Noise-tolerant modular neural network system for classifying ECG signal

Q: What is the clinical evidence from this study?

Population: Cardiac abnormalities (myocardial infarction, dysrhythmia, myocardial hypertrophy, right atrial enlargement, junctional rhythm, cardiac ischemia) (n=221). Intervention: Noise-tolerant modular neural network (NN) classifier trained with artificial Gaussian white-noise injection (1-12%) vs. True clinical diagnosis / reference ECG annotations (and literature comparison against prior filtering-based NN approaches). Primary outcome: Overall classification accuracy for discriminating normal heartbeat vs. cardiac abnormality on single-lead (V1) raw real ECG signals.

Q: What does this research mean for the field?

A modular neural network trained with artificial noise injection accurately classifies raw single-lead ECG signals into normal and abnormal patterns with 95.7% overall accuracy and 99.5% precision without relying on adaptive noise-filtering preprocessing. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Alberto Ochoa-Brust; Luis J. Mena; Vanessa G. Félix; A. Ojeda González; Walter Mata; Gladys E. Maestre

doi:10.31449/inf.v43i1.1605

March 1, 2019InformaticaOpen Access

Noise-tolerant modular neural network system for classifying ECG signal

View Full Paper

Ask AI

Bookmark

Share

Key result

Noise-tolerant modular neural network classifies raw single-lead ECGs with ~96% overall accuracy without adaptive filtering.

n=221

Why the study?

Current strategies for preprocessing and noise removal in automated neural network-based ECG signal classification remain unsatisfactory.

Does a modular neural network system trained with artificial noise injection accurately classify normal and abnormal heartbeat patterns in noisy simulated and raw single-lead real ECG signals without preprocessing filters?

Key Points

To develop a modular neural network system trained with artificial noise injection to accurately classify cardiac abnormalities in noisy and raw electrocardiogram signals.
Engineered a modular neural network system composed of four three-layered feedforward micro neural networks corresponding to the P, Q, ST, and T segments of an electrocardiogram.
Trained networks using simulated electrocardiogram signals injected with 1% to 12% Gaussian white noise across minor, moderate, severe, and extreme noise tiers.
Evaluated performance on a corrupted testing set of 21,600 simulated segments and a real-world testing set of 884 lead V1 electrocardiogram traces from 221 subjects in the PTB Diagnostic ECG Database.
Achieved average classification accuracies on simulated noisy electrocardiogram signals of 99.2% for minor, 95.1% for moderate, 91.4% for severe, and 85.2% for extreme noise, with an overall accuracy of 93.9%.
Discriminated normal from abnormal heartbeat patterns on single-lead raw electrocardiogram signals from the PTB dataset with 95.7% overall accuracy and 99.5% precision.

Structured PICO

Does a modular neural network system trained with artificial noise injection accurately classify normal and abnormal heartbeat patterns in noisy simulated and raw single-lead real ECG signals without preprocessing filters?

P

Population

221 subjects (884 single-lead V1 ECG traces; 52 normal heartbeat, 148 myocardial infarction, 14 dysrhythmia, and 7 myocardial hypertrophy) from the German PTB Diagnostic ECG Database, plus 21,600 simulated contaminated ECG segments.

I

Intervention

Modular neural network (NN) system consisting of four 3-layered feedforward micro NNs (one each for P, Q, ST, and T ECG intervals) trained with artificial Gaussian white-noise injection (1% to 12% noise levels: minor 1-2%, moderate 4-6%, severe 8-10%, extreme 12%) without adaptive signal filtering

C

Comparator

True clinical classification / reference annotations in the PTB database and simulated datasets (also benchmarked against published NN approaches using wavelet transform, FFT, cubic spline interpolation, or band-pass filtering)

O

Outcome

Classification performance (overall classification accuracy, precision, sensitivity, and specificity) for discriminating normal heartbeat from cardiac abnormality in raw single-lead (V1) real ECG signals and contaminated simulated ECG segmentssurrogate

Limitations

Tested for the detection of only four types of cardiac pathologies in the simulated dataset
Used artificially generated random Gaussian white noise without specific assumptions on the origin of the noise
Additive noise model may differ from real ECG records corrupted by physiological noise and exhibiting spatial correlation across individual ECG signals
Requires further verification with information about actual noise sources and other specific types of cardiac disorders
The system was tested for the detection of only four types of cardiac pathologies
Training used artificially generated random Gaussian white noise without any specific assumption on the origin of the noise
The additive noise model can differ from real ECG records corrupted by physiological noise and exhibiting spatial correlation across individual ECG signals
Requires further verification using actual ECG data with known noise sources and other specific types of cardiac disorders

Implication

Training a modular neural network with artificially injected noise enables accurate classification of raw single-lead ECG signals (95.7% accuracy, 99.5% precision) without requiring preprocessing adaptive filters that can distort clinically relevant waveform morphologies.

Abstract

Millions of electrocardiograms (ECG) are interpreted every year, requiring specialized training for accurate interpretation. Because automated and accurate classification ECG signals will improve early diagnosis of heart condition, several neural network (NN) approaches have been proposed for classifying ECG signals. Current strategies for a critical step, the preprocessing for noise removal, still are unsatisfactory. We propose a modular NN approach based on artificial noise injection, to improve the generalization capability of the resulting model. The NN classifier initially performed a fairly accurate recognition of four types of cardiac anomalies in simulated ECG signals with minor, moderate, severe, and extreme noise, with an average accuracy of 99.2%, 95.1%, 91.4%, and 85.2% respectively. Ultimately we discriminated normal and abnormal heartbeat patterns for single lead of raw ECG signals, obtained 95.7% of overall accuracy and 99.5% of Precision. Therefore, is a useful tool for the detection and diagnosis of cardiac abnormalities.

Discussion

No takes yet. Share an insight, caveat, or question.

Cite This Study

Ochoa-Brust et al. (2019) studied Cardiac abnormalities (myocardial infarction, dysrhythmia, myocardial hypertrophy, right atrial enlargement, junctional rhythm, cardiac ischemia) (n=221). Noise-tolerant modular neural network (NN) classifier trained with artificial Gaussian white-noise injection (1-12%) vs. True clinical diagnosis / reference ECG annotations (and literature comparison against prior filtering-based NN approaches) was evaluated on Overall classification accuracy for discriminating normal heartbeat vs. cardiac abnormality on single-lead (V1) raw real ECG signals. A noise-tolerant modular neural network trained with artificial noise injection achieved 95.7% overall accuracy and 99.5% precision in classifying raw single-lead ECG signals without adaptive filtering.

synapsesocial.com/papers/6a74766401b447f0dadbd6ad https://doi.org/10.31449/inf.v43i1.1605

View Full Paper

Ask AI

Bookmark

Share

Noise-tolerant modular neural network system for classifying ECG signal

Key Points

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Noise-tolerant modular neural network system for classifying ECG signal

Key Points

Structured PICO

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading