Lightweight model reconstructs heart signals from wearable sensors
TinyCardioUNet: IMU-to-ECG Translation with Graph-Encoded Inter-Axis Dependencies and Tensor Decomposition-Based Parameter Reduction
Machine Learning
Summary
Measuring heart signals usually needs sticky electrodes that can be uncomfortable. This paper shows how to get similar heart readings from a small device worn on the chest that measures motion and orientation instead. The authors designed a compact neural network that cleverly uses all sensor directions and cuts down on model size without losing accuracy. This could make heart monitoring easier and less annoying for users.
What this means in practice
- •For wearable device developers: Build small, accurate models that translate chest motion sensor data into ECG signals to monitor heart rate without electrodes.$Commercial implications: Enables companies to sell comfortable, electrode-free heart monitoring wearables using compact models that reduce hardware demands.
- •For health monitoring service providers: Use graph-encoded dependencies and tensor-based compression to improve continuous heart rate tracking accuracy from IMU data in real-world settings.
Authors
Seungwoo Han, Ingon Chanpornpakdi, Motoi Noda, Puwadej Leelasiri, Ibuki Hiruma, Toshihisa Tanaka
Abstract
Estimating electrocardiography (ECG) from a chest-worn inertial measurement unit (IMU) enables continuous heart rate (HR) monitoring without the discomfort of electrodes. We propose TinyCardioUNet, a lightweight UNet that uses all six IMU axes without prior channel selection, refines its bottleneck with a graph neural network that encodes inter-axis dependencies, and employs tensor decomposition with automatic variational Bayesian rank selection for parameter reduction. On a public dataset, TinyCardioUNet achieves an RMSE of $0.098$ and a Pearson correlation coefficient of $0.677$ with only $36.0$k parameters and remains comparatively robust to additive noise, demonstrating accurate ECG reconstruction with a compact model.