Skip to content

PHANTOM User Manual

Jump to bottom
Choonsik Lee edited this page Jul 2, 2026 · 19 revisions

PHANTOM — NCI Computational Human Phantom Libraries

Introduction

PHANTOM is a collection of reference-grade computational human phantom libraries developed by physicists at the National Cancer Institute (NCI) in collaboration with external partners. These phantoms are designed to support population-based radiation dose estimation, benchmarking, and research or regulatory-facing analyses, and serve as the anatomical foundation for NCI dose tools (NCICT, NCINM, and NCIRF).

The PHANTOM libraries represent reference and body size–dependent pediatric, adult, and pregnant populations and are intended for Monte Carlo–based radiation transport simulations, rather than patient-specific clinical modeling.

Intended use
PHANTOM libraries are intended for reference dosimetry and computational research.
They are not intended for patient-specific clinical treatment planning or site-customized anatomical modeling.

Available Phantom Libraries

The PHANTOM libraries currently available are summarized below.
A cross mark (×) indicates availability in the current release folder.

File format UF/NCI reference size phantoms UF/NCI body size-dependent phantoms1 ICRP reference phantoms UF/NCI pregnant women phantoms
Number of phantoms 12 362 12 8
Arms2 w and w/o w and w/o w and w/o w and w/o
Binary voxel (legacy) × × (low res) Available from ICRP 110 and 143
NIfTI (.nii.gz) × × (incl. xy-low/z-high) ×
DICOMRT3 × × ×
MC input available MCNP (w arms) Geant4 (w/o arms)
Reference Lee 2010 Geyer 2014 ICRP 110, ICRP 143 Maynard 2014

1 The reference-size, body size–dependent, and pregnant woman phantom libraries are available in compressed NIfTI (.nii.gz) format, the recommended format for new downloads. Legacy binary voxel files are retained under each library's bin/ folder for compatibility with existing Monte Carlo workflows.

2 Phantoms with the arms removed are intended for imaging geometries in which the patient’s arms are raised.

3 DICOMRT files are provided for selected supported libraries under each library's dicomrt folder, e.g. nci_reference/armless_highres/dicomrt, icrp_reference/armless_highres/dicomrt, and nci_pregnant/arm_highres/dicomrt. DICOM-CT and RT STRUCTURE files can be directly imported into Treatment Planning Systems (TPS).
Recommended citation for DICOMRT files: Griffin 2019

Folder Naming Convention

The PHANTOM download directory uses a phantom-library folder first, then a posture/resolution subfolder, then a file-format subfolder:

III_SSS/AAA_RRR/format

  • nci_size/arm_highres/niigz, nci_size/armless_lowres/{bin, niigz}
    Body size–dependent phantom folders. armless_xylow_zhigh is a mixed resolution (xy-plane at low resolution, z-axis at high resolution).

  • nci_reference/arm_highres/{bin, niigz, mc-input}, nci_reference/armless_highres/{bin, dicomrt}
    Reference-size phantom folders.

  • nci_pregnant/arm_highres/{niigz, dicomrt}
    Pregnant woman phantom folders.

  • icrp_reference/armless_highres/{dicomrt, mc-input}
    ICRP reference phantom folders.

Not every posture/resolution folder includes every format; see Available Phantom Libraries above for what is available per library.

Monte Carlo input folders use code-specific names rather than the voxel-library naming pattern: mcnp-nci-reference-arm (under nci_reference/arm_highres/mc-input) and geant4-icrp-noarm (under icrp_reference/armless_highres/mc-input).

The naming fields are:

  • III_SSS: phantom library, one of nci_size, nci_reference, nci_pregnant, icrp_reference
  • AAA_RRR: arm posture and voxel resolution, e.g. arm_highres, armless_lowres, armless_xylow_zhigh
  • format: one of bin, niigz, dicomrt, mc-input

Master Table

The file _mastertable_ref&size.xlsx contains anatomical and material data required for Monte Carlo radiation transport simulations. The file includes the following tabs:

  • about Contents of this master table

  • summary_arm
    Organ voxel tags, organ names, and material card identifiers for UF/NCI phantoms with arms attached

  • density_reference
    Organ and tissue density for the reference size phantoms

  • material
    Material identifiers, material types, elemental compositions, and mass fractions

  • dose response function
    Photon dose response functions (absorbed dose per unit fluence) for active marrow and endosteum

  • marrow_fraction
    Fractions of active marrow and endosteum assigned to individual skeletal sites

These tables ensure consistent, reproducible mapping between voxelized anatomy, material definitions, and dosimetric response functions across NCI dose tools.

