DRW-CNN: A Dilated Residual Wavelet CNN for ECG-based Severe Heart Failure Screening

Q: What is the clinical evidence from this study?

Population: Severe Congestive Heart Failure (NYHA class III-IV) vs Normal Sinus Rhythm (n=33). Intervention: Dilated Residual Wavelet Convolutional Neural Network (DRW-CNN) ECG screening model vs. Clinically confirmed cohort membership (BIDMC severe CHF vs MIT-BIH normal sinus rhythm ground truth). Primary outcome: Discrimination of confirmed severe CHF from normal sinus rhythm on held-out test segments at prespecified threshold 0.5 (Accuracy and ROC-AUC) (Accuracy 95.238%, ROC-AUC 0.988, 95% CI 91.071%-98.810%).

Q: What does this research mean for the field?

A dilated residual wavelet convolutional neural network (DRW-CNN) discriminates severe congestive heart failure from normal sinus rhythm with 95.2% accuracy and a 0.988 ROC-AUC under a record-independent evaluation framework. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

Amit M. Sahu; Jayant P. Mehare

doi:10.14445/23488379/ijeee-v13i9p106

October 8, 2026International Journal of Electrical and Electronics EngineeringOpen Access

DRW-CNN: A Dilated Residual Wavelet CNN for ECG-based Severe Heart Failure Screening

View Full Paper

Ask AI

Bookmark

Share

Key result

DRW-CNN discriminates severe CHF from normal sinus rhythm on ECG with ~95% accuracy.

Accuracy 95.238%, ROC-AUC 0.988
95% CI 91.071%-98.810%
n=33

Why the study?

ECG-based heart-failure screening requires clinically valid labels, record-independent evaluation, and uncertainty analysis at the level of independent records.

Can a compact dilated residual wavelet convolutional neural network (DRW-CNN) accurately discriminate severe congestive heart failure (NYHA III-IV) from normal sinus rhythm using single-channel ambulatory ECG segments under record-independent validation?

Key Points

Evaluate a dilated residual wavelet convolutional neural network (DRW-CNN) for discriminating clinically confirmed severe congestive heart failure from normal sinus rhythm using record-independent evaluation.
Analyzed 15 BIDMC severe congestive heart failure records (NYHA class III-IV) and 18 MIT-BIH Normal Sinus Rhythm Database records partitioned at the record level before segmentation into 528 training, 96 validation, and 168 held-out test segments.
Applied level-4 Daubechies-4 wavelet denoising, z-score normalization, residual blocks with dilation factors 1, 2, and 4, global average pooling, sigmoid classification, and class-weighted binary cross-entropy.
On 168 held-out test segments from seven unseen records (72 CHF, 96 normal), the model achieved 95.238% accuracy (95% bootstrap CI, 91.071% to 98.810%), 95.833% sensitivity (95% CI, 87.500% to 100.000%), and 94.792% specificity (95% CI, 88.542% to 100.000%) at a prespecified 0.5 threshold.
Held-out classification achieved a precision of 93.243% (95% CI, 63.636% to 100.000%), F1-score of 94.521% (95% CI, 75.000% to 99.174%), MCC of 0.903, ROC-AUC of 0.988, and average precision of 0.989 with a Brier score of 0.0418.

Structured PICO

Can a compact dilated residual wavelet convolutional neural network (DRW-CNN) accurately discriminate severe congestive heart failure (NYHA III-IV) from normal sinus rhythm using single-channel ambulatory ECG segments under record-independent validation?

P

Population

33 independent ambulatory ECG recordings (15 with confirmed severe NYHA class III-IV congestive heart failure from BIDMC and 18 normal sinus rhythm controls from MIT-BIH), partitioned at the record level into 22 training, 4 validation, and 7 held-out test records (792 total segments).

E

Exposure

Dilated Residual Wavelet Convolutional Neural Network (DRW-CNN) incorporating level-4 Daubechies-4 (db4) wavelet denoising, z-score normalization, three residual 1D convolutional blocks with dilation factors {1, 2, 4}, global average pooling, dense layer with 0.30 dropout, and sigmoid classification trained with class-weighted binary cross-entropy.

C

Comparator

Ground-truth record-level cohort labels (confirmed severe CHF in BIDMC vs normal sinus rhythm in MIT-BIH NSRDB; no matched baseline model comparator evaluated on the same split).

