DCentNet: Decentralized Multistage Biomedical Signal Classification using Early Exits

TL;DR

This work tackles the energy, latency, and bandwidth challenges of centralized biomedical signal inference for wearable IoT by proposing DCenNet, a decentralized multistage CNN architecture with early exits and inter-node encoder–decoder compression. By partitioning a large CNN into multiple sub-networks and distributing them across edge, fog, and cloud nodes, DCenNet minimizes wireless data transmission and processing load while preserving high accuracy; a genetic algorithm optimizes the placement of early exits to balance accuracy, sensitivity, and FLOPs. Evaluations on MIT-BIH ECG data show that a single EEP can reduce data transmission by up to 94.54% and complexity by 21% without sacrificing performance, while two EEPs achieve sensitivity 98.36% and accuracy 97.74% with further reductions in data transfer and FLOPs; embedded hardware experiments on a Nano 33 BLE Sense and PPK2 validate substantial real-world energy savings, averaging 73.6% compared to continuous wireless transmission. The approach offers a scalable, robust, and energy-efficient framework for IoMT, enabling responsive monitoring and potential adaptation to other biosignals such as EEG or EMG.

Abstract

DCentNet is a novel decentralized multistage signal classification approach designed for biomedical data from IoT wearable sensors, integrating early exit points (EEP) to enhance energy efficiency and processing speed. Unlike traditional centralized processing methods, which result in high energy consumption and latency, DCentNet partitions a single CNN model into multiple sub-networks using EEPs. By introducing encoder-decoder pairs at EEPs, the system compresses large feature maps before transmission, significantly reducing wireless data transfer and power usage. If an input is confidently classified at an EEP, processing stops early, optimizing efficiency. Initial sub-networks can be deployed on fog or edge devices to further minimize energy consumption. A genetic algorithm is used to optimize EEP placement, balancing performance and complexity. Experimental results on ECG classification show that with one EEP, DCentNet reduces wireless data transmission by 94.54% and complexity by 21%, while maintaining original accuracy and sensitivity. With two EEPs, sensitivity reaches 98.36%, accuracy 97.74%, wireless data transmission decreases by 91.86%, and complexity is reduced by 22%. Implemented on an ARM Cortex-M4 MCU, DCentNet achieves an average power saving of 73.6% compared to continuous wireless ECG transmission.

Figures (9)

• Figure 1: Illustration of the decentralized multistage biomedical inferencing using Early Exits.
• Figure 2: Huge CNN architecture for biomedical signal inferencing.
• Figure 3: The architecture of the Edge-Cloud continuum for decentralized multistage biomedical signal inferencing using single EEP.
• Figure 4: The architecture of the Edge-Fog-Cloud continuum for decentralized multistage biomedical signal inferencing using two EEPs.
• Figure 5: Hardware implementation: PPK2 for measuring power consumption, Nano 33 (left) running the model, and another Nano 33 (right) serving as a receiver for the BLE signal.