What is the clinical evidence from this study?

Study design: Other. Population: ECG multi-label classification (n=18885). Intervention: Neural network model compression techniques (quantization, structured pruning, knowledge distillation) vs. Uncompressed neural network models. Primary outcome: Multi-label classification AUROC (macro-averaged) (Absolute AUROC increase of 0.0022).

What does this research mean for the field?

Artificial neural network compression techniques improved model efficiency for ECG multi-label classification, enabling effective deployment on resource-constrained devices. Novelty: ClaimNovelty.INCREMENTAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

This research aims to evaluate various artificial neural network compression techniques for improving classification efficiency on wearable devices.

February 12, 2026Informatica0 citationsOpen Access

A Case Study of Artificial Neural Network Compression Methods for Resource-Constrained Multi-Label Classification

PHPrzemysław HołdaSystems Research Institute KWKatarzyna Wasielewska-MichniewskaSystems Research Institute MGMaria GanzhaWarsaw University of Technology

Key Result

Knowledge distillation improved RNN model AUROC from 0.9186 to 0.9208 at 61.45% parameter retention on ECG multi-label classification.

Key Points

This research aims to evaluate various artificial neural network compression techniques for improving classification efficiency on wearable devices.
Experimental evaluation of compression techniques on three neural network models
Utilized quantization, structured pruning, and knowledge distillation
Applied methods to the PTB-XL dataset for ECG signal classification
Compression methods enhanced model responsiveness while maintaining quality
Significant reduction in model size observed
Accelerated inference times indicated improved deployment feasibility on edge devices

Structured PICO

Do artificial neural network compression techniques maintain multi-label ECG classification performance while reducing model size and inference time compared to uncompressed models?

Population

18,885 patients contributing 21,837 12-lead ECG 10-second recordings (PTB-XL dataset) for multi-label classification (NORM, MI, CD, HYP, STTC).

Intervention

Artificial neural network compression techniques including quantization, structured pruning, and knowledge distillation applied to deep learning models (LSTM-based RNN, 1D CNN, 1D ResNet).

Comparator

Uncompressed (100% of parameters) floating-point baseline models.

Outcome

Macro-averaged AUROC (Area Under the Receiver Operating Characteristic curve) for multi-label classification.

Neural network compression techniques like pruning and knowledge distillation can significantly reduce the size of ECG classification models, sometimes even improving performance (as seen in RNNs), enabling deployment on resource-constrained edge devices.

Main Result

Effect estimate: Absolute AUROC increase of 0.0022

Absolute Event Rate: 0.9208% vs 0.9186%

Limitations

Study conducted on pre-existing dataset without clinical trial enrollment or control
Results based on computational experiments on models, not clinical outcomes
Inference time and model size measured on general-purpose hardware, not wearable ECG devices
No patient demographic details or diversity reported

Abstract

Proliferation of wearable healthcare devices has created the need to deliver artificial intelligence applications for these resource-constrained devices to achieve faster, localized decision-making, by bringing computation closer to the data sources, for improved responsiveness and privacy. This contribution presents the results of an experimental evaluation of artificial neural network compression techniques, including quantization, structured pruning, and knowledge distillation, applied to multi-label classification of electrocardiogram (ECG) signals. The experiments were carried out on the PTB-XL dataset using three deep learning models, i.e. an LSTM-based recurrent neural network, a 1D convolutional neural network, and a 1D residual neural network. The results show how the compression methods impact model quality and highlight opportunities to reduce model size and accelerate inference, thereby enabling effective deployment on resource-constrained, edge devices.

KI fragen

Bookmark

View Full Paper