PaperlayerView exact revision on arXiv

Source-linked AI summary

Validation of HRV Studio: A Transparent and Quality-Control-Aware Platform for Heart Rate Variability Analysis

Cyrus Mexon Evrard Djindot, Faliang Liu, Sylvain Laborde, Yinjia Zhang, Jessie Chen, Ming Li, Congrong Wang, Weixiong Rao, Qinpei Zhao

arXiv:2608.24241v1physics.med-phcs.LGcs.SEq-bio.QM

TL;DR

HRV analysis reproducibility is limited by differences in preprocessing and computational conventions across software platforms. This study developed and evaluated HRV Studio through cross-platform benchmarking and controlled robustness tests, finding strong agreement when NN sequences and analytical conventions were harmonized, while stress tests demonstrated computational robustness rather than clinical validity.

  • Problem

    Differences in preprocessing and computational conventions across HRV software platforms limit reproducibility.

  • Method

    The study evaluated an open-source HRV Studio platform through NeuroKit2 and Kubios benchmarking, spectral-method comparisons, duration analysis, synthetic perturbations, and arrhythmia-focused QC stress testing.

  • Results

    HRV Studio showed consistent cross-platform agreement across time-domain, frequency-domain, and nonlinear indices when preprocessing and analytical conventions were explicitly harmonized.

  • Takeaways & Limitations

    HRV Studio supports transparent and reproducible HRV research workflows under explicitly defined preprocessing and spectral conventions.

  • Takeaways & Limitations

    The Kubios benchmark was relatively small, manually exported, and manually curated, limiting generalizability; residual spectral variability also remained under harmonized settings.

Abstract

from arXiv · show

Reproducibility of heart rate variability (HRV) analysis is limited by differences in preprocessing and computational conventions across software platforms. We developed HRV Studio, an open-source PyQt6-based desktop application integrating transparent HRV analysis with automated quality-control (QC) diagnostics. Validation included large-scale agreement with NeuroKit2, targeted Kubios benchmarking, spectral-method comparison, synthetic perturbation testing, recording-duration sensitivity analysis, and arrhythmia-focused QC stress testing. HRV Studio showed near-identical agreement for the widely used time-domain indices RMSSD and SDNN under matched conditions. In the primary five-minute NeuroKit2 comparison, frequency-domain median relative errors were 1.35% for LF, 0.18% for HF, and 1.41% for LF/HF, while VLF remained more convention-sensitive (37.79%). Nonlinear Poincaré indices also demonstrated high consistency. Sequence-harmonized Kubios benchmarking confirmed near-identical agreement for time-domain and nonlinear indices and strong agreement for most frequency-domain measures. Extended ten-minute analyses reproduced the same overall pattern with lower disagreement for some convention-sensitive spectral outputs. Synthetic and arrhythmia stress tests maintained 100% numerical stability while consistently triggering QC warnings. Overall, HRV Studio provides a transparent and reproducible platform for HRV research, with strong cross-platform consistency when NN sequences, preprocessing, and analytical conventions are harmonized. Stress-test results indicate computational robustness rather than clinical validation.

1. Introduction

HRV analysis is broadly used, but reproducibility is constrained by methodological differences across preprocessing and spectral-analysis conventions. HRV Studio was developed to provide transparent, quality-control-aware analysis and was evaluated through staged cross-platform and robustness comparisons.

  • HRV quantifies variation in timing between successive cardiac intervals and is used across biomedical, physiological, and clinical research.Applications include cardiovascular physiology, stress and recovery, sleep, exercise, neurophysiology, and disease monitoring.
  • Methodological choices involving artifact correction, interpolation, detrending, spectral estimation, frequency bands, and recording duration can materially alter HRV values.Frequency-domain differences may also reflect PSD computation, DC handling, frequency-bin integration, and segment processing.
  • Welch, FFT, and AR spectral methods use different assumptions about segmentation, smoothing, resolution, model order, and low-frequency behavior.These implementation differences can produce materially different frequency-domain outputs from identical RR interval data.
  • HRV Studio integrates interval preprocessing, artifact correction, spectral and nonlinear analysis, visualization, export, and user-visible QC diagnostics in an open-source Python/PyQt6 desktop platform.The platform is intended to support research workflows rather than function as a diagnostic system.
  • The validation framework assessed agreement with NeuroKit2 and Kubios, FFT and AR methods, synthetic perturbations, recording duration, and arrhythmia-focused QC behavior.The evaluation emphasized matched preprocessing and spectral settings, robustness, transparency, and warning behavior.

