prCARS

prCARS is a Python toolkit for phase retrieval, non-resonant-background correction, preprocessing, and Raman-like signal reconstruction from Coherent Anti-Stokes Raman Scattering (CARS/BCARS) spectra.

It provides a modular pipeline for recovering Raman-like information using classical retrieval methods — Kramers-Kronig and the Maximum Entropy Method — with an optional neural-network retrieval interface.

---

Why this project matters

CARS and BCARS spectra carry chemically meaningful Raman-like information, but the measured signal is distorted by the non-resonant background, interference effects, instrument response, noise, and baseline artifacts. Recovering the underlying Raman-like signal is therefore an inverse problem, and the answer depends on the preprocessing and retrieval choices made along the way.

prCARS makes those choices explicit and comparable. It provides a research-oriented workflow for:

• preprocessing CARS/BCARS spectra
• estimating and correcting background contributions
• retrieving Raman-like spectral content
• comparing retrieval methods against a known target
• testing phase-retrieval pipelines on synthetic examples

The goal is to make CARS/BCARS retrieval easier to test, compare, and integrate into scientific machine-learning workflows.

---

Example output

---

Quickstart

prCARS ships with a synthetic data generator, so this runs with no data files:

import prcars as pc
from prcars.utils import synthetic_cars

wavenumbers, cars_raw, im_true = synthetic_cars(seed=0)

result = pc.retrieve(wavenumbers, cars_raw, method="kk")

print(result.im_chi3.shape)

The recovered im_chi3 signal is the Raman-like target extracted from the CARS/BCARS spectrum. Comparing it against im_true shows how well retrieval recovered the known ground truth.

With your own data:

import numpy as np
import prcars as pc

wavenumbers = np.load("wavenumbers.npy")
cars_raw = np.load("cars_spectrum.npy")

result = pc.retrieve(wavenumbers, cars_raw)
im_chi3 = result.im_chi3

---

Key features

Retrieval methods

• Kramers-Kronig phase retrieval
• Maximum Entropy Method retrieval
• Optional neural-network retrieval interface (experimental)

Background estimation: ALS, polynomial fitting, SNIP, rolling-ball Background correction: subtract, divide, square-root divide Denoising: Savitzky-Golay, Wiener, wavelet-based

Also: optional phase-matching correction, automatic phase correction with silent-region optimization, synthetic CARS example generation, benchmark utilities for comparing retrieval results, and a reusable Pipeline object for reproducible workflows.

---

Project status

prCARS is alpha-stage research software.

Component	Status
Kramers-Kronig retrieval	Implemented
MEM retrieval	Implemented
Background estimation	Implemented
Background correction	Implemented
Denoising utilities	Implemented
Synthetic CARS utility	Implemented
Benchmark helper utilities	Implemented
Unit tests	Implemented
GitHub Actions CI	Implemented
Citation metadata	Implemented
Neural-network retrieval interface	Experimental / optional
Changelog	Planned
Real-data validation workflow	Planned
Integration with CARSBench	Planned
Integration with CARSGuard	Planned
Full documentation site	Planned

---

Installation

git clone https://github.com/rhouhou/prCARS.git
cd prCARS

Create and activate a virtual environment:

python -m venv .venv

# macOS / Linux
source .venv/bin/activate

# Windows
.venv\Scripts\activate

Install the package in editable mode:

python -m pip install --upgrade pip setuptools wheel
python -m pip install -e .

Optional extras:

python -m pip install -e ".[dev]"       # development tools
python -m pip install -e ".[plot]"      # plotting helpers
python -m pip install -e ".[wavelet]"   # wavelet denoising
python -m pip install -e ".[torch]"     # optional PyTorch backend

For most local development:

python -m pip install -e ".[dev,plot,wavelet]"

On systems where the interpreter is python3, substitute python3 throughout.

