INN-PAR: Invertible Neural Network for PPG to ABP Reconstruction

TL;DR

The paper addresses non-invasive arterial BP reconstruction from PPG by tackling information loss in prior PAR models. It introduces INN-PAR, an invertible neural network that jointly learns bijective mappings between ($X$, $\nabla X$) and ($Y$, $\nabla Y$), enabling simultaneous forward and inverse reconstruction to preserve information. A Multi-Scale Convolution Module (MSCM) within invertible blocks and an invertible $1\times1$ convolution enable multi-scale feature extraction and channel mixing, while a gradient-focused loss aligns both ABP values and their gradients with ground truth. Experiments on two public datasets show INN-PAR outperforms state-of-the-art methods in ABP waveform reconstruction and SBP/DBP accuracy with lower computational cost, suggesting strong potential for continuous, non-invasive BP monitoring and applicability to other physiological signal reconstruction tasks.

Abstract

Non-invasive and continuous blood pressure (BP) monitoring is essential for the early prevention of many cardiovascular diseases. Estimating arterial blood pressure (ABP) from photoplethysmography (PPG) has emerged as a promising solution. However, existing deep learning approaches for PPG-to-ABP reconstruction (PAR) encounter certain information loss, impacting the precision of the reconstructed signal. To overcome this limitation, we introduce an invertible neural network for PPG to ABP reconstruction (INN-PAR), which employs a series of invertible blocks to jointly learn the mapping between PPG and its gradient with the ABP signal and its gradient. INN-PAR efficiently captures both forward and inverse mappings simultaneously, thereby preventing information loss. By integrating signal gradients into the learning process, INN-PAR enhances the network's ability to capture essential high-frequency details, leading to more accurate signal reconstruction. Moreover, we propose a multi-scale convolution module (MSCM) within the invertible block, enabling the model to learn features across multiple scales effectively. We have experimented on two benchmark datasets, which show that INN-PAR significantly outperforms the state-of-the-art methods in both waveform reconstruction and BP measurement accuracy. Codes can be found at: https://github.com/soumitra1992/INNPAR-PPG2ABP.

Figures (3)

• Figure 1: INN-PAR architecture and Invertible Block (IB) structure.
• Figure 2: Multi-Scale Convolution Module (MSCM).
• Figure 3: Visual comparison with SOTA methods. The first row shows a signal from the Sensors dataset sensors, and the second row shows a signal from the BCG dataset bcg. Signals are best viewed in $200\%$ zoom.