This project extends the Bioacoustic Topology framework into the domain of biomedical acoustic signal analysis.
Rather than treating cardiac auscultation as isolated waveform classification, the framework investigates whether cardiac acoustic activity may be modeled as:
- latent physiological manifolds,
- recurrent trajectory systems,
- dynamical state-space traversal,
- cyclic physiological topology,
- and distributed acoustic occupation structure.
The project explores whether phonocardiogram recordings exhibit measurable latent organization when analyzed through:
- nonlinear manifold learning,
- physiological trajectory reconstruction,
- probabilistic state transitions,
- anomaly topology,
- toroidal cyclic embedding,
- orbital phase analysis,
- and holographic physiological density rendering.
Importantly, this framework is exploratory and computational.
It does not provide clinical diagnosis or medical decision support.
The framework investigates whether physiological acoustic systems may exhibit:
- recurrent latent organization,
- partially stabilized acoustic states,
- directional manifold traversal,
- cyclic physiological recurrence,
- and distributed topological structure.
Within this perspective:
- cardiac acoustic events become state-space coordinates,
- physiological transitions become latent trajectories,
- and recurrent cardiac dynamics become structured traversal through latent physiological topology.
The broader objective is not diagnostic classification alone, but exploration of whether physiological acoustic systems may exhibit measurable organizational principles when modeled through latent geometric state-space.
Continuous cardiac waveforms are segmented into discrete computational acoustic events suitable for manifold analysis.
Each cardiac acoustic event is transformed into a multidimensional physiological descriptor vector integrating:
- MFCCs,
- spectral descriptors,
- energetic structure,
- and temporal acoustic properties.
The feature space is projected into a three-dimensional latent manifold using UMAP.
This enables investigation of:
- acoustic similarity organization,
- recurrent physiological regions,
- and trajectory evolution through latent topology.
Unsupervised clustering identifies recurrent physiological acoustic states distributed throughout latent manifold structure.
Sequential event reconstruction enables probabilistic analysis of:
- state persistence,
- recurrent transition corridors,
- and physiological trajectory evolution.
Trajectory-based anomaly scoring identifies dynamically unusual cardiac acoustic events occupying atypical latent-space regions.
Latent trajectories are additionally projected onto cyclic toroidal geometry in order to investigate recurrence-preserving physiological organization.
Orbital phase statistics are estimated to explore cyclic physiological recurrence behavior within toroidal topology.
Volumetric density rendering approximates recurrent physiological manifold occupation regions and distributed acoustic structure.
This project uses a subset of the:
PhysioNet: https://physionet.org/content/circor-heart-sound/
The dataset contains pediatric cardiac auscultation recordings collected for computational heart sound analysis research.
- Python
- Google Colab
- Librosa
- UMAP
- Scikit-learn
- Plotly
- NumPy
- Pandas
- Matplotlib
latent-physiological-topology/
│
├── notebooks/
│ └── latent_physiological_topology.ipynb
│
├── images/
│ ├── latent_physiological_manifold.png
│ ├── physiological_state_discovery.png
│ ├── physiological_anomaly_topology.png
│ ├── toroidal_physiological_topology.png
│ └── holographic_physiological_topology.png
│
├── README.md
├── LICENSE
└── requirements.txt
The framework remains exploratory and hypothesis-generating.
The project does not claim:
- clinical diagnosis,
- pathological inference,
- physiological mechanism discovery,
- or biologically validated cardiac topology.
Similarly:
- toroidal geometry,
- orbital synchronization,
- and attractor interpretations
should be understood as computational abstractions rather than direct physiological mechanisms.
Potential future extensions include:
- larger-scale physiological datasets,
- comparative cardiac manifold analysis,
- multimodal physiological integration,
- graph neural network trajectory modeling,
- persistent homology and topological data analysis,
- real-time physiological topology rendering,
- and streaming biomedical manifold visualization systems.
This project contributes toward an emerging computational perspective combining:
- biomedical acoustic analysis,
- manifold learning,
- nonlinear dynamical systems,
- geometric machine learning,
- computational topology,
- and physiological trajectory modeling.
More broadly, the framework explores whether physiological systems may be investigated not only as signals, but as evolving dynamical structures embedded within latent geometric state-space.
If referencing this work:
@software{palis2026latentphysiologicaltopology,
author = {Palis, Sabrina},
title = {Latent Physiological Topology: Dynamic State-Space Modeling of Cardiac Acoustic Signals},
year = {2026},
publisher = {Zenodo},
doi = {10.5281/zenodo.MY_DOI_HERE}
}Sabrina Palis
Independent Researcher
AI • Computational Topology • Dynamical Systems • Bioacoustic Modeling
This project is intended solely for exploratory computational research and educational purposes.
It is not a medical device and should not be used for clinical diagnosis, treatment, or healthcare decision-making.