Key concepts

Cardiology in Gradient Biotech is organized around physiological recordings, derived cardiovascular metrics, cohort context, and reproducible analysis runs. This page explains the main cardiovascular concepts behind the workflows and the app concepts used to manage them.

Physiological signals

Most Cardiology workflows begin with one or more time-series recordings. A time-series is a sequence of measurements collected over time, often at a known sample rate.

Common signal types:

• Electrocardiogram (ECG): electrical activity of the heart, used for R-peak detection, RR intervals, heart rate variability, and arrhythmia burden analysis.
• RR intervals: beat-to-beat intervals derived from ECG R-peaks or exported by a device. The name comes from the interval between two R-peaks.
• Photoplethysmography (PPG): optical pulse waveform, often from wearables, used for pulse peaks and pulse rate variability.
• Blood pressure: systolic, diastolic, and mean arterial pressure measurements over time.
• Respiration: breathing waveform or breath timing used for respiratory rate and coupling analysis.

The raw waveform is not the final result. Pipelines first estimate signal quality, detect events such as R-peaks or pulse peaks, and then compute metrics that can be compared across subjects or groups.

Electrocardiogram and R-peaks

An electrocardiogram (ECG) records cardiac electrical activity. In many research workflows, the most important event is the R-peak, the dominant peak in a QRS complex. R-peaks define the timing of heartbeats.

Key electrocardiogram concepts:

Concept	Meaning
Sample rate	Number of waveform samples recorded per second.
Channel	One signal trace in a file; European data format (EDF) files may contain multiple channels.
R-peak	Detected heartbeat timing point used to derive RR intervals.
RR interval	Time between consecutive R-peaks.
Signal quality index (SQI)	Metric used to flag usable and low-quality segments.
Artifact rate	Fraction of signal or beats affected by noise, motion, dropouts, or detection problems.

Electrocardiogram preprocessing turns a waveform into R-peaks, RR intervals, quality metrics, and a preview that can be inspected in the dataset explorer.

Heart rate variability

Heart rate variability (HRV) summarizes beat-to-beat variation in RR intervals. HRV is often used as a research marker of autonomic regulation, recovery, stress physiology, or disease-associated cardiovascular control.

Common heart rate variability metric families:

• Time-domain metrics: direct summaries of RR intervals, such as mean heart rate, standard deviation of normal-to-normal intervals (SDNN), and root mean square of successive differences (RMSSD).
• Frequency-domain metrics: spectral power estimates such as low-frequency power (LF), high-frequency power (HF), and the LF/HF ratio.
• Nonlinear metrics: measures of rhythm structure and complexity, when supported by the pipeline.

HRV interpretation depends on recording length, posture, sleep/wake state, medications, breathing pattern, ectopy, and signal quality. A high or low value is not automatically diagnostic.

Arrhythmia burden

Arrhythmia burden summarizes irregularity patterns from beat intervals. In this product area, arrhythmia outputs are research metrics, not automatic diagnoses.

Examples include:

• Irregularity percentage: proportion of beats or intervals flagged as irregular.
• Pause burden: long intervals that may indicate pauses or missing detections.
• Beat flags: beat-level annotations used to inspect where irregularity occurred.

These metrics should be reviewed alongside waveform quality. Noise, missed peaks, motion artifact, and device export issues can look like rhythm irregularity if not checked.

Photoplethysmography and pulse rate variability

Photoplethysmography (PPG) measures blood volume changes optically, often from wearable or fingertip sensors. PPG does not directly measure cardiac electrical activity, but pulse peaks can estimate inter-beat intervals.

Pulse rate variability (PRV) is analogous to HRV but derived from pulse timing rather than ECG R-peaks. PRV can be useful when ECG is not available, but it can differ from HRV because vascular tone, pulse transit time, motion artifact, and sensor placement affect the waveform.

The photoplethysmography workflow preprocesses the waveform, detects pulse peaks, estimates quality, and computes pulse rate variability metrics from inter-beat intervals.

Blood pressure dynamics

Blood pressure workflows analyze systolic blood pressure, diastolic blood pressure, and mean arterial pressure over time.

Important concepts:

• Systolic blood pressure (SBP): pressure during cardiac contraction.
• Diastolic blood pressure (DBP): pressure during cardiac relaxation.
• Mean arterial pressure (MAP): mean pressure estimate across the cardiac cycle.
• Variability: fluctuation in blood pressure over time.
• Dipping: nighttime blood pressure reduction relative to daytime values, when timing data supports the classification.

Blood pressure interpretation depends on measurement schedule, device type, posture, sleep period definitions, antihypertensive medications, and cohort context.

Respiration and autonomic coupling

Cardiovascular signals are coupled to respiration and autonomic reflexes. Cardiology supports multimodal workflows that combine electrocardiogram or RR interval, blood pressure, and respiration recordings for a subject.