Installation check

python -c "import prcars; print(prcars.__name__)"
python -m pytest

Expected output from the first command:

prcars

---

Retrieval methods

Kramers-Kronig

result = pc.retrieve(wavenumbers, cars_raw, method="kk")

With full control:

result = pc.retrieve(
    wavenumbers,
    cars_raw,
    method="kk",
    background="als",
    correction="divide",
    denoise="savgol",
    auto_phase=True,
    silent_region=(2700, 2730),
    retriever_kw={"zero_pad_factor": 8},
)

Maximum Entropy Method

result = pc.retrieve(
    wavenumbers,
    cars_raw,
    method="mem",
    background="snip",
    correction="divide",
    retriever_kw={
        "order": 128,
        "solver": "burg",
        "phase_method": "kk",
    },
)

Neural-network retrieval (experimental)

Requires an additional backend such as PyTorch or TensorFlow.

result = pc.retrieve(
    wavenumbers,
    cars_raw,
    method="nn",
    background="als",
    retriever_kw={
        "model_name": "cars_unet_v1",
        "backend": "torch",
    },
)

This interface is experimental and intended for future learned retrieval workflows.

---

Pipeline object

For reproducible workflows, apply the same configuration to many spectra:

pipeline = pc.Pipeline(
    method="kk",
    background="als",
    correction="divide",
    denoise="savgol",
    auto_phase=True,
    silent_region=(2700, 2730),
)

result_1 = pipeline.run(wavenumbers, spectrum_1)
result_2 = pipeline.run(wavenumbers, spectrum_2)

---

Preprocessing and retrieval options

Step	Options
Background estimation	als, polynomial, snip, rolling_ball, none
Background correction	subtract, divide, sqrt_divide, none
Denoising	savgol, wiener, wavelet, none
Retrieval	kk, mem, nn
Phase correction	automatic silent-region optimization

---

Benchmarking

Compare retrieval methods against a known synthetic target:

from prcars.utils import synthetic_cars, benchmark

wavenumbers, cars_raw, im_true = synthetic_cars(seed=0)

scores = benchmark(wavenumbers, cars_raw, im_true, methods=["kk", "mem"])

for method, values in scores.items():
    print(method, values)

These utilities are intended for sanity checks and method comparison, not as a substitute for validation on experimental data.

---

Typical workflow

Load CARS/BCARS spectrum
Preprocess / denoise signal
Estimate background
Apply background correction
Run phase retrieval
Apply optional phase correction
Inspect Raman-like output
Compare against reference or synthetic target

---

Repository structure

prCARS/
  examples/       Example scripts and usage demos
  prcars/
    __init__.py
    pipeline.py
    result.py
    methods/      kk.py, mem.py, nn.py
    corrections/  background.py, denoise.py, phase.py, phase_matching.py
    networks/     Optional neural-network model utilities
    utils/        Synthetic data, benchmarking, plotting helpers
  tests/          Unit and pipeline tests
  sanity_check.py
  pyproject.toml

---

Documentation

• docs/methods.md
• docs/preprocessing.md
• docs/examples.md
• docs/integration.md

---

The CARS/BCARS ecosystem

prCARS is the retrieval layer of a three-part workflow:

CARSBench  → simulate benchmark spectra under controlled domain shifts
prCARS     → retrieve Raman-like spectra
CARSGuard  → validate plausibility, consistency, and artifact risk

Project	Role
CARSBench	Simulates CARS/BCARS spectra under controlled domain shifts
prCARS	Retrieves Raman-like signals from CARS/BCARS spectra
CARSGuard	Validates spectra and retrieval outputs

---

Limitations

prCARS is research and educational software. Current limitations:

• it is not a substitute for experimental validation
• retrieval quality depends on preprocessing choices and input signal quality
• neural-network retrieval is experimental and optional
• real-data validation workflows are still planned
• results should be interpreted carefully when applied to real biological spectra

This project is not intended for clinical diagnosis, medical decision-making, or deployment in real healthcare settings.

---

Roadmap

• Improve test coverage for retrieval methods and correction utilities
• Expand CI with optional backend tests for PyTorch or TensorFlow
• Add a changelog and release notes
• Expand documentation pages for retrieval methods and preprocessing choices
• Add stronger synthetic benchmark reports
• Add integration examples with CARSBench and CARSGuard
• Add real CARS/BCARS data examples where licensing allows
• Improve neural-network retrieval documentation

---

Citation

If you use prCARS in research, education, or benchmarking work, please cite it using the metadata in CITATION.cff.

@misc{prcars2026,
  title={prCARS: Phase Retrieval and Raman-like Signal Reconstruction for CARS/BCARS Spectroscopy},
  author={Houhou, Rola},
  year={2026},
  note={Alpha research software},
  url={https://github.com/rhouhou/prCARS}
}

---

License

MIT. See LICENSE.

---

Part of my research on biophotonics and machine learning — biophotonics-ai.de