WIPSNet: Deep Learning for Paediatric Wheeze Detection from Overnight Impedance Pneumography
Authors: Felix Oury, Harley Day, Karina Mayoral, Ville-Pekka Seppä, Sejal Saglani, Reiko J. Tanaka
Organizations: Department of Computing, Imperial College London, London, United Kingdom · Department of Bioengineering, Imperial College London, London, United Kingdom · National Heart and Lung Institute and Centre for Paediatrics and Child Health, Imperial College London, London, United Kingdom; Department of Respiratory Paediatrics, Royal Brompton Hospital, London, United Kingdom · Icare Finland Ltd, Vantaa, Finland
Overnight impedance pneumography (IP) is used to monitor paediatric respiratory health. Its current clinical readout, the Expiratory Variability Index (EVI), compresses each IP recording into a single scalar and achieves an AUC of 0.633 for night-level wheeze classification. We introduce Wheeze Impedance Pneumography Scalogram Network (WIPSNet), a 3D ResNet operating on stacked continuous wavelet transform scalograms of overnight IP signals. On a 15-patient cohort (60 nights, 281 hours), WIPSNet achieves an AUC of 0.783±0.026, outperforming EVI, a state-space model (Mamba), and two modern sleep-staging architectures. Performance peaks at a volumetric depth corresponding to 32 minutes of temporal context, suggesting that multi-scale temporal aggregation is important for modelling nocturnal respiratory dynamics. Overall, these results indicate that structured time-frequency representations combined with 3D convolutional architectures provide an effective approach for learning from long, irregular physiological time series.
Figures & tables
Figure 1: WIPSNet framework and architecture. (a) The IP signal is filtered, transformed into scalograms, grouped into 3D volumes, and passed through a 3D CNN; volume-level probabilities are aggregated into a night-level prediction. (b–d) Raw signal, processed signal, and 4-minute scalogram. (e) Volumetric input construction: D consecutive scalograms stacked along the depth axis. (f) The bottleneck 3D ResNet backbone.
Model
Input & backbone
AUC
AUPRC
WIPSNet
CWT + 3D CNN
.783±.026
.634±.047
AttnSleep
Raw + MRCNN-attn
.764±.030
.603±.048
catch22+XGBoost
22 canonical TS features
.704±.002
.481±.002
Mamba
Raw + SSM
.689±.047
.509±.060
EVI †
Clinical scalar
.633
.433
SleePyCo
Raw + Pyramid-Transformer
.577±.152
.444±.101
Table 1 : Night-level wheeze classification results. Mean ± std over 5 seeds, 15-fold patient-level LOOCV. AUPRC random baseline at our class prevalence is 0.35. Mamba and Mamba-Scalogram both use a 32-min window. † Deterministic; 57/60 valid nights.
Figure 2 : WIPSNet’s depth-scaling behaviour. Both AUC and AUPRC peak at D=8 (32 min of temporal context) and decline at longer windows, indicating an optimal macro-timescale rather than monotone improvement. Error bars: standard deviation over 5 seeds. Top axis: temporal context in minutes.
Appendix figures & tables3 assets
Supplementary material from the paper’s appendix.
Appendix
Parameter
Search Space
Selected Value
Scalogram window length
2–16 min
60,000 (4 min)
Volume depth D
1–10 scalograms
8
Stride (scalogram)
1– D scalograms
8 ( =D , non-overlapping)
Backbone
ResNet / EfficientNet
ResNet
Batch size
{8, 16, 32, 64}
32
Learning rate
10−5 – 10−2
3.00×10−4
Appendix
Table 2 : WIPSNet hyperparameter search space and selected values.
Parameter
Search Space
Selected Value
Context window W (4-min segments)
{1, 2, 4, 8} sweep
8 ( =32 min)
Stride (segments)
=W (non-overlapping)
8
Per-segment length
—
5,000 samples (4 min) 3 3 3 The raw 250 Hz IP signal is downsampled 12× to ≈ 20.8 Hz before Mamba, so a 4-min segment is 5,000 samples and a 32-min window is 8×5,000=40,000 samples. This keeps the SSM sequence length tractable; on the full-rate 480,000-sample window Mamba collapses to near-chance.
Sequence length
W× seg. length
40,000 samples (32 min)
dmodel
16–192
32
dstate
8–48
32
Appendix
Table 3 : Mamba SSM hyperparameter search space and selected values. The context window W (number of consecutive 4-min segments) was swept separately, analogous to the WIPSNet volume depth D , while the architecture was tuned by Optuna.
