Craniospinal irradiation (CSI) is used when there is suspected widespread dissemination of disease

throughout the central nervous system (CNS). Pediatric patients with medulloblastoma are the most
frequently seen patients who undergo this particular treatment. Historically, patients were treated
primarily in a prone position, with two opposed lateral fields for the brain treatment and 1 (or several,
depending upon the length of the PTV) spinal field, matched to the brain fields. To mitigate the over- or
under-dosing typically seen at field junctions within a patient, the junctions were feathered so as to
create a smoother dose gradient. Although I have seen many whole-CNS treatments previously, I
haven’t planned one. Clinically, our department treats few of these types of patients, and our most
recent patient was treated using a mixed-technique approach with forced dose gradients across the
brain-spinal field junction using IMRT fields at the junction and 3D fields elsewhere. As a department,
we are still not completely comfortable using this technique, so I first decided to try a more traditional
3DCRT plan on a supine patient.

The ProKnow website provided datasets of both supine and prone patients for whole-CNS planning. I
chose the patient in the supine position as supine set-ups are typically more reproducible and stable. In
my previous clinic, we tried to avoid anesthesia for irradiation wherever possible. Due to the long
treatment times associated with CSI, pediatric patients were much more comfortable and less likely to
move when in the supine position – where they were able to watch a movie projected above the linac.
ProKnow also provided ideal and mandatory goals and constraints for the brain and spinal PTVs, and
associated OAR.1

Within my department, we have an in-house calculator for matching brain and spinal fields, and I used
this initially to determine my field placement.2

Our in-house craniospinal field matching calculator. The isocenter coordinates and superior spine jaw
position are entered, and the calculator returns the required brain field jaw positions, and couch and
collimator rotations necessary for a perfect match of field divergence.
Using the results from the calculator, I positioned my isocentric fields and ran a quick 3D field-in-field
plan for the brain PTV, to ensure that I could achieve the desired coverage and constraints with this
technique. The prescription for the patient was 36 Gy in 20 fractions at 1.8 Gy per fraction. Since the
optic nerves extended into the PTV and the maximum tolerated dose was less than the prescription
dose, it was necessary to carefully control the dose to the PTV in this area. The superior and posterior
parts of the brain fields were left open to ensure there was enough flash to compensate for any small
variations in treatment position.

Field information for the 3DCRT matched field technique.

My spinal field needed only one isocenter as I was able to fully encompass the length of the PTV at
extended SSD while respecting machine limitations. I chose the junction site so that the lateral brain
fields would not enter through the shoulders, and the posterior spinal field would not exit through the
mandible. This allowed me a 4cm area to feather the fields. After calculating an initial open spinal field, I
soon realized the inherent difficulties with this technique. Achieving the necessary depth for adequate
PTV dose-coverage increased the exit dose through the patient, violating the esophagus, thyroid, and
bowel constraints. There was also a large area of 110% of the prescription dose posterior to the spinal
column, through the entire length of the patient. Changing the energy from 6MV to 18MV decreased
the area of 110%, but resulted in increased exit dose. It also resulted in difficulty in covering the most
inferior part of the PTV at the cauda equina area, as it was very superficial relative to the dmax depth of
an 18MV beam.
Figure depicting the isodose coverage from a single spinal field.

At this time, I was also reading an article that compared 5 different techniques for craniospinal
irradiation: 3DCRT photons, IMRT photons, VMAT photons, Tomotherapy photons, and pencil-beam
scanning protons.3 The article included a side-by-side comparison of the dose distribution for each of
these techniques within the same patient. The researchers also evaluated the techniques using a range
of metrics and indices, such as PTV coverage, homogeneity and conformity indices, and OAR sparing.
Their results showed that with the exceptions of protons, VMAT produced the best plan quality overall. I
therefore decided to try a VMAT approach instead of the more traditional 3DCRT technique, so that I
could more easily meet the ProKnow plan objectives.
Figure taken from Seravalli et al.3

The brain PTV and spine PTV were combined to make a single PTV. Three isocenters were chosen, one
each for the brain, thoracic spine, and lumber spine, all on the same lateral and vertical planes to reduce
the amount of shifts necessary on a treatment machine. They were also chosen to ensure that each arc
field overlapped by a minimum of 4cm, as recommended in research articles concerning junction dose
gradients with IMRT.4 The overlapping fields take advantage of Eclipse’s auto-feathering tool wherein
the optimizer attempts to create a smooth, homogenous dose distribution across the overlapped fields.