2.1 Software Architecture and Implementation

HRV Studio is a Python/PyQt6 desktop application and modular analysis library supporting transparent HRV processing across interval and waveform data formats. Waveform-derived interval extraction is implemented but was not a primary validation endpoint.

  • HRV Studio combines a PyQt6 GUI with a modular analysis library for preprocessing, HRV computation, visualization, QC diagnostics, and export.The source code and implementation are publicly available.
  • Supported inputs include CSV and TXT interval files, EDF waveforms, Polar HRM, Garmin FIT, Suunto SML, Movesense JSON, and compatible Biopac ACQ recordings.Inputs may contain RR or pulse-to-pulse intervals, ECG, PPG, and respiratory signals.
  • When explicit intervals are unavailable, HRV Studio can derive interval series from ECG or PPG using NeuroKit2-based signal processing.The validation package primarily evaluated downstream NN processing and metric agreement, not independent raw-waveform peak-detection accuracy.

Visualization, Editing, and Reproducibility Features

HRV Studio provides an integrated workflow for configuring, inspecting, editing, analyzing, and exporting HRV data. Its interface combines visualization and QC review with reproducibility-oriented outputs.

  • The GUI supports data loading, parameter configuration, threaded analysis, signal visualization, manual editing, and export in one environment.Threaded processing prevents the interface from blocking during computationally intensive analyses.
  • Users can review warnings and quality indicators, inspect HRV outputs, and visualize RR tachograms, heart-rate distributions, and PSD plots.The workflow supports investigator-supervised review of potentially problematic recordings.
  • Editing features include undo operations and reanalysis, while exports include PDF reports, metric tables, SPSS-compatible CSV files, and JSON audit trails.Exported outputs can include preprocessing summaries, analysis parameters, quality indicators, and manual-editing information.
  • Figure 1 presents the integrated combination of RR preprocessing, time-domain, frequency-domain, nonlinear analysis, visualization, QC diagnostics, and export.

2.2 RR-to-NN Interval Preprocessing and Quality Control

HRV Studio converts raw RR intervals into cleaned NN sequences through validation, artifact handling, and preprocessing diagnostics. Its QC layer flags conditions that may compromise interpretation without treating finite numerical output as physiological validity.

  • RR-to-NN Interval Preprocessing and Quality Control: The preprocessing pipeline integrates diagnostics and warnings with HRV computation to reduce nonphysiological distortions and identify convention-sensitive outputs.
  • RR-to-NN Interval Preprocessing and Quality Control: RR intervals denote raw beat-to-beat sequences, whereas NN intervals are the cleaned sequences remaining after artifact handling and exclusion of non-sinus beats.This terminology follows Task Force recommendations.
  • RR-to-NN Interval Preprocessing and Quality Control: Input validation normalizes units when necessary and excludes non-finite or nonpositive intervals before downstream processing.
  • RR-to-NN Interval Preprocessing and Quality Control: Artifact detection screens for missed, ectopic, and extra beats and abrupt timing discontinuities using interval thresholds and local variability criteria.When enabled, correction may reconstruct intervals through interpolation, including cubic splines where appropriate.
  • RR-to-NN Interval Preprocessing and Quality Control: QC diagnostics report signal quality, duration warnings, frequency-band sufficiency, rhythm irregularity, and method-specific spectral behavior.Warnings identify insufficient resolution, unstable VLF estimation, excessive artifact burden, correction dependence, and irregular rhythm structure.
  • RR-to-NN Interval Preprocessing and Quality Control: Figure 2 depicts a modular pipeline from physiological signals and interval series through preprocessing, QC assessment, HRV computation, visualization, and export.

2.3 HRV Metric Computation

HRV Studio computes time-domain, frequency-domain, nonlinear, and respiratory-derived metrics from validated NN interval sequences, with transparent preprocessing traceability and method-specific diagnostics. Welch PSD is the primary spectral reporting method, while FFT and AR methods support methodological comparison.