Key coupling concepts:

• Respiratory rate: breaths per minute or breath timing summary.
• Respiratory sinus arrhythmia (RSA): heart rate variation linked to the breathing cycle.
• Baroreflex sensitivity: relationship between blood pressure changes and heart period responses, estimated by sequence methods in the pipeline.
• Autonomic balance: a research interpretation of sympathetic and parasympathetic patterns using multiple metrics, not a single definitive number.

Coupling metrics are strongest when recordings are synchronized, signal quality is high, and the protocol is clear.

Subjects, cohorts, and outcomes

A subject represents one participant in a study. A subject may have multiple linked datasets across electrocardiogram, RR interval, blood pressure, photoplethysmography, and respiration recordings. Linking datasets to subjects lets the app combine metrics across modalities.

A cohort is a group of subjects analyzed together. Cohort analysis can compare treatment groups, diagnostic groups, timepoints, or outcome-defined groups.

Important cohort concepts:

• External identifier (external ID): study-facing subject identifier, such as P001.
• Group: metadata field used for comparisons, such as treatment arm or diagnosis.
• Outcome: flexible endpoint or follow-up field imported from comma-separated values, such as mace_event or followup_days.
• Batch processing: running a pipeline across many datasets in one request.
• Outlier: subject or metric value that differs strongly from the group distribution.

Cohort results should be interpreted with sample size, missingness, protocol differences, and subject-level covariates in mind.

Research risk stratification

Signal-derived risk scoring ranks subjects using computed physiological metrics and optional outcomes metadata. It is designed for research stratification, cohort exploration, and hypothesis generation.

Risk scores are not clinical predictions. They depend on the available signals, the cohort distribution, preprocessing quality, and the outcome definitions supplied by the study team. Use them to prioritize review or compare groups, not to make care decisions.

Artificial intelligence interpretation

Artificial intelligence (AI) interpretation summarizes completed metrics and links them into readable narratives.

Mode	When	Output
Descriptive	Single run or cohort summary	Plain-language summary citing computed metrics
Mechanistic	Subject with two or more metric domains, such as heart rate variability plus blood pressure	Cross-domain hypotheses with pathway or physiology tags

Interpretations should cite computed values from completed runs. Low signal quality or missing prerequisites should trigger caveats. The interpretation is a research aid, not a substitute for reviewing the waveform, artifacts, and study design.

Study

A study is the top-level container for one cardiovascular research project. It holds datasets, subject records, pipeline runs, and cohort analyses. Create one study per protocol, trial arm collection, or analysis project.

Dataset

A dataset is one uploaded recording attached to a study. Each dataset records:

• Signal type, such as electrocardiogram, RR intervals, photoplethysmography, blood pressure, or respiration
• File format, such as waveform database (WFDB), comma-separated values (CSV), European data format, rr_csv, or bp_csv
• Detected metadata, such as sample rate, duration, and channel count
• Storage path on the local filesystem

Open a dataset to run pipelines, explore waveforms, annotate time ranges, and view run history.

Pipeline run

A run is one execution of an analysis pipeline.

Each run stores:

• Unique run_id
• Pipeline name and version
• JavaScript Object Notation (JSON) parameter record
• Status: pending -> running -> completed or failed
• Output artifacts, such as JSON metrics, previews, and interpretations
• Timestamps and duration

Runs chain through provenance. For example, heart rate variability runs reference their preprocess run, baroreflex runs reference RR interval and blood pressure runs, and interpretations reference the metrics they summarize.

Job polling

All analysis runs asynchronously in backend-analysis. The frontend polls GET /jobs/{run_id} until the job completes. Active jobs appear in the navbar Jobs panel.

Artifacts

Pipeline outputs are written to data/artifacts/{run_id}/ as JavaScript Object Notation and supporting arrays where needed.

Examples:

• peaks.json, rr_intervals.json, quality.json, preview.json from preprocessing
• hrv_metrics.json from heart rate variability analysis
• arrhythmia_burden.json from arrhythmia burden analysis
• cohort_summary.json from group comparison
• risk_scores.json from cohort risk scoring
• interpretation.json from artificial intelligence interpretation

Download artifacts from the run detail Artifacts tab.

Provenance and reproducibility

Every analysis should be reproducible. Open a run to review parameters on the Inputs tab, inspect artifacts, and use Reproduce run to re-submit identical work. Compare two completed runs from a dataset Runs tab via Compare selected.

Interpretation scope

The Cardiology area supports research workflows for physiological signal analysis. It is not a regulated clinical decision support device, does not diagnose arrhythmias, and does not validate risk predictions for care decisions. Treat outputs as structured evidence for research review.