Carlos Torres Teran
Clinical Practicum III
CSI ProKnow Paper
Craniospinal irradiation (CSI) is a treatment planning technique used to treat cancer
seeding along the entire length of cerebral spinal fluid (CSF). Orthogonal junctions between
lateral brain fields and posterior spine fields treat the entire craniospinal axis, making use of
divergence on the skin at the point of intersection of the field borders.1 The goal is to treat CSF
to a uniform dose while reducing hot and cold spots that can manifest near field match lines
between the cranial and spine fields due to incorrect field matching. Exaggerated gaps or
overlaps between the orthogonal fields can lead to underdosed or overdosed target volumes.
Common diseases associated with CSI include medulloblastoma, ependymoma, germinoma,
infratentorial glioblastoma, and leptomeningeal carcinomas.2
The CSI ProKnow patient was positioned head-first supine in the planning computed
tomography (CT) scan and provided two planning target volumes (PTV) in the brain and spine
prescribed to 3600 cGy in 20 fractions. To meet the designated ProKnow metrics, I ultimately
combined two forms of Memorial Sloan Kettering Cancer Center (MSK)-approved planning
techniques: a three-dimensional conformal radiation therapy (3D-CRT) plan for the brain and an
intensity modulated radiation therapy (IMRT) plan for the spine.
Cranial Plan
At MSK, CSI has traditionally been treated using a 3D-CRT technique. After simulation,
the radiation oncologist delineates the target volumes, cribriform plate, eyes, and lenses for
treatment and shielding purposes. The cribriform plate is of particular importance due to its
association with medulloblastoma recurrence.3 A treatment plan is designed by the medical
dosimetrist to treat the cranium and upper cervical spine with opposed lateral fields, and the
lower cervical, thoracic, and lumbar spine with 1-2 posterior fields depending on the length of
the spine (whether greater than 38 cm). The medical dosimetrist plans out the 1 cm shifting of
the field junctions (feathering) that the radiation therapists do every 5 fractions by creating a
phase plan per field junction shift.
In addition to our standard 3D-CRT for CSI treatments, MSK is currently piloting a new
procedure to treat with volumetric modulated arc therapy (VMAT). When treating with VMAT,
the radiation oncologist delineates the target volumes, cribriform plate, eyes, lenses, optic nerves,
chiasm, cochlea, parotids, esophagus, kidneys, bowel, and stomach. Typically, three isocenters
(four if necessary to accommodate for spine length per radiotherapy physicist) are placed in one
plane for planning and optimization. The first isocenter is placed at the center of the brain, the
second 20 cm inferior to the brain isocenter, and the third 50 cm inferior to the brain isocenter.
Arm and anterior tissue avoidance sectors are applied in the optimizer for the upper and lower
spine fields because the target is posterior and entry through the arms can lead to variation in
path entry if setup is not a complete match to the simulated CT.
If planning with VMAT, the patient is treated with a total of seven arcs, ensuring at least
5 cm of overlap between fields of the first and second isocenter, and the fields of the second and
third isocenter. The technique makes use of the Eclipse treatment planning system auto-
feathering feature, an algorithm that activates when a plan is optimized with multiple isocenters
and overlapping fields. Like the manual feathering method, the algorithm tapers the dose
delivered per field within overlapping regions so that the sum of the fields does not result in an
unreasonably high dose detrimental to the patient’s quality of life. In the future at MSK, the
radiation oncologist will evaluate the patient’s clinical criteria to decide whether to treat with
VMAT or 3D-CRT.
For the ProKnow case, my first round of planning followed the MSK-VMAT technique
but violated ProKnow’s clinical criteria for global max dose and PTV coverage. Additionally,
VMAT took more than 3 hours to complete one-run. A combined consideration of meeting
clinical criteria, understanding that VMAT generally produces a “hotter” plan, and requires the
use of “continue optimization” tricks to boost coverage to the PTV, I determined the best
approach would be to sum a 3D-CRT utilizing Field-in-Field (FiF) technique for the cranial field
matched to an IMRT spine treatment.
