simind-python-connector

Python SIMIND Monte Carlo Connector.

A Python toolkit that lets you run SIMIND from Python and use the outputs in common reconstruction ecosystems (STIR, SIRF, PyTomography).

Disclaimer

This project is an independent Python connector toolkit and is not affiliated with, endorsed by, or maintained by the SIMIND project or Lund University.

SIMIND is not distributed with this package. You must separately obtain and install a licensed SIMIND installation.

Quick Links

• Full Documentation
• Installation
• Backend Support - adaptor dependency matrix
• Reconstruction with SIMIND corrections

What This Package Does

1. Runs SIMIND from Python with a minimal, explicit API.
2. Adapts SIMIND data for widely used Python reconstruction packages.

Key Features

• Connector-first API - SimindPythonConnector for direct SIMIND execution from Python
• Package Adaptors - STIR/SIRF/PyTomography adaptors for reconstruction workflows
• Native reconstruction workflows - Use STIR/SIRF/PyTomography reconstruction tools with generated SIMIND data
• Reconstruction with SIMIND corrections (optional) - simind_python_connector.recon: Monte Carlo scatter and Fu & Qi residual correction in SIRF OSEM and CIL algorithms
• Dual scoring support - SCATTWIN and PENETRATE
• DICOM builders - DICOM-driven setup utilities for scanner/input preparation
• Advanced Schneider2000 density conversion - 44-segment HU-to-density mapping

Installation

Basic Installation

pip install simind-python-connector

Import with:

import simind_python_connector

For example plotting dependencies:

pip install "simind-python-connector[examples]"

Adaptor Dependencies

SimindPythonConnector works without SIRF/STIR/PyTomography.

Install optional packages only for the adaptor paths you need:

• STIR Python for StirSimindAdaptor workflows (example 07A)
• SIRF for SirfSimindAdaptor workflows (example 07B)
• PyTomography for PyTomographySimindAdaptor workflows (example 07C)

Option 1: STIR Python (Recommended for basic usage)

Install from conda:

conda install -c conda-forge stir "numpy<2.0"

Note: If STIR from conda-forge fails with NumPy binary-compatibility errors, install numpy<2.

Or build from source:

git clone https://github.com/UCL/STIR.git
cd STIR
# Follow build instructions in the repository

Option 2: SIRF

Install from source:

git clone https://github.com/SyneRBI/SIRF.git
cd SIRF
# Follow build instructions in the repository

Note: SIRF includes STIR, so a separate STIR install is usually unnecessary.

Option 3: PyTomography

Install PyTomography for the PyTomography adaptor workflow:

pip install pytomography

SIMIND Requirement

SIMIND is not included with this package and must be installed separately.

Use the official SIMIND resources:

• SIMIND site (Medical Radiation Physics, Lund University): https://www.msf.lu.se/en/research/simind-monte-carlo-program
• SIMIND manual/docs: https://www.msf.lu.se/en/research/simind-monte-carlo-program/manual

For local use with this package's scripts, place your SIMIND installation under:

./simind

and make sure the executable is available at:

./simind/simind

and ensure SIMIND data files are available under:

./simind/smc_dir/

The Docker helper scripts use this layout and automatically wire SIMIND paths when the binary exists at ./simind/simind.

For direct local runs outside the repository helpers, make sure the SIMIND executable is available as simind on PATH.

Quick Start

import numpy as np
from simind_python_connector import SimindPythonConnector
from simind_python_connector.configs import get
from simind_python_connector.core.types import DEFAULT_QUANTIZATION_SCALE

source = np.zeros((32, 32, 32), dtype=np.float32)  # z, y, x
source[12:20, 12:20, 12:20] = 1.0
mu_map = np.zeros_like(source)
mu_map[source > 0] = 0.15

connector = SimindPythonConnector(
    config_source=get("Example.yaml"),
    output_dir="output/basic",
    output_prefix="case01",
    quantization_scale=0.05 * DEFAULT_QUANTIZATION_SCALE,
)

connector.configure_voxel_phantom(
    source=source,
    mu_map=mu_map,
    voxel_size_mm=4.0,
)
connector.set_energy_windows([126], [154], [0])  # Tc-99m ± 10%
connector.add_runtime_switch("FI", "tc99m")
connector.add_runtime_switch("CC", "ma-lehr")
connector.add_runtime_switch("NN", 1)
connector.add_runtime_switch("RR", 12345)

outputs = connector.run()
total = outputs["tot_w1"].projection
print(total.shape)

Advanced Density Conversion

from simind_python_connector.converters.attenuation import hu_to_density_schneider
import numpy as np

# Convert HU image to densities using Schneider2000 model
hu_image = np.array([[-1000, 0, 500], [800, 1200, 2000]])
density_map = hu_to_density_schneider(hu_image)  # 44-segment clinical model

Pure Python Connector (NumPy Outputs)

from simind_python_connector import RuntimeOperator, SimindPythonConnector
from simind_python_connector.configs import get

connector = SimindPythonConnector(
    config_source=get("AnyScan.yaml"),
    output_dir="output/python_connector",
    output_prefix="case01",
    quantization_scale=1.0,
)

outputs = connector.run(RuntimeOperator(switches={"NN": 1, "RR": 12345}))
total = outputs["tot_w1"]
print(total.projection.shape)
print(total.header_path)

For minimal toy runs with smaller projection/image settings, use get("Example.yaml").

quantization_scale controls source integer quantization before writing SIMIND .smi files:

• the default (500 / 32767) preserves the legacy 500-weight source budget;
• 1.0 uses the full signed 16-bit source range (32767), the best numeric precision;
• < 1.0 reduces the integer scale for faster toy runs but increases rounding error.

SIMIND reads the source and density maps as signed 16-bit integers, so values of 32768 and above wrap negative and silently corrupt the map; the connector caps both maps at 32767. Explicit quantization_scale values are used as given, with no automatic migration of scales calibrated for the former 500-weight maximum: rescale an old scale with old_scale * DEFAULT_QUANTIZATION_SCALE. SIMIND treats source maps as integer weights; absolute activity/time scaling is controlled separately by simulation settings.

Docker Environments

Dedicated container definitions are provided in docker/:

• docker/stir/Dockerfile
• docker/sirf/Dockerfile
• docker/pytomography/Dockerfile

Use Docker Compose:

docker compose -f docker/compose.yaml build stir
docker compose -f docker/compose.yaml run --rm stir

Run per-backend validation and separated examples:

bash scripts/run_container_validation.sh
bash scripts/run_container_validation.sh --with-simind
bash scripts/run_container_examples.sh

Contributing

Please see the Contributing Guide.

License

Apache License 2.0