Cases/CROSS/C14
Cross-Domain SignalsMedical-device makers / primary care / AI-assisted diagnosis

Heart-Sound (PCG) Anomaly Screening

01

Pain Points

  • !Stethoscope recordings need S1/S2 segmentation and murmur spectral features, but lack a standardized workflow.
  • !Normal/abnormal binary classification depends on experience and is hard to scale and reproduce.
  • !Low-sample-rate, single-channel, low-frequency signals need dedicated preprocessing and feature engineering.
02

The Tinia Approach

Low-frequency band-pass + S1/S2 localization + murmur-segment spectral features + AutoML discrimination — the same operators carry directly to physiological signals.

20–400 Hz band-passS1/S2 localizationMurmur spectrumMel featuresAutoML discrimination
03

Nodes Used

IIR Filteriir_filter
20–400 Hz band-pass preprocessing
Active Segment Detectionactive_segment
S1 / S2 beat localization
FFT Spectrum Analysisfft_spectrum
Murmur-segment spectrum
Spectral Feature Extractionfbank_extract
Mel spectral features
Feature Mergefeature_merge
Feature merging

Core nodes; add or remove steps as needed for your data and standards — the node graph is always editable.

04

Expected Output

  • S1 / S2 beat periods and murmur-segment spectral features.
  • Normal / abnormal binary classification (AutoML discriminant function).
  • A scalable, reproducible screening workflow.
Reference Standards / Engineering PracticePhysioNet open datasets · clinical guidelines

Want to run this workflow on your own data?

Request a Trial
CROSS

Other cases in this domain