Radiotherapy Distortion QA Analysis

[RFranklin-MR]

This issue covers the addition of radiotherapy-specific geometric accuracy test analysis. The current geometric distortion and linearity test uses a grid of 9 points, positioned within 10cm of the isocentre. However, when the images will be used for radiotherapy treatment planning, it is valuable to quantify the image distortion over the whole patient volume. The recent IPEM Topical Report on use of MRI in EBRT treatment planning (Speight, et al., 2021) recommends measuring the gradient-field related geometric distortion “over the useful FOV of the scanner”, using a large phantom. In addition, this measurement should be performed “more frequently than annually until the stability of the MR scanner and the measurement procedure used has been demonstrated locally”. Therefore, automation of this QA will be valuable to the Physics team

Plan for the software

The phantom used for this test is called the “flat phantom”:

Which is custom-built locally. It measures approximately 60cm width x 30cm height x 1.5cm thickness, contains a grid of spheres with 25mm separation. In order to quantify the geometric accuracy at different positions within the bore, a full dataset will consist of 7 sets (at z = -75mm, -50mm, -25mm, isocentre, +25mm, 50mm, 75mm) of 10 slices (1.1m + 0.3mm gap). The geometric accuracy of the images, defined as magnitude difference between expected and imaged sphere locations (Wang, Strugnell, Cowin, Doddrell, & Slaughter, 2004), is extracted for each sphere at each z-location. This can be displayed as:

• scatter plots of distortion with distance from isocentre
• Median, SD and maximum distortion at different radial distances from the isocentre (0-5cm, 5-10cm, 10-15cm, 15-20cm, 20-25cm, 25-30cm)
• 2D contour maps of distortion.

It is recommended (Speight, et al., 2021) that the tolerance for distortion is “2mm within clinically relevant volume”, and at acceptance testing, “document volume where displacement is > 1 mm and 2 mm”. Similar tolerances are recommended by the ACR (<2mm, (ACR, 2018)) and AAPM (2%, (AAPM, 2010)

Design outline

• The location of grid of spheres from MR image, in DICOM co-ordinate system is determined
  • Likely using 3D version of current algorithm (threshold, binarise and verifying number of features) – fit sphere to grid of spheres or measure COM
  • The edges of the images will be cropped (given top always cropped off by distortion correction).
• The MR image grid will be initially aligned to the exact locations grid using a pre-determined shift, based on the DICOM coordinates of central grid points
  • The exact locations (based on CT) of the grid of spheres will be extracted offline, and stored in a config file in Hazen.
• A rigid registration will then be performed using the central 10cm of grid points.
• The distortion (distance) of each grid point between MR image and exact location will be calculated
• This will be output as:
  • scatter plots (maybe a contour plot) with the graph option
  • max, median and SD for distortion within set radial distances from isocentre
  • distortion for each grid point with the verbose option
• To build a map of distortion at different z-position, images are grouped according to sequence name (string at the end indicates bed position) and analysed as a group. The z-position is extracted from DICOM coordinates, and the analysis outputs reported for each z-position

.

[hshuaib90]

Thanks, @RFranklin-MR - looks good to me. Couple of things to consider in implementation:

• [ ] Robust definition of 'clinically relevant volume'
• [ ] Making sure initial image crop doesn't include clinically relevant volume
• [ ] Exact locations of spheres can be stored at top of file as a global variable rather than dealing with separate file probably
• [ ] Does analysis need all z-positions or just one. Can it be smart?
• [ ] Plot to include ROI where displacement exceeds 1 and 2 mm

[Robert-Johnstone]

Hi @RFranklin-MR - this is me replying, as you suggested, so that you can tag me on future updates.

[laurencejackson]

@RFranklin-MR as discussed please can you update on this issue?

[RFranklin-MR]

This functionality is desired for Radiotherapy QA with the flat phantom. Oscar will be working on this going forwards.

[RFranklin-MR]

This has been implemented outside of hazen, so closing.