Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TransConv-DDPM: Enhanced Diffusion Model for Generating Time-Series Data in Healthcare

Md Shahriar Kabir, Sana Alamgeer, Minakshi Debnath, Anne H. H. Ngu

TL;DR

TransConv-DDPM addresses data scarcity in healthcare time-series by enhancing diffusion-based generation with a transformer bottleneck and multi-scale convolutions, enabling robust modeling of local and global temporal dependencies. The method outperforms TimeGAN and Diffusion-TS across stick-balancing, SmartFallMM, and EEG datasets, exhibiting high distribution fidelity and predictive utility. A notable finding is the significant performance boost in fall-detection tasks when synthetic data generated by TransConv-DDPM is combined with real data, underscored by substantial gains in F1-score and accuracy. These results support the practical value of high-fidelity synthetic physiological time-series for training robust clinical AI systems and point to avenues for broader, conditional, and multi-modal data generation in healthcare.

Abstract

The lack of real-world data in clinical fields poses a major obstacle in training effective AI models for diagnostic and preventive tools in medicine. Generative AI has shown promise in increasing data volume and enhancing model training, particularly in computer vision and natural language processing (NLP) domains. However, generating physiological time-series data, a common type in medical AI applications, presents unique challenges due to its inherent complexity and variability. This paper introduces TransConv-DDPM, an enhanced generative AI method for biomechanical and physiological time-series data generation. The model employs a denoising diffusion probabilistic model (DDPM) with U-Net, multi-scale convolution modules, and a transformer layer to capture both global and local temporal dependencies. We evaluated TransConv-DDPM on three diverse datasets, generating both long and short-sequence time-series data. Quantitative comparisons against state-of-the-art methods, TimeGAN and Diffusion-TS, using four performance metrics, demonstrated promising results, particularly on the SmartFallMM and EEG datasets, where it effectively captured the more gradual temporal change patterns between data points. Additionally, a utility test on the SmartFallMM dataset revealed that adding synthetic fall data generated by TransConv-DDPM improved predictive model performance, showing a 13.64% improvement in F1-score and a 14.93% increase in overall accuracy compared to the baseline model trained solely on fall data from the SmartFallMM dataset. These findings highlight the potential of TransConv-DDPM to generate high-quality synthetic data for real-world applications.

TransConv-DDPM: Enhanced Diffusion Model for Generating Time-Series Data in Healthcare

TL;DR

TransConv-DDPM addresses data scarcity in healthcare time-series by enhancing diffusion-based generation with a transformer bottleneck and multi-scale convolutions, enabling robust modeling of local and global temporal dependencies. The method outperforms TimeGAN and Diffusion-TS across stick-balancing, SmartFallMM, and EEG datasets, exhibiting high distribution fidelity and predictive utility. A notable finding is the significant performance boost in fall-detection tasks when synthetic data generated by TransConv-DDPM is combined with real data, underscored by substantial gains in F1-score and accuracy. These results support the practical value of high-fidelity synthetic physiological time-series for training robust clinical AI systems and point to avenues for broader, conditional, and multi-modal data generation in healthcare.

Abstract

The lack of real-world data in clinical fields poses a major obstacle in training effective AI models for diagnostic and preventive tools in medicine. Generative AI has shown promise in increasing data volume and enhancing model training, particularly in computer vision and natural language processing (NLP) domains. However, generating physiological time-series data, a common type in medical AI applications, presents unique challenges due to its inherent complexity and variability. This paper introduces TransConv-DDPM, an enhanced generative AI method for biomechanical and physiological time-series data generation. The model employs a denoising diffusion probabilistic model (DDPM) with U-Net, multi-scale convolution modules, and a transformer layer to capture both global and local temporal dependencies. We evaluated TransConv-DDPM on three diverse datasets, generating both long and short-sequence time-series data. Quantitative comparisons against state-of-the-art methods, TimeGAN and Diffusion-TS, using four performance metrics, demonstrated promising results, particularly on the SmartFallMM and EEG datasets, where it effectively captured the more gradual temporal change patterns between data points. Additionally, a utility test on the SmartFallMM dataset revealed that adding synthetic fall data generated by TransConv-DDPM improved predictive model performance, showing a 13.64% improvement in F1-score and a 14.93% increase in overall accuracy compared to the baseline model trained solely on fall data from the SmartFallMM dataset. These findings highlight the potential of TransConv-DDPM to generate high-quality synthetic data for real-world applications.
Paper Structure (19 sections, 8 equations, 2 figures, 3 tables)

This paper contains 19 sections, 8 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Non-Generative AI-based Techniques
  4. Generative AI-based Techniques
  5. Methodology
  6. Forward Diffusion (Training)
  7. Reverse Denoising (Generation)
  8. Multi-Scale Convolution for Local Dependencies
  9. Transformer Layer for Global Dependencies
  10. Experimental Setup
  11. Datasets
  12. Data Preparation
  13. Training and Testing Parameters
  14. Results
  15. Qualitative Analysis: Distribution Comparison
  16. ...and 4 more sections

Figures (2)

  • Figure 1: Illustration of the proposed model architecture. (a) Training: The diffusion process adds noise to the input accelerometer data for feature extraction. (b) Generation: The denoising process generates synthetic accelerometer data using a Transformer-enhanced model.
  • Figure 2: Distribution comparison for the SmartFallMM, EEG, and Stick-balance datasets with synthetic data generated from (a) TransConv-DDPM (our), (b) TimeGAN, and (c) Diffusion-TS methods.