The Cranial isocenter was placed at the junction of the Cranial and Spine PTVs to
minimize divergence into the superior spine field. The respective jaws at the junction were
collimated to 0° to match between the Cranial and Spine plans. The Cranial plan used two lateral
fields (gantries 90° and 270°) within the same isocenter group with a 0° collimator and 0° couch
rotation, 6 MV with equal field weighting. The collimator and multi-leaf collimator (MLC)
leaves were fitted with a 0.5 cm margin in the X direction and 1.0 cm in the Y direction around
the PTVCranial to allow for proper scatter. After the forward dose calculation, the plan was
normalized so that 100% of the prescription covered 98% of the PTVCranial volume, then
assessed the dose distribution and DVH. The left and right lens doses were 2201 cGy and 2841
cGy, respectively. To reduce lenses to the required ProKnow constraint of 700 cGy, I had to
completely shield the lenses with a 0.5 cm margin field 01 and 02. To do so I created a 0.5 cm
expansion of the combined lenses, made this visible in BEV and pulled the leaves to cover
(Figure 1 and 2). This resulted in left and right lens dose of 583 and 667.8 cGy, respectively, and
98% of the PTV was covered by 100% of the dose.
The right and left optic nerves needed to be constrained to 3600 cGy and without
blocking, they received 3933 cGy and 3930 cGy, respectively. To gently reduce the dose below
3600 cGy without significantly reducing the PTV coverage, I used a FiF technique. Subfields
01.1 and 02.1 were created from parent fields 01 and 02 to shield the optic nerves with the leaves
(Figure 3 and Figure 4). I adjusted the normalization method to “lock” to the original value of
103.08 before adjusting the subfield weighting. Subfields 01.1 and 02.1 were then weighted
6.5% and 7.2% respectively.
To reduce the global maximum of the Cranial plan to 3955.9 cGy, subfields 01.2 and
02.2 were created to block 3960 cGy and each weighted 3.3% (Figure 5 and Figure 6). With this
method, I was able to reduce the global max to 3955.9 cGy without reducing PTV coverage.
The parotid doses were trickier to assess because the lateral fields partially treat the
parotid (they were split between the 3D-CRT and IMRT plans) so I was unable to adjust the FiF
until after I completed the IMRT. I created a plan sum with both plans to assess the OAR dose
then determined if I had to adjust my optimizer or 3D-CRT plan to further reduce the parotid
doses. The right parotid was most difficult to meet because it was further in the field for the 3D-
CRT plan than the left parotid. After reducing the parotid dose in the IMRT the lowest I could go
without impacting the PTV coverage, I went back to the 3D-CRT plan and partially blocked the
right parotid in the parent field 02 and sub-field 02.2 and assessed the DVH in the plan sum to
determine whether to adjust further. Upon assessing the dose, I manipulated the MLC placement
and weighting until the right parotid in the plan sum was less than 1500 cGy mean dose.
Figure 1: Parent field for Left lateral cranial field (01) with MLCs pulled to completely
shield lenses (green).
Figure 2: Parent field for Right lateral cranial field (02) with MLCs pulled to completely
shield lenses (green).
Figure 3: Left lateral cranial field (01.1) with MLCs pulled to shield lenses (green), parotids
(orange and pink), and optic nerves (purple and brown)
Figure 4: Right lateral cranial field (02.1) with MLCs pulled to shield lenses (green), parotids
(orange and pink), and optic nerves (purple and brown)
Figure 5: Left lateral cranial field (01.2) with MLCs pulled to shield lenses (green), parotids
(orange and pink), and reduce hotspots (locations identified by arrows) by manipulating beam
weighting.
Figure 6: Right lateral cranial field (02.2) with MLCs pulled to shield lenses (green), parotids
(orange and pink), and reduce hotspots (locations identified by arrows) by manipulating beam
weighting.
Figure 7: Field arrangement details of 3D Cranial plan
Figure 8: Cranial field % isodose distribution with PTV turned (dark blue) on in axial view.
