Multi-modal Framework for Fetal Heart Rate Estimation: Fusion of Low-SNR ECG and Inertial Sensors.

Abstract

This study presents a novel multi-modal framework for fetal heart rate extraction, which incorporates wearable seismo-cardiography (SCG), gyro-cardiography (GCG), and electrocardiography (ECG) readings from ten pregnant women. Firstly, a signal refinement method based on empirical mode decomposition (EMD) is proposed to extract the desired signal components associated with fetal heart rate (FHR). Afterwards, two techniques are developed to fuse the information from different modalities. The first method, named early fusion, is intended to combine the refined signals of different modalities through intra-modality fusion, intermodality fusion, and FHR estimation. The other fusion approach, i.e., late fusion, includes FHR estimation and intermodality FHR fusion. FHR values are estimated and compared with readings from a simultaneously-recorded cardiotocography (CTG) sensor. It is demonstrated that the best performance belongs to the late-fusion approach with 87.00% of positive percent agreement (PPA), 6.30% of absolute percent error (APE), and 10.55 beats-per-minute (BPM) of root-meansquare-error (RMSE).Clinical Relevance- The proposed framework allows for the continuous monitoring of the health status of the fetus in expectant women. The approach is accurate and cost-effective due to the use of advanced signal processing techniques and lowcost wearable sensors, respectively.