A Performance Study of 1D CNN Variants for 12-Lead ECG Classification and Edge Inference on NVIDIA Jetson Orin Nano

DOI

10.2991/978-94-6239-828-3_27How to use a DOI?

Keywords

Electrocardiogram; 1D-CNN; Edge Computing; PTB-XL

Abstract

Most prior electrocardiogram (ECG) deep-learning studies focus on classification performance and are evaluated on server-class GPUs, leaving the accuracy-efficiency trade-off on edge devices under- characterized. This study compares four 1D Convolutional Neural Net- work (1D-CNN) variants—baseline 1D-CNN, 1D-ResNet, Multi-Scale CNN, and Attention-Based CNN—for diagnostic 12-lead ECG classification on the PTB-XL dataset at 100 Hz. Each complete 10-s, 12-lead recording is bandpass-filtered, Z-score normalized per channel, and fed directly as a (1000, 12) tensor to classify five diagnostic super- classes (NORM, MI, STTC, CD, HYP), without segmentation. Models are trained in TensorFlow/Keras and converted to TensorRT FP16 for batch inference on the NVIDIA Jetson Orin Nano. On the held-out fold- 10 test set, 1D-ResNet attains the highest accuracy (79.94%) and Macro AUC (0.9354), while the baseline 1D-CNN delivers the lowest latency (0.5291 ms per sample) and smallest parameter count. All four mod- els achieve sub-millisecond latency within the device’s 8 GB memory envelope. Model selection uses the Weighted Sum Model (WSM), with weights of 0.40 accuracy, 0.30 latency, 0.15 CPU, 0.10 GPU, and 0.05 RAM, verified via the Analytic Hierarchy Process (AHP). 1D-ResNet is identified as the optimal model under both schemes (WSM 0.600; AHP 0.641), offering the best balance between performance and efficiency for accuracy-priority edge deployment.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Najwa Muthia Aziz
AU  - Athiya Jannati Zumaro
AU  - Muhammad Nadhil Luthfirrahman
AU  - Kasyful Amron
AU  - Dany Primanita Kartikasari
AU  - Rizal Setya Perdana
AU  - Eko Sakti Pramukantoro
PY  - 2026
DA  - 2026/11/04
TI  - A Performance Study of 1D CNN Variants for 12-Lead ECG Classification and Edge Inference on NVIDIA Jetson Orin Nano
BT  - Proceedings of the 2026 11th International Conference on Sustainable Information Engineering and Technology (SIET 2026)
PB  - Atlantis Press
SP  - 305
EP  - 315
SN  - 2589-4900
UR  - https://doi.org/10.2991/978-94-6239-828-3_27
DO  - 10.2991/978-94-6239-828-3_27
ID  - Aziz2026
ER  -