Repurposing Foundation Model for Generalizable Medical Time Series Classification

2024-10-03

arXiv.org

AI Summary

Medical time-series classification suffers from severe generalization limitations due to strong heterogeneity—both cross-dataset (e.g., varying numbers of physiological channels, sequence lengths, and task definitions) and within-dataset (e.g., inter-patient variability). To address this, we propose FORMED, the first foundation-model reuse framework for medical time series. FORMED leverages a pre-trained temporal backbone and introduces three core techniques: (i) multi-source Medical Time-Series (MedTS) data co-distillation, (ii) channel-agnostic feature alignment, and (iii) task-agnostic representation disentanglement. These enable zero-shot adaptation across diverse channel counts, sequence lengths, and clinical tasks—without any task-specific training. Empirically, FORMED matches or surpasses 11 specialized baselines in zero-shot settings. With only lightweight fine-tuning, it rapidly adapts to entirely unseen datasets, significantly enhancing both generalizability and clinical scalability.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Computer Vision: Medical and Biological Imaging
• Multiagent Systems: Other Foundations of Multi Agent Systems

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Scalable techniques for the creation, curation, publication, maintenance, and consumption of large, Web-based, structured, reusable, knowledge graphs and ontologies

Abstract

Medical time series (MedTS) classification is critical for a wide range of healthcare applications such as Alzheimer's Disease diagnosis. However, its real-world deployment is severely challenged by poor generalizability due to inter- and intra-dataset heterogeneity in MedTS, including variations in channel configurations, time series lengths, and diagnostic tasks. Here, we propose FORMED, a foundation classification model that leverages a pre-trained backbone and tackles these challenges through re-purposing. FORMED integrates the general representation learning enabled by the backbone foundation model and the medical domain knowledge gained on a curated cohort of MedTS datasets. FORMED can adapt seamlessly to unseen MedTS datasets, regardless of the number of channels, sample lengths, or medical tasks. Experimental results show that, without any task-specific adaptation, the repurposed FORMED achieves performance that is competitive with, and often superior to, 11 baseline models trained specifically for each dataset. Furthermore, FORMED can effectively adapt to entirely new, unseen datasets, with lightweight parameter updates, consistently outperforming baselines. Our results highlight FORMED as a versatile and scalable model for a wide range of MedTS classification tasks, positioning it as a strong foundation model for future research in MedTS analysis.

Problem

• Addressing poor generalizability in medical time series classification due to dataset heterogeneity
• Repurposing foundation models for adaptable classification without full fine-tuning
• Enhancing performance across diverse clinical tasks and data configurations

Innovation

• Repurposing foundation model for medical time series
• Dynamic channel embeddings and label queries
• Lightweight adaptation with shared decoding layer