GADM: Data augmentation using Generative Adversarial Diffusion Model for pulse-based disease identification

doi:10.1016/j.bspc.2024.107005

GADM: Data augmentation using Generative Adversarial Diffusion Model for pulse-based disease identification

2025-02-01

0

PRE

AI

DOI：10.1016/j.bspc.2024.107005

Original

Original request for help

Save

Abstract

En 中文

Pulse is an important health indicator that includes physiological parameters like heart rate, rhythm, and vascular elasticity. While deep learning methods for pulse diagnosis have been widely used, they are often constrained by limited data, especially for rare diseases. Data augmentation using diffusion models can address these issues, but their sampling process consumes a lot of time and is difficult to apply in practice. To address these challenges, we introduce a novel generative adversarial diffusion model (GADM) that significantly enhances data sampling efficiency and achieves superior results in data augmentation. Our model employs a timestep-dependent discriminator to model each stage of the diffusion process. By utilizing the adversarial loss of generative adversarial networks (GANs), GADM achieves tighter lower bounds on the log-likelihood compared to the original diffusion model, resulting in an implicit model that can generate high-quality samples faster. Experimental results demonstrate that the proposed GADM obtains better sample quality and is also 10 times faster compared to the denoised diffusion probability model (DDPM). Additionally, compared to the accuracy of 88.28% obtained from training on the original imbalanced dataset, training on the class-balanced dataset enhanced using GADM achieved an accuracy of 94.33% in the hypertension classification task.

Keywords:

Pulse

Data augmentation

Diffusion model

Generative adversarial network

Medical data generation

GADM: Data augmentation using Generative Adversarial Diffusion Model for pulse-based disease identification

Abstract

Journal

Organization