Isocenter positions throughout the patient.

Patients undergoing craniospinal irradiation in the supine position are simulated with a shell
immobilizing the head, and shoulders if the immobilization equipment allows. The neck is
hyperextended to allow for optimal beam placement, avoiding the mandible and oral cavity due to
concerns about the long-term effects of irradiating this area in pediatric patients. This position is evident
in the image below.

Patient position and planned arcs.

6 full arcs were used – 1 clockwise and 1 counter-clockwise for each isocenter. The collimators were
rotated 15 degrees from gantry 0 in opposing directions for each arc-pair. There was no couch rotation
necessary, as the arcs did not need to be matched to one another with regards to field divergence. The Y
field jaws were opened enough to provide a minimum of 4cm overlap between arcs. The field sizes were
modified slightly after an initial run so that the thoracic and lumber arc overlap wasn’t directly over the
kidneys. This was done so as to minimize the modulation required within that area to both reduce dose
to the kidneys as well as create a smooth homogenous dose across the arc junctions. 6 MV was chosen
due to the small patient separation.
Parameters for each of the 6 arcs.

Avoidance sectors for each spinal arc were added to minimize entry dose in the patient’s arms, which
were by his side. This also allowed for a reduction in entry dose to the kidneys. The use of 2 full arcs for
each isocenter, rather than 1, was designed to compensate for the avoidance sectors, allowing more
space for modulation solutions within the optimizer.

Avoidance sector parameters for each of the spinal arcs.

The following optimization objectives were used, in accordance with the dose criteria and OAR
constraints issued by ProKnow for this particular patient. The combined brain and spine PTV was used
for dose evaluation and optimization, in addition to the other target structures – individual brain and
spine CTVs and PTVs.
Optimization objectives for the 3-isocenter VMAT plan.

Point, mean and gEUD constraints were used for the critical structures. From the 3DCRT attempt, I knew
the optic nerves and lenses would need a higher priority in the optimization, and I also created an
anterior avoidance structure to minimize the entry and overall dose in this area. Although I am relatively
new to using gEUD optimization parameters, I chose to utilize this feature in order to reduce the number
of objectives within the optimizer. The fewer objectives there are, the more solutions the optimizer can
evaluate before convergence.

I ran through the first two levels, tweaking the optimization parameters as I did so. I found that the
kidney constraint was very difficult to meet. I therefore turned off my NTO so that the optimizer could
squeeze the dose away from the kidneys and into other normal structures and tissue. With the NTO
enabled, it was causing conflict within the optimizer.
Optimization window, showing the plan objectives.

I also noticed that I was having areas of heterogenous dose around the junction sites.
The areas of dose heterogeneity.

The Eclipse v15.5 treatment planning manual states5:

From Eclipse Treatment Planning 15.5 MR1 Customer Release Note.

I tried splitting the PTVs into smaller sections at each junction and ran a quick optimization with those
PTVs instead – however, I did not notice enough of an improvement in dose to continue using this
technique. With more time, I would have explored separating the PTVs at other levels within the arc
overlaps, or just creating one split to see if that resulted in any improvement.

Image depicting the separation of PTVs at the junction sites.

Although I tried successive attempts at improving the mean kidney dose, I was unsuccessful. Any
reduction in kidney dose resulted in a colder plan overall, meaning that once normalized to the
combined PTV volume, the plan maximum dose exceeded 113%. The maximum dose in my final plan is
113% and is within the PTV. Recent studies demonstrate that excessive dose to the vertebrae in
pediatric CSI patients can affect later growth and spine mobility, although there are no
recommendations as to what excessive dose may be in any particular population. I therefore opted to
keep my plan dose under 113% of the prescription, at the cost of the mean kidney dose. The ideal
maximum dose was 110%, and I am uncomfortable going higher than that. With less restrictive OAR
constraints, the maximum dose wouldn’t be as much of an issue, but in any case, I would seek physician
input before proceeding with a plan that has a maximum dose of greater than 107% of the prescription
dose. The mean kidney dose was close to the ideal tolerance.