O

Outcome

Binary classification performance for discriminating confirmed severe CHF from normal sinus rhythm on 168 held-out test segments (from 7 unseen records) at a prespecified probability threshold of 0.5, evaluated via accuracy, sensitivity, specificity, precision, F1-score, Matthews correlation coefficient (MCC), ROC-AUC, and average precision with 95% record-cluster bootstrap confidence intervals.surrogate

Main Result

Effect estimate: Accuracy 95.238%, ROC-AUC 0.988 (95% CI 91.071%-98.810%)

Provides a leakage-resistant, record-independent deep learning benchmark for ECG-based severe heart failure screening that avoids conflating generic arrhythmias with heart failure and properly accounts for record-level clustering.

Limitations

Small number of independent records (33 total records, with only 7 records in the held-out test set)
Database-source confounding because positive (BIDMC CHF) and negative (MIT-BIH NSRDB) cohorts originate from separate databases
Individual ejection-fraction phenotypes (HFrEF vs HFpEF) and complete NYHA class I-IV labels are not available
No independent external confirmed-HF target cohort validation or NSTDB wearable-noise stress testing was performed
No matched fold-wise head-to-head comparison against contemporary TCN, transformer, or Mamba architectures on the 33-record cohort
Small cohort size of 33 total independent records and only 7 held-out test records
Database-source confounding due to using separate databases for positive (BIDMC) and negative (MIT-BIH NSRDB) classes
Lack of per-record ejection fraction phenotypes (HFpEF vs HFrEF) and complete NYHA I-IV staging labels
Absence of external hospital validation cohort with confirmed HF and lack of NSTDB wearable noise-stress testing
Local theoretical convolutional receptive field is 31 samples (~124 ms at 250 Hz), which does not directly model long-term RR dynamics prior to global average pooling

Implication

Supports DRW-CNN for record-independent severe CHF screening on ECG; leaves open clinical adoption pending prospective validation.

Abstract

Electrocardiogram (ECG)-based heart-failure screening requires clinically valid labels, record-independent evaluation, and uncertainty analysis at the level of independent records. This study evaluates a Dilated Residual Wavelet Convolutional Neural Network (DRW-CNN) for discrimination of clinically confirmed severe Congestive Heart Failure (CHF) from normal sinus rhythm. The primary cohort comprises all 15 available BIDMC CHF records and 18 MIT-BIH Normal Sinus Rhythm Database records. Records are partitioned before segmentation, yielding 528 training, 96 validation, and 168 held-out test segments from mutually exclusive records. The network applies level-4 Daubechies-4 wavelet denoising, z-score normalization, residual blocks with dilation factors 1, 2, and 4, global average pooling, and sigmoid classification. Class-weighted binary cross-entropy addresses imbalance without deleting independent records. At the prespecified threshold of 0.5, held-out predictions achieve 95.238% accuracy, 95.833% sensitivity, 94.792% specificity, 93.243% precision, 94.521% F1-score, MCC 0.903, ROC-AUC 0.988, and average precision 0.989. Record-cluster bootstrap intervals, calibration, threshold sensitivity, record-level aggregation, leave-one-record-out stability, and auxiliary non-HF domain-shift stress analyses are also reported. The study provides a leakage-resistant severe-CHF screening evaluation that separates disease-specific evidence from auxiliary abnormality testing and quantifies uncertainty at record level.

Discussion

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

Cite This Study

Sahu et al. (2026) studied Severe Congestive Heart Failure (NYHA class III-IV) vs Normal Sinus Rhythm (n=33). Dilated Residual Wavelet Convolutional Neural Network (DRW-CNN) ECG screening model vs. Clinically confirmed cohort membership (BIDMC severe CHF vs MIT-BIH normal sinus rhythm ground truth) was evaluated on Discrimination of confirmed severe CHF from normal sinus rhythm on held-out test segments at prespecified threshold 0.5 (Accuracy and ROC-AUC) (Accuracy 95.238%, ROC-AUC 0.988, 95% CI 91.071%-98.810%). DRW-CNN achieved 95.238% accuracy, 95.833% sensitivity, 94.792% specificity, and a 0.988 ROC-AUC in discriminating severe CHF from normal sinus rhythm on held-out ECG records.

synapsesocial.com/papers/6ae28e38cbadf70daf6c40fd https://doi.org/10.14445/23488379/ijeee-v13i9p106

View Full Paper

Ask AI

Bookmark

Share

DRW-CNN: A Dilated Residual Wavelet CNN for ECG-based Severe Heart Failure Screening

Key Points

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

DRW-CNN: A Dilated Residual Wavelet CNN for ECG-based Severe Heart Failure Screening

Key Points

Structured PICO

Main Result

Limitations

Implication

Abstract

Discussion

Cite This Study

Keep reading

Keep reading