Figure 9: Cranial field % isodose distribution with PTV (dark blue) turned on in frontal view
Figure 10: Cranial field % isodose distribution with PTV (dark blue) turned on in sagittal view
Spinal Plan
To accommodate for the length of the patient’s spine, three spinal isocenters groups were
created on the same plane for treatment planning and optimization. Per 3D-CRT departmental
procedure, the first Spine plan isocenter was matched to the Cranial plan at the junction of the
Cranial and Spine PTV with Y2 yaw set to 0°. The second spinal isocenter was placed 30 cm
inferior from the first isocenter, while the third was placed 60 cm inferior from the first (Figure
11).
Figure 11: Spine length and placement of CSI isocenters
For an MSK 3D-CRT clinical patient, the radiation oncologist is consulted to manipulate
the MLCs to block surrounding organs at risk (OAR) because there are no PTVs drawn. The
prescription indicates calculation depths, as the cranial field is prescribed to midplane while the
spinal fields are prescribed to a specific depth based on the average depth of the anterior portion
of the thoracic and lumbar spine.
For my final Proknow case I referenced a 2013 paper by Wang et al4 who conducted a
comparative study using a three-isocenter CSI intensity modulated radiation therapy (IMRT)
technique. The IMRT spine plan had a greater conformity index with better modulation of the
dose to OAR. With IMRT technique I utilized beam fluence editing to further manipulate the
OAR doses since I was unable to meet the 3 Gy mean kidney dose and volumetric esophageal
constraints with optimization techniques alone. The IMRT technique required less plan setup for
manual feathering because I used the auto feathering built into the optimization to account for
potential discrepancies in dose.4 With 3D-CRT, radiation therapists are required to measure the
gaps on skin between the feathered plans during treatment but with IMRT the field junctions
between the cranium and spine are modulated appropriately as a result of the optimization.
However, when setting up an IMRT spine case, the radiation therapist must ensure the spine
matches appropriately throughout the entire length of the field between beginning treatment
because they cannot shift during treatment. Shifting during treatment can cause unwanted
overlaps or gaps on the treatment area.
Energy of 6 MV was selected for each field to limit bowel exit dose, and a total of 18
posteroanterior (PA) fields separated into 3 different isocenter groups (8 fields in the first and 5
in the second and third isocenter group) with a 0-degree collimator and gantry rotation were
utilized for the spinal plan (Figure 12).
Figure 12: Axial view of spinal plan field selection showing central axis only
The posterior fields were set up to reduce dose spilling, to avoid entering through the
patient’s arms, and reduce kidney and esophageal doses. The field set with the first spinal
isocenter had gantry angles of 180°, 125°, 105°, 20°, 340°, 250°, 200° and 200°, respectively.
The field set with the second spinal isocenter had gantry angles of 180°, 160°, 135°, 235° and
205°, respectively. Finally, the field set with the third spinal isocenter had gantry angles of 180°,
150°, 125°, 240° and 220°, respectively. The collimators were then fitted with a 5.0 mm circular
margin around the “zzPTV_Total” structure and the default field weighting of 1 was kept for
each field (Figure 13).
Figure 13: Field arrangement details of IMRT Spinal plan
Each of the target and OAR contours provided by ProKnow were then assessed and
optimization structures were created to assist in achieving constraints. Using the extract wall
tool, 1.0 cm, 1.5 cm, 2.0 cm, and 3.0 cm rings with a 2.0 mm margin from the PTV were created
to assist in shaping the dose (Figure 14).
Figure 14: The 1.0 cm, 1.5 cm, 2.0 cm, and 3.0 cm rings with a 2.0 mm margin from the spinal
PTV
The optimizer was then accessed, and a preliminary run was done with the spinal clinical
target volume (CTV_Spinal), spinal PTV (PTVspine), and first ring (1.0 cm) only to ensure
correct field selection. A lower objective was assigned to the spinal CTV, and it was
communicated to the optimizer that 100% of the structure is to receive 3672 cGy (102% of the
prescription dose). A priority value of 150 was given. The same process was repeated for the
spinal PTV except that an extra lower objective with a 150-priority was added in addition to the
100% volume objective. This was done to communicate that additionally, 99.0% of the structure
is to receive 3725 cGy (approximately 103.5% of the prescription dose). Two upper objectives
with a 150-priority value were then assigned to the spinal PTV where it was communicated to
the optimizer that 2.0% of the structure is to receive 3750 cGy (approximately 104% of the
prescription dose) while 0% is to receive 3780 cGy (105% of the prescription dose). Finally, an
upper objective was assigned to the first ring where it was communicated that 0% of the structure
is to receive 3625 cGy (47 cGy under what was assigned to the 100.0% volume).
