Paperdoi:10.3390/s26227229

TinyML-Based Photoplethysmography (PPG) Analysis for Atrial Fibrillation Detection

Venue listedDated Nov 16, 2026Added Nov 17, 2026

  • KC

Authors: Kyoungtaek Choi

Sensors (MDPI), published 2026-11-16

TL;DR

Compares CNN, autoencoder and Transformer PPG AF classifiers on MIMIC PERform AF and ports the Transformer to an nRF52840. Post-training quantization dropped average accuracy from 98.66% to 95.23%; quantization-aware training avoided the drop. Unsupported TFLite Micro layers were replaced and multi-head attention approximated with a CNN block, giving a 107 KB model at 98.09% accuracy on device.

  • Form factor: fingertip ICU PPG for training (MIMIC PERform AF); edge target nRF52840

Abstract

This paper proposes an atrial fibrillation detection algorithm using PPG signals and presents a method to port it to edge devices such as the nRF52840. To classify the PPG signals, simple CNN-based, Autoencoder-based, and Transformer-based Deep Neural Networks (DNNs) were implemented and comparatively analyzed using the MIMIC PERform AF dataset. As a result, the Transformer-based algorithm exhibited the best performance with an average accuracy of 98.66%. While post-training quantization (PTQ) of the Transformer-based algorithm caused the average accuracy to degrade to 95.23%, applying Quantization-Aware Training (QAT) prevented this performance drop. Because some layers of the Transformer are not supported in TensorFlow Lite for Microcontrollers (TFLite Micro), they were modified, and the Multi-Head Attention was approximated using a CNN-based network for porting to the edge device, resulting in a final model size of 107 KB. The ported network demonstrated excellent performance, maintaining an average accuracy of 98.09%.

Verbatim from the publisher's DOI record (opens in a new tab) · Publisher licence: https://creativecommons.org/licenses/by/4.0/

Results on ChatPPG leaderboards

All leaderboards

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.

Tasks & signals

Tasks

AF / rhythm

Signals

Keywords

AF DetectionTinyMLTFLite MicronRF52840Cortex M4fTransformerQuantization Aware TrainingPTQ

Datasets used

How it works

How Method works: Setup → Method → TargetSetup: Signals (PPG); Form factor (fingertip ICU PPG for training (MIMIC PERform AF); edge target nRF52840). Method: Method (Transformer). Target: AF / rhythm.SetupSignals: PPGSignalsPPGForm factor: fingertip ICU PPG for training (MIMIC PERform AF); edge target nRF52840Form factorfingertip ICU PPG for training (MIMIC…Method×NMethodTransformerTargetAF / rhythmAF / rhythm

Cite this paper

Cite this paper with the fields recorded here. Check it against the publisher before submitting.

Choi, K. (2026). TinyML-Based Photoplethysmography (PPG) Analysis for Atrial Fibrillation Detection. Sensors. https://doi.org/10.3390/s26227229
Choi, Kyoungtaek. 2026. "TinyML-Based Photoplethysmography (PPG) Analysis for Atrial Fibrillation Detection." Sensors. https://doi.org/10.3390/s26227229.
Choi, K. (2026) 'TinyML-Based Photoplethysmography (PPG) Analysis for Atrial Fibrillation Detection', Sensors. doi: 10.3390/s26227229.
Choi, Kyoungtaek "TinyML-Based Photoplethysmography (PPG) Analysis for Atrial Fibrillation Detection." Sensors, 2026. https://doi.org/10.3390/s26227229.
K. Choi, "TinyML-Based Photoplethysmography (PPG) Analysis for Atrial Fibrillation Detection," Sensors, 2026, doi: 10.3390/s26227229.
@article{choi2026tinymlbased,
  title         = {{TinyML-Based Photoplethysmography (PPG) Analysis for Atrial Fibrillation Detection}},
  author        = {Choi, Kyoungtaek},
  year          = {2026},
  journal       = {Sensors},
  publisher     = {MDPI},
  doi           = {10.3390/s26227229}
}

Added to the ChatPPG research hub