Monte Carlo Input Files

Monte Carlo input files are available under an mc-input subfolder within the relevant library folder in the PHANTOM download directory. These files are code-specific resources for selected established reference workflows.

The current release includes:

  • nci_reference/arm_highres/mc-input/mcnp-nci-reference-arm/ MCNP input decks for the NCI reference-size phantoms with arms. This folder contains 12 numbered input files (01–12) and 12 corresponding lattice files (*.lat).

  • icrp_reference/armless_highres/mc-input/geant4-icrp-noarm/ Geant4 input package for ICRP reference phantoms without arms. This folder contains Ref_noa_icrp_bin_G4.tar.gz.

These inputs are provided as reference starting points for established Monte Carlo workflows. Users should review and modify local file paths, source definitions, scoring setup, compiler settings, and code-version assumptions before running simulations.

NIfTI (.nii.gz) Voxel Phantom Files

NIfTI (.nii.gz) is the recommended format for newly downloaded voxel phantom files. Each file is a gzip-compressed NIfTI image containing the 3D integer voxel-label array for one phantom. The voxel values correspond to organ or tissue tag numbers used by the PHANTOM master table.

The current NIfTI release is located under each library's niigz subfolder in the PHANTOM download directory:

Reference-size phantoms:

  • nci_reference/arm_highres/niigz

Body size–dependent phantoms:

  • nci_size/arm_highres/niigz
  • nci_size/arm_lowres/niigz
  • nci_size/armless_highres/niigz
  • nci_size/armless_lowres/niigz
  • nci_size/armless_xylow_zhigh/niigz — xy-plane at low resolution, z-axis at high resolution

Pregnant woman phantoms:

  • nci_pregnant/arm_highres/niigz

Legacy binary voxel files are retained under each library's bin subfolder. NIfTI was adopted as the recommended distribution format for several reasons:

  • NIfTI is a standard medical-imaging format that can be opened directly by common tools such as 3D Slicer, ImageJ/Fiji, ITK-SNAP, MATLAB, and Python packages.
  • NIfTI headers store image dimensions, voxel spacing, data type, and orientation information, while raw binary files require separate metadata to be interpreted correctly.
  • Compressed NIfTI files are smaller and easier to transfer than the corresponding raw binary voxel files.
  • In previous releases, the raw binary file size was too large to practically release the full high-resolution NCI body size–dependent phantom library (n = 362) via Google Drive. The compressed .nii.gz format makes this high-resolution release practical while preserving the integer organ-label voxel data.

What Information Is Included

Each NIfTI file stores:

  • 3D voxel-label data The image array contains integer organ or tissue labels. A voxel value should be interpreted as a label ID, not as CT number, attenuation, or density.

  • Image dimensions The number of voxels along each image axis is stored in the NIfTI header.

  • Voxel spacing The physical voxel size is stored in the NIfTI header and can be read by standard NIfTI-compatible software.

  • Data type The label array is stored as an integer image. When processing the files, preserve the integer labels and avoid interpolation unless a label-preserving method is used.

  • Spatial orientation information NIfTI headers include affine/orientation information used by visualization and image processing software.

The NIfTI file does not replace the master table. Organ names, material assignments, tissue densities, elemental compositions, and marrow or dose-response data should be obtained from _mastertable_ref&size.xlsx.

Reading NIfTI Files

3D Slicer

  1. Open 3D Slicer.
  2. Use Add Data or drag the .nii.gz file into the Slicer window.
  3. Load the file as a volume. For label-based visualization, set or convert the volume to a label map as needed.
  4. Use the volume display or segmentation tools to inspect individual label regions.

