Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-scale Frequency-Aware Adversarial Network for Parkinson's Disease Assessment Using Wearable Sensors

Weiming Zhao, Xulong Wang, Jun Qi, Yun Yang, Po Yang

TL;DR

This paper tackles automated Parkinson's disease severity assessment from wearable sensor time series by introducing MFAM, a framework that combines a knowledge-driven Frequency Decomposition Module with a Multi-scale Channel Attention Encoder and an Attention-based MIL aggregator, all trained jointly with a Conditional Adversarial Domain Adaptation network. By isolating pathology-related frequency bands, MFAM improves feature specificity; its MIL component focuses on sparse, diagnostically valuable segments, enhancing sensitivity to transient PD signals. Validation on both a public PD vs. differential-diagnosis dataset and a private four-class severity dataset shows MFAM achieving superior or competitive performance against strong baselines, with interpretable attention maps that highlight key time windows. The approach holds promise for robust, objective PD assessment in real-world, long-term monitoring and can be extended to regression, multi-modal data fusion, and improved clinician interpretability.

Abstract

Severity assessment of Parkinson's disease (PD) using wearable sensors offers an effective, objective basis for clinical management. However, general-purpose time series models often lack pathological specificity in feature extraction, making it difficult to capture subtle signals highly correlated with PD.Furthermore, the temporal sparsity of PD symptoms causes key diagnostic features to be easily "diluted" by traditional aggregation methods, further complicating assessment. To address these issues, we propose the Multi-scale Frequency-Aware Adversarial Multi-Instance Network (MFAM). This model enhances feature specificity through a frequency decomposition module guided by medical prior knowledge. Furthermore, by introducing an attention-based multi-instance learning (MIL) framework, the model can adaptively focus on the most diagnostically valuable sparse segments.We comprehensively validated MFAM on both the public PADS dataset for PD versus differential diagnosis (DD) binary classification and a private dataset for four-class severity assessment. Experimental results demonstrate that MFAM outperforms general-purpose time series models in handling complex clinical time series with specificity, providing a promising solution for automated assessment of PD severity.

Multi-scale Frequency-Aware Adversarial Network for Parkinson's Disease Assessment Using Wearable Sensors

TL;DR

This paper tackles automated Parkinson's disease severity assessment from wearable sensor time series by introducing MFAM, a framework that combines a knowledge-driven Frequency Decomposition Module with a Multi-scale Channel Attention Encoder and an Attention-based MIL aggregator, all trained jointly with a Conditional Adversarial Domain Adaptation network. By isolating pathology-related frequency bands, MFAM improves feature specificity; its MIL component focuses on sparse, diagnostically valuable segments, enhancing sensitivity to transient PD signals. Validation on both a public PD vs. differential-diagnosis dataset and a private four-class severity dataset shows MFAM achieving superior or competitive performance against strong baselines, with interpretable attention maps that highlight key time windows. The approach holds promise for robust, objective PD assessment in real-world, long-term monitoring and can be extended to regression, multi-modal data fusion, and improved clinician interpretability.

Abstract

Severity assessment of Parkinson's disease (PD) using wearable sensors offers an effective, objective basis for clinical management. However, general-purpose time series models often lack pathological specificity in feature extraction, making it difficult to capture subtle signals highly correlated with PD.Furthermore, the temporal sparsity of PD symptoms causes key diagnostic features to be easily "diluted" by traditional aggregation methods, further complicating assessment. To address these issues, we propose the Multi-scale Frequency-Aware Adversarial Multi-Instance Network (MFAM). This model enhances feature specificity through a frequency decomposition module guided by medical prior knowledge. Furthermore, by introducing an attention-based multi-instance learning (MIL) framework, the model can adaptively focus on the most diagnostically valuable sparse segments.We comprehensively validated MFAM on both the public PADS dataset for PD versus differential diagnosis (DD) binary classification and a private dataset for four-class severity assessment. Experimental results demonstrate that MFAM outperforms general-purpose time series models in handling complex clinical time series with specificity, providing a promising solution for automated assessment of PD severity.

Paper Structure

This paper contains 17 sections, 2 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deep Learning Models for Time Series Classification
  4. Multi-scale and Frequency Analysis of Time Series
  5. Multiple Instance Learning and its Application in the Medical Field
  6. Methodology
  7. Problem Definition
  8. Frequency Decomposition Module
  9. Multi-scale Channel Attention Encoder (MS-CAE)
  10. Attention-based Multi-Instance Learning Aggregator
  11. Joint Training and Conditional Adversarial Domain Adaptation
  12. Experiments
  13. Datasets and Evaluation Protocol
  14. Implementation Settings
  15. Results and Analysis
  16. ...and 2 more sections

Figures (3)

  • Figure 1: Comparison between general-purpose time series models and the proposed MFAM method. (a) General-purpose time series models rely on generic feature extraction, while MFAM integrates (b) domain-specific prior knowledge with (c) advanced machine learning paradigms to address pathological feature specificity and temporal symptom sparsity.
  • Figure 2: Overall architecture of the MFAM model.
  • Figure 3: Attention weight distribution comparison for two samples. The upper panel shows a PD patient (label 1), and the lower panel shows a DD patient (label 0).