Model
Input & backbone
AUC
AUPRC
WIPSNet-CWT ( D=8 )
CWT + 3D CNN
.783±.026
.634±.047
AttnSleep
Raw 1D + MRCNN + attention
.764±.030
.603±.048
WIPSNet-CWT ( D=4 )
CWT + 3D CNN
.747±.032
.556±.035
WIPSNet-CWT ( D=1 )
CWT + 3D CNN
.744±.022
.527±.026
WIPSNet-CWT ( D=32 )
CWT + 3D CNN
.744±.042
.546±.049
WIPSNet-CWT ( D=2 )
CWT + 3D CNN
.742±.035
.558±.044
Appendix
Table 4 : All models and ablations, sorted by AUC. Mean ± std over 5 seeds for DL models. AUC is the primary ranking metric; AUPRC reflects performance on the minority (wheeze) class under class imbalance. Depth-sweep rows ( D=8 ) and front-end rows share WIPSNet’s backbone; the headline model is WIPSNet-CWT ( D=8 ).
Apnoea of prematurity is characterised by recurrent episodes of cessation of breathing and remains difficult to detect reliably using routinely monitored physiological signals in the Neonatal Intensive Care Unit (NICU). Existing bedside monitors rely primarily on respiratory rate and oxygen saturation thresholds, often generating high false-positive alarm rates and missing short or irregular events. Improving automated detection using routinely acquired clinical signals could enhance identification of clinically meaningful events without additional sensing hardware. We evaluated deep learning-based detection of apnoea-related Cessation Of BrEathing (COBE) events using impedance pneumography (IP), electrocardiography (ECG), and photoplethysmography (PPG) signals from approximately 430 hours of NICU recordings collected from 24 pre-term infants. Three independent reviewers annotated COBE events, producing a dataset of 346 COBE and 608 non-COBE events. We compared a shallow convolutional neural network (CNN), residual networks (ResNets), and a ConvNeXt architecture using an independent held-out test set. Across all architectures, detection performance was influenced more strongly by signal modality than by architectural complexity. Unimodal IP-based models achieved balanced accuracies of 86.8-88.0%, outperforming ECG-derived (62.6-69.7%) and PPG-derived (65.1-66.4%) respiratory surrogates. Multimodal fusion yielded modest improvements over IP alone. The best-performing model, a ConvNeXt architecture combining IP and PPG inputs, achieved 88.7% balanced accuracy and an F1 score of 0.75 on the independent test set. These findings demonstrate that deep learning models applied to routinely monitored NICU signals can reliably detect COBE events and highlight the importance of signal modality in data-constrained neonatal monitoring settings.
False ventricular tachycardia (VT) alarms are a leading contributor to alarm fatigue in intensive care units. We propose a deep learning framework combining a 1D SE-ResNet with ICU-realistic data augmentations and a physics-informed auxiliary reconstruction task based on the three-element Windkessel hemodynamic model, implemented as a differentiable forward simulation. By requiring the network's latent representation to produce physiologically plausible arterial pressure waveforms, artifact-driven ECG patterns are penalized while true VT remains coherent across modalities. Evaluated on the VTaC benchmark under a strict real-time protocol (10-second pre-alarm window), our method achieves a 5-point Challenge Score improvement over prior state-of-the-art. Ablation studies confirm that the physics-informed objective is the primary performance driver, providing gains in accuracy, 2x label efficiency, and more localized and clinically meaningful ECG segments.
Athanasios Papastathopoulos-Katsaros, Alexandra Stavrianidi, Zhandong Liu
Department of Pediatrics, Baylor College of Medicine, Houston, TX, USA · Jan and Dan Duncan Neurological Research Institute, Texas Children’s Hospital, Houston, TX, USA · Institute for Analysis and Numerics, University of Münster, Germany +1
Sleep is essential for health, yet studying its dynamics requires manual sleep staging, a labor-intensive step in research and clinical care. Across centers, polysomnography (PSG) recordings are traditionally scored in 30-s epochs for pragmatic, not physiological, reasons and vary in electrode count, montage, and subject characteristics. These constraints challenge harmonized multi-center studies and the discovery of robust biomarkers on shorter timescales. We present AnySleep, a deep neural network that scores sleep from any electroencephalography (EEG) or electrooculography (EOG) data at adjustable temporal resolutions. We trained and validated the model on over 20,000 overnight recordings (> 200,000 hours of EEG and EOG) from 28 datasets across multiple clinics to promote robust generalization across sites. The model attains state-of-the-art performance and surpasses or equals established baselines at 30-s epochs. Performance improves with more channels, yet remains strong when EOG is absent or only EOG or single EEG derivations (frontal, central, or occipital) are available. On sub-30-s timescales, the model captures short wake intrusions consistent with arousals and improves prediction of pathophysiological conditions (obstructive sleep apnea, narcolepsy type 1, insomnia) over 30-s scoring. We make the model publicly available to facilitate large-scale studies with heterogeneous electrode setups and accelerate biomarker discovery in sleep.
Department of Medical Engineering and Technomathematics, FH Aachen University of Applied Sciences, 52428 Jülich, Germany · Department of Information and Computing Sciences, Utrecht University, Utrecht, The Netherlands · Institute for Data-Driven Technologies, FH Aachen University of Applied Sciences, 52428 Jülich, Germany