Because PHANTOM files are label images, avoid smoothing or linear interpolation when resampling. Use nearest-neighbor interpolation for any label-preserving operation.

ImageJ / Fiji

  1. Open ImageJ or Fiji.
  2. Import the .nii.gz file using a NIfTI-compatible importer, such as the NIfTI plugin or Bio-Formats importer, depending on the local installation.
  3. Inspect the image stack as a 3D label volume.
  4. Use thresholding or label-value selection to isolate a specific organ ID.

When using ImageJ/Fiji, keep the image as an integer label image where possible. Some operations may convert the data to floating point; this is acceptable for display but should be avoided for saving label-preserving phantom data.

ITK-SNAP

  1. Open ITK-SNAP.
  2. Load the .nii.gz file as the main image.
  3. Use the label-inspection tools to identify voxel values and inspect organ regions.

Python

The recommended Python package for reading NIfTI files is nibabel.

import nibabel as nib
import numpy as np

nii_path = "30f165060_armless_highres.nii.gz"
img = nib.load(nii_path)

# Preserve integer labels. Avoid get_fdata() unless floating-point data are desired.
labels = np.asanyarray(img.dataobj)

print("shape:", labels.shape)
print("voxel spacing:", img.header.get_zooms()[:3])
print("data type:", labels.dtype)
print("affine:")
print(img.affine)

organ_id = 125
organ_mask = labels == organ_id
print("organ voxel count:", int(organ_mask.sum()))

MATLAB

MATLAB can read NIfTI files with niftiread and inspect header metadata with niftiinfo.

info = niftiinfo("30f165060_armless_highres.nii.gz");
labels = niftiread(info);

disp(size(labels))
disp(info.PixelDimensions)
disp(class(labels))

organ_id = 125;
organ_mask = labels == organ_id;
nnz(organ_mask)

Converting Legacy Binary Voxel Files to NIfTI

Legacy binary voxel files are retained under each library's bin folder for users who need to reproduce previous workflows. These files are raw arrays and do not carry their own metadata. To convert a binary file to NIfTI, the following information must be known from the phantom documentation or master table:

  • array dimensions, e.g., number of voxels in x, y, and z
  • voxel spacing in physical units
  • integer data type, such as uint16 or int32
  • byte order
  • array storage order and orientation

Example Python conversion:

import nibabel as nib
import numpy as np

binary_path = "phantom.raw"
nii_path = "phantom.nii.gz"

# Replace these values with the dimensions and voxel spacing for the phantom.
shape = (512, 256, 180)       # x, y, z
spacing = (1.0, 1.0, 1.0)     # mm
dtype = np.dtype("<u2")       # little-endian unsigned 16-bit integer

data = np.fromfile(binary_path, dtype=dtype)
data = data.reshape(shape, order="C")

affine = np.diag([spacing[0], spacing[1], spacing[2], 1.0])
img = nib.Nifti1Image(data, affine)
img.header.set_data_dtype(data.dtype)
nib.save(img, nii_path)

The shape, spacing, dtype, and array order must match the original binary phantom. If these values are incorrect, the converted NIfTI file may open but the anatomical labels can be transposed, flipped, or assigned to the wrong physical locations.

Converting NIfTI Back to Raw Binary

Some Monte Carlo pipelines may still require raw binary input. A NIfTI label image can be exported back to a raw binary array after confirming the required data type and array order for the target code.

import nibabel as nib
import numpy as np

img = nib.load("phantom.nii.gz")
labels = np.asanyarray(img.dataobj)

# Match the data type expected by the receiving Monte Carlo workflow.
labels = labels.astype(np.uint16, copy=False)
labels.tofile("phantom.raw")

After conversion, confirm the voxel count, unique label values, and organ-label mapping against _mastertable_ref&size.xlsx before using the file for dose calculation.

Features Under Development

The following enhancements are under active development and may be released in future versions:

  • Additional organ substructure models beyond the current cardiac substructure release (e.g., brain substructures)
  • Implementation of lymphatic node models in the body size–dependent phantom library

Clone this wiki locally