  • Overview: HRV Studio computes time-domain, frequency-domain, nonlinear, and respiratory-derived metrics from validated NN interval sequences.The framework emphasizes transparent computation, method-specific diagnostics, and explicit preprocessing traceability.
  • Time-Domain Analysis: SDNN, RMSSD, pNN50, and pNN20 quantify overall or short-term variability from preprocessed NN interval sequences.SDNN reflects broader variability across the recording, whereas RMSSD, pNN50, and pNN20 primarily characterize short-term variability.
  • Frequency-Domain Analysis: Welch power spectral density estimation is the primary frequency-domain method, supplemented by whole-signal FFT periodograms and adaptive AR estimation.Welch-derived estimates were prioritized for interpretation and reporting; FFT and AR outputs supported methodological comparison and diagnostic evaluation.
  • Frequency-Domain Analysis: 0 < f ≤ 0.40 Hz defines total physiological power, while mutually exclusive ULF, VLF, LF, and HF bands prevent double counting.The standard convention excludes the direct-current component and assigns adjacent boundaries exclusively.
  • Frequency-Domain Analysis: Band-specific spectral power is obtained by integrating the PSD over each corresponding frequency interval.The framework also records relative powers, peak frequencies, segment behavior, frequency resolution, band sufficiency, and spectral consistency indicators.
  • Nonlinear Analysis: Nonlinear analyses characterize signal complexity and temporal organization using Poincaré descriptors including SD1, SD2, and their ratio.SD1 and SD2 summarize short- and longer-term variability components, respectively.

2.4 Validation Design and Comparator Framework

Validation used a staged framework spanning matched cross-platform agreement, spectral-method comparison, duration sensitivity, synthetic perturbations, and arrhythmia-focused robustness and QC testing. Comparator analyses were designed to harmonize preprocessing and spectral conventions, while arrhythmic segments were reserved for engineering stress tests rather than primary agreement analyses.

  • Validation Design: Validation progressed from controlled agreement assessment to robustness and stress testing across multiple datasets and comparators.The design examined matched-setting agreement, methodological sensitivity, degraded-signal robustness, and QC transparency.
  • Validation Design: Large-scale NeuroKit2 agreement used relatively clean normal-sinus-rhythm recordings to minimize rhythm-related confounding.These recordings supported agreement, duration sensitivity, and spectral-method evaluations.
  • Validation Phases: The validation phases included NeuroKit2 agreement, targeted Kubios benchmarking, FFT and AR evaluation, synthetic corruption testing, duration sensitivity, and arrhythmia QC stress testing.Endpoints included relative error, duration stability, finite-output rate, warning behavior, and preprocessing sensitivity.
  • Comparator Framework: The Kubios benchmark comprised 50 exported and 44 reviewed quality-controlled files for matched-setting comparison.The benchmark evaluated agreement using relative error, correlation, median agreement, and outlier sensitivity.
  • Robustness Testing: Synthetic tests introduced missed beats, ectopy, jitter, dropouts, and artifacts to evaluate numerical stability, warning visibility, and correction behavior.Arrhythmia testing used selected MITDB recordings to assess finite outputs and QC behavior under non-normal rhythms.
  • Robustness Testing: Arrhythmia stress-test segments were intentionally selected from heterogeneous rhythm conditions and were not intended to represent MITDB’s overall arrhythmia distribution.The cohort comprised 12 predefined segments from 8 recordings and served software-robustness evaluation.
  • Comparator Harmonization: Comparator frameworks were harmonized using explicit preprocessing and spectral-processing conventions to reduce configuration-driven disagreement.Kubios benchmarking used comparator-compatible bands of 0–0.04 Hz for VLF, 0.04–0.15 Hz for LF, 0.15–0.40 Hz for HF, and 0–0.40 Hz for total power.
  • Comparator Harmonization: The primary Kubios benchmark used no detrending as a comparator-matching strategy, although Kubios Standard normally recommends Smoothness Priors for routine short-term HRV analysis.Exact implementation equivalence could not be assumed because all internal Kubios processing details were not exposed.

2.5 Statistical Analysis and Validation Endpoints

Validation endpoints combined metric-level agreement, correlation, numerical finiteness, warning behavior, and robust error summaries. Interpretation remained cautious for convention-sensitive spectral measures and excluded arrhythmic recordings from primary agreement analyses.

