Proton CT#

TIGRE includes a MATLAB extension, written in CUDA, that turns proton CT measurements into optimised proton radiographies, following the method of Collins-Fekete et al. (2016). Protons scatter as they cross the object, so their paths are not straight lines. The extension estimates each proton’s path with a cubic spline and bins its water-equivalent path length (WEPL) along that path.

The radiographies can then be reconstructed with TIGRE’s usual algorithms to give a full proton CT image, or used on their own. A Python version is not available yet.

Building the extension#

From TIGRE’s MATLAB folder, run:

CompilePCT

It asks for a pixel length and an intercepts vector size; press Enter to keep the defaults. If the pixel lengths in x and y differ, enter the smaller one. When the code runs, it reports how many protons it rejected because of the intercepts vector size.

Input data#

Load the measured or simulated data in any format MATLAB reads (.mat or raw binary files work well). All lengths must be in millimetres.

Parameter

Name in the demo

Unit

Water-equivalent path length

WEPL

mm

Entry position, x and y

posIn

mm

Exit position, x and y

posOut

mm

Entry direction, x and y

dirIn

mm, with the z component equal to 1

Exit direction, x and y

dirOut

mm, with the z component equal to 1

Before computing the radiography, remove protons with large angular or energy deviations from the data set (3-sigma cuts), or the result will not be reliable.

Computing a radiography#

proj = pCTCubicSpline_mex(posIn, posOut, dirIn, dirOut, Wepl, eIn, geo);
  • Wepl is a vector of length N with the WEPL of each proton.

  • posIn, posOut, dirIn and dirOut are vectors of length 2N: first the x values of all protons, then the y values.

  • eIn is the initial beam energy, in MeV.

geo uses these fields:

Field

Meaning

Unit

geo.dDetector

Pixel size of the radiography

mm

geo.sDetector

Size of the radiography in x and y (two values)

mm

geo.DSO

Distance from the source to the origin (positive)

mm

geo.DSID

Distance from the source to the entry detector (positive)

mm

geo.DSD

Distance from the source to the exit detector (positive)

mm

geo.mode

'parallel' or 'cone'

geo.hull

Convex hull parameters [a, b, alpha, h]

mm

The origin is at the centre of the object. Set h = 0 to compute without a convex hull. Otherwise the hull is the elliptical cylinder

\[\frac{(x\cos\alpha - z\sin\alpha)^2}{a^2} + \frac{(x\sin\alpha + z\cos\alpha)^2}{b^2} = 1, \qquad -\frac{h}{2} \le y \le \frac{h}{2},\]

that is, an ellipse rotated by \(\alpha\) around the y axis.

Testing#

MATLAB/Demos/pRad_demo.m computes one optimised radiography in parallel geometry from a simulated test data set (loaded with pCTdata, in MATLAB/Test_data/pCT_test_data). Run it to check that the extension works on your system.

Notes#

  • If CompilePCT fails with missing symbols, check that the libstdc++ shipped with MATLAB is compatible with your CUDA version and your system’s gcc.

  • The extension was tested on Debian 10 with CUDA 11.2, an NVIDIA RTX 4000, and MATLAB R2020b and R2021b.