation on sparse data