  • Agreement Endpoints: RMSSD, SDNN, pNN50, SD1, and SD2 were evaluated alongside frequency-domain metrics to assess consistency across HRV representations.The endpoint set covered time-domain, nonlinear, and spectral measures.
  • Agreement Endpoints: Relative error was the primary quantitative endpoint for comparing HRV Studio with matched-setting comparator frameworks.Comparator values came from NeuroKit2, Kubios, or method-specific reference outputs depending on the validation phase.
  • Statistical Analysis: Correlation assessed inter-recording metric structure, but it was interpreted alongside relative-error distributions because correlation alone does not establish numerical agreement.Finite-output rate measured the proportion of analyses producing numerically finite outputs without computation failure.
  • Statistical Analysis: Median statistics were prioritized because frequency-domain errors can be skewed and sensitive to retained pathological or convention-sensitive recordings.Means were reported secondarily as outlier-sensitive summaries, especially for the limited Kubios benchmark subset.
  • Interpretation Boundaries: VLF and absolute spectral power were interpreted cautiously because preprocessing, detrending, DC handling, interpolation, PSD integration, duration, and frequency-bin treatment can affect them.The interpretation boundaries restrict frequency-domain claims to the completed evidence package.
  • Interpretation Boundaries: Arrhythmic recordings were used exclusively for engineering robustness and QC stress testing, not for primary HRV agreement analyses.This boundary separates computational stress-test evidence from cross-platform agreement evidence.

3. Results

Matched-condition validation showed strong agreement between HRV Studio and NeuroKit2 across major HRV domains, while VLF and absolute-power measures remained more convention-sensitive. Kubios comparisons likewise improved substantially when effective NN interval sequences and analytical conventions were harmonized.

  • NeuroKit2 agreement: 1.35% LF, 0.18% HF, and 1.41% LF/HF median relative errors demonstrated strong five-minute agreement with NeuroKit2.Total power and VLF were more convention-sensitive, with median relative errors of 13.57% and 37.79%, respectively.
  • NeuroKit2 agreement: 1.21% LF, 0.11% HF, and 1.25% LF/HF median relative errors reproduced the five-minute agreement pattern in the ten-minute analysis.VLF and total-power errors decreased to 22.72% and 6.03%, while correlations were 0.9998 or higher for LF, HF, LF/HF, LFnu, and HFnu.
  • Cross-platform pattern: Strong matched-setting consistency extended across time-domain, frequency-domain, and nonlinear indices, with VLF and absolute power remaining more method-sensitive.The sensitivity involved detrending, spectral resolution, and low-frequency estimation choices.
  • Kubios benchmark: 3.49% LF and 2.54% HF median relative errors followed sequence harmonization against Kubios, with correlations of r = 0.973–0.998 across frequency-domain metrics.LF/HF showed 5.55% median relative error and r = 0.982, while VLF remained higher at 14.01%.
  • Kubios benchmark: Below 0.001% median relative errors for RMSSD, SDNN, pNN50, and SD1 demonstrated near-identical Kubios agreement after using identical NN sequences.SD2 showed 0.0705% median relative error with Pearson correlation 0.9999.

3.3 Smoothness Priors Sensitivity Analysis

The sensitivity analysis found that spectral agreement depended strongly on the estimator and implementation conventions. Welch PSD supported the primary reporting framework, AR comparisons were close, and FFT outputs were more variable under the evaluated settings.

  • Comparison constraints: Kubios Welch estimates were unavailable, and the Kubios FFT window function was not exposed, preventing exact implementation matching.Comparisons were therefore estimator-specific: Kubios FFT versus HRV Studio FFT and Kubios AR versus HRV Studio AR.
  • Smoothness Priors sensitivity: 1.40% VLF, 1.39% LF, and 3.18% HF median relative errors showed close AR-based Kubios agreement, with Pearson r ranging from 0.989 to 0.999.FFT median relative errors ranged from 4.10% for LFnu to 18.41% for LF, with VLF at 17.44% and r = 0.586.
  • Smoothness Priors sensitivity: Agreement depended strongly on spectral estimation method, with larger FFT discrepancies likely reflecting remaining implementation differences.The analysis indicates estimator-specific implementation differences contributed more to cross-platform variability than detrending alone.
  • Spectral-method comparison: Welch PSD showed strong matched-setting agreement with NeuroKit2 and was selected as HRV Studio’s primary frequency-domain reporting method.FFT and AR outputs were treated as methodological checks rather than interchangeable replacements for Welch-derived estimates.
  • Spectral-method comparison: 25/100 recordings showed more than 50% AR model-order spread, while 3/100 triggered AR instability flags.The figure compares Welch, FFT, and AR PSD-area consistency and identifies FFT sensitivity under the no-detrend baseline.

3.5 Synthetic Robustness and QC Behavior

Stress testing evaluated computational stability, QC visibility, and duration sensitivity under synthetic perturbations and non-normal rhythms. HRV Studio produced finite outputs and surfaced warnings, while short recordings produced larger and metric-dependent spectral disagreement.