The plan was normalized so that 95% of the combined PTV structure received 100% of the prescription
dose. The individual brain and spine PTVs were then evaluated to ensure that they each also met the
D95 goal. Due to differences in dose binning and calculation between Eclipse and ProKnow, I actually
normalized to 95.2% volume receiving 100% of the dose. This is just to ensure that the plan passes the
target criteria when uploaded to the ProKnow site.
Final dose distribution.
I created a scorecard within Eclipse to evaluate the plan objectives. In addition to exceeding the
maximum plan dose and the kidney mean dose, one of my orbital lenses exceeded the maximum
tolerance dose of 7 Gy. I was not concerned with this clinically, due to the relatively benign endpoint of
the lens, although I was conscious about the effects of cataracts on pediatric patients.

Scorecard, evaluating the plan with regards to targets and constraints.

ProKnow scorecard.
Final plan DVH.
As an interesting note – my final plan was not the first one that I submitted. The maximum dose of my
initial plan, also a 6 arc 3 isocenter VMAT plan, was under 110% of the prescription dose, and met all
organ constraints with the exception of the maximum dose to the lenses and mean dose to the kidneys.
The plan was close to meeting these tolerances, and I felt that clinically, this plan was of sufficient
quality to treat.

My Eclipse scorecard, showing the plan criteria and results.

However, when importing into the ProKnow website, the plan quality results look quite different than
they did in Eclipse. In ProKnow, my lens doses are approximately 2 Gy higher than reported by Eclipse,
and my optic nerves are 1 Gy higher, which pushes them beyond the ideal tolerance dose. I discovered
that Eclipse segmentation rendering consistently underestimates the volumes of structures. On smaller
structures such as the lenses and optic nerves, and in areas of steep dose gradient, this difference can
have a significant impact on the doses – as seen within my initial plan. This makes me uncomfortable
using an Eclipse-generated plan to treat a site where OAR might be close to tolerance, as in this case.
Firstly, the OAR may be exceeding the tolerance dose unbeknownst by the planner and physicist, and
secondly, in areas of steep-dose gradient, this complicates the process of image-matching and accurate
set-up.

Treating 3 isocenters is difficult to implement clinically, due to imaging safeguards in place on the
treatment machines. It requires the separation of arcs into different plans that are treated sequentially.
There exists great potential for error in shifting, imaging, and setup – especially if the therapy team is
unfamiliar with the technique. For this reason, I would not be comfortable with this plan being treated
clinically, unless there was a significant advantage over 3DCRT. As a future exercise, without any time
constraints, I would like to complete the 3DCRT plan that I started and compare it to the VMAT plan –
particularly to see if the 3DCRT plan dose experiences the same issues when uploaded to ProKnow.
Perhaps the typical VMAT dose distribution is more sensitive to the contouring underestimation in
Eclipse.
Scorecard from ProKnow, showing different values to those reported by Eclipse.
References:

1. ProKnow Systems. 3. https://proknowsystems.com/planning/studies. Published 2019. Accessed

October 13, 2019.

2. Ochsner Health System, New Orleans. Internal document. Updated April 2019.

3. Seravalli E, Bosman M, Lassen-Ramshad Y et al. Dosimetric comparison of five different

techniques for craniospinal irradiation across 15 European centers: analysis on behalf of the SIOP-E-
BTG (radiotherapy working group). Acta Oncol (Madr). 2018;57(9):1240-1249.
doi:10.1080/0284186x.2018.1465588

4. Hadley A, Ding G. A single-gradient junction technique to replace multiple-junction shifts for

craniospinal irradiation treatment. Medical Dosimetry. 2014;39(4):314-319.
doi:10.1016/j.meddos.2014.05.004

5. Eclipse Treatment Planning Customer Release Note 15.5 MR1. Varian Medical Systems. Palo Alto. First
published 2017.