Bayesian-DSC

Bayesian-DSC is a small portfolio project that demonstrates Bayesian fitting of Dynamic Susceptibility Contrast (DSC) MRI concentration-time curves with CmdStanPy. The core design goal is to compile the Stan model once when the fitter is instantiated, then reuse that compiled executable for repeated curve fitting.

Features

• Uses a pre-compiled Stan gamma-variate model through CmdStanPy
• Exposes a backend-swappable BaseDSCFitter abstraction for future PyMC or NumPyro implementations
• Fits noisy DSC concentration-time curves with Bayesian inference
• Returns posterior means and 95% HDI bounds for A, alpha, beta, t0, sigma, CBV, and MTT
• Includes a reproducible synthetic-data simulation and plot export

Installation

1. Create a virtual environment in the project root:

python -m venv .venv

2. Activate it.

On Windows PowerShell:

.\.venv\Scripts\Activate.ps1

3. Install the Python dependencies:

pip install -r requirements.txt

4. Install the underlying CmdStan toolchain.

On Windows, the simplest option is to let cmdstanpy install the required compiler toolchain if needed:

python -m cmdstanpy.install_cmdstan -c

This downloads and builds CmdStan under the user CmdStan directory, typically C:\Users\<username>\.cmdstan.

Usage

Run the synthetic demonstration script from the project root:

python simulate_dsc.py

On your Windows setup, the easiest option is the included helper script, which configures the RTools toolchain before launching the demo:

.\run_simulation.ps1

The script will:

• generate a synthetic DSC concentration-time curve
• add Gaussian noise
• compile and sample the Stan model
• print true parameters, posterior means, and 95% HDI intervals
• save fit_result.png in the project root

If you prefer to avoid activating the environment explicitly on Windows, you can run:

.\.venv\Scripts\python simulate_dsc.py

If Stan compilation on Windows picks up the wrong C:\MinGW toolchain, prefer the helper script above because it prepends the correct RTools paths automatically.

Tests

Run the lightweight unit tests from the project root:

.\.venv\Scripts\python -m unittest discover -s tests

Project Structure

.
|-- .gitignore
|-- README.md
|-- requirements.txt
|-- run_simulation.ps1
|-- simulate_dsc.py
|-- tests
|   `-- test_bayesian_fitter.py
`-- src
    |-- __init__.py
    |-- bayesian_fitter.py
    `-- stan_models
        `-- gamma_variate.stan

Notes

• The first DSCBayesianFitter instantiation may take longer because Stan compilation happens at that point.
• Subsequent calls to .fit_curve(...) on the same fitter instance reuse the already compiled executable.
• If the Stan source is unchanged and the executable already exists, CmdStanPy can also reuse that executable on later runs.
• If compilation or sampling fails, the wrapper safely returns NaN outputs instead of crashing.
• The repository .gitignore excludes the local virtual environment, Python caches, CmdStan directories, and the generated fit_result.png.
• On Windows systems that already have an older C:\MinGW toolchain installed, make sure the RTools paths provided by cmdstanpy.install_cmdstan -c take precedence when compiling Stan models.