  • Synthetic robustness and QC: 60/60 synthetic perturbation cases produced finite outputs before and after correction, and all 60/60 corrected cases surfaced QC warnings.Perturbations included missed beats, ectopic short–long patterns, Gaussian jitter, dropouts, and extreme artifacts.
  • Synthetic robustness and QC: 60/60 cases maintained numerical robustness while diagnostic or warning indicators consistently identified degraded or potentially unreliable recordings.The stress tests assessed numerical stability, quality-control visibility, and preprocessing sensitivity under degraded signals.
  • Duration sensitivity: 60.29% LF/HF median relative error at 30 seconds declined to 11.98% at five minutes under duration-matched comparisons.The corresponding values were 27.66% at 60 seconds, 28.61% at 2 minutes, 19.20% at 3 minutes, and 12.57% at 10 minutes.
  • Arrhythmia robustness: The MIT-BIH cohort contained 12 segments from 8 recordings and was used for engineering stress tests rather than agreement or clinical validation.The cohort evaluated numerical robustness and QC behavior under non-normal rhythm conditions.

4. Discussion

HRV Studio showed strong cross-platform agreement when effective NN sequences, preprocessing, and spectral conventions were harmonized, while VLF and some absolute spectral outputs remained convention-sensitive. Quality-control stress testing demonstrated computational robustness and warning visibility, not physiological or clinical validity.

  • Cross-platform agreement: Median relative errors remained below 6% for LF, HF, LF/HF, normalized measures, and total power after sequence harmonization, while VLF remained highest at 14.01%.These results indicate greater sensitivity of VLF to low-frequency preprocessing and spectral-processing conventions.
  • Cross-platform agreement: RMSSD, SDNN, and SD1 showed near-identical NeuroKit2 agreement under matched preprocessing, with median relative errors approaching zero and correlations of 1.000.pNN50 also showed highly consistent agreement, although relative-error interpretation was more sensitive when reference values were very small or zero.
  • Cross-platform agreement: Harmonizing effective NN interval sequences and analytical conventions substantially reduced apparent cross-platform disagreement.In the Kubios benchmark, LF/HF median relative error decreased from 17.56% to 5.55%, while VLF decreased from 31.92% to 14.01%.
  • Spectral-method sensitivity: Matched Smoothness Priors preprocessing produced closer agreement for AR-derived than FFT-derived metrics, with residual differences concentrated in VLF and absolute spectral powers.The FFT implementations differed in signal coverage and windowing: HRV Studio used whole-signal estimation, whereas Kubios reported 120-s windows with 75% overlap.
  • Spectral-method sensitivity: HRV Studio provides multiple detrending options, with Smoothness Priors using the Tarvainen formulation and a default λ = 500, while no detrending primarily serves comparator matching.The platform uses AR order 16 under the standard 4 Hz interpolation framework, although LF/HF remained strongly order-dependent in some recordings.
  • Duration and interpretation: Ultra-short recordings showed greater disagreement across several frequency-domain measures, while VLF error remained 33.71% at five minutes and 38.75% at ten minutes.Extending recording duration alone did not resolve VLF instability.
  • Quality control and scope: Synthetic perturbation and arrhythmia-focused tests consistently triggered warnings for degraded interpretability while maintaining finite numerical output.Stable computation under degraded conditions indicates operational robustness, not physiological validity or clinical validation.
  • Quality control and scope: The Kubios benchmark was small, manually exported, and partly dependent on inaccessible internal spectral-processing details, limiting generalizability and interpretation of residual differences.The exact Kubios FFT window function was not exposed in the available exports.

5. Conclusion

This study validated HRV Studio through staged cross-platform, spectral-method, duration-sensitivity, and stress-testing analyses. The findings support transparent and reproducible HRV research workflows, with agreement strongly dependent on methodological harmonization.

  • 5. Conclusion: HRV Studio is a Python/PyQt6 platform designed for transparent, reproducible, and quality-control-aware HRV analysis.Its validation used NeuroKit2 comparison, Kubios benchmarking, spectral-method evaluation, synthetic testing, duration analysis, and arrhythmia-focused QC stress testing.
  • 5. Conclusion: Harmonized conditions supported strong cross-platform reproducibility across frequency-domain, time-domain, and nonlinear HRV measures.Supported measures included LF, HF, total power, normalized spectral measures, LF/HF, RMSSD, SDNN, pNN50, and nonlinear Poincaré descriptors.
Loading 2608.24241v1…