To achieve great coverage and better control the dose, exaggerated values were inserted
into the optimizer from the beginning. Then after the first preliminary optimization run, coverage
to the PTV was assessed and upon noticing close to perfect coverage, upper objectives were
assigned to the OAR and rings for the subsequent runs. A priority value of 125 was assigned to
the rings and the priority values of the OAR varied depending on how hard the optimizer needed
to be pushed to achieve desired ProKnow metrics. To achieve desired results, the structure
volumes that were optimized to were exaggerated and subject to change based on DVH
assessments done post optimization.
Finishing touches included creating an optimization structure to achieve better target
coverage, incorporating the body contour into the optimizer to decrease the global maximum
dose of the plan, and creation of an additional optimization structure to further decrease the
global maximum dose. The first optimization structure was a “cold structure” named “zzzCold,”
created by converting the 100% isodose line (IDL) into a contour, subtracting that 100% IDL
from the spinal PTV utilizing the boolean tool, and inserting the result of the operation into a
new structure called “zzzCold.” This structure was added to the optimizer with a lower objective
and 150-priority, and it was communicated that 100% of it (which represented the cold/under-
covered areas of the plan) was to receive 3725 cGy (approximately 103.5% of the prescription
dose). To decrease the global maximum dose of the plan, two upper objectives with a 250-
priority were assigned to the body structure and it was communicated that 0 and 2% of the
structure was to receive no more than 3800 and 3960 cGy, respectively. To further reduce the
global max dose, the “tail” of the PTV was assessed in the DVH using the crosshair tool to
identify the percentage with a volume of approximately 2 cc, and the percentage that
corresponded to that volume of the tail was converted to a contour named “zzzHot.” It was added
to the optimizer with a lower objective and a 250-priority value, and it was communicated that
0% of the structure was to receive 3900 cGy. When boosting or cooling a plan with optimization
structures, it is important to not assign the zzzCold structure an excessively high dose value or an
excessively low one (below prescription dose) because it can over or under-dose depending on
what is communicated.
In the end, the optimizer was able to achieve most of the desired ProKnow constraints by
pushing on the two targets, OAR, and optimization structures (Figure 15 – Figure 17). I fluence
edited to decrease the global maximum dose of the plan sum (hotspots) and decrease any
additional dose to OAR such as the kidneys. The spinal plan was normalized to increase target
coverage while maintaining appropriate OAR constraints. Upon completion the isodose
distribution of the plan was assessed (Figure 18 – Figure 20).
Figure 15: Optimizer interphase demonstrating structures and manipulated settings
Figure 16: Optimizer interphase demonstrating structures and manipulated settings
Figure 17: Optimizer interphase demonstrating structures and manipulated settings
Figure 18: Isodose distribution of the superior portion of the spine with spinal PTV
(blue) on in axial view.
Figure 19: Isodose distribution of the middle portion of the spine with spinal PTV (blue)
on in axial view
Figure 20: Isodose distribution of the inferior portion of the spine with spinal PTV (blue)
on in axial view
Plan Evaluation
The use of 3D-CRT for the brain and IMRT for the spine proved to be successful in
meeting ProKnow constraints (Figure 21). The only constraints that could only be marginally
met per ProKnow score include the volume of the cranial PTV covered by 3960 cGy, volume of
the spinal PTV covered by 3960 cGy, ideal maximum dose to the right lens of 700 cGy, and
ideal maximum dose to the right optic nerve of 3600 cGy . It should be noted that the ProKnow
dose statistics differ from the actual Eclipse values because of the difference in calculation grid
between the two systems.
Figure 21: ProKnow scorecard of composite plan
In the plan summary, my plan’s hotspot manifested within the PTV in the patient’s sacral
region at the level of S3 (Figure 22). This location is appropriate as due to the nature of a 6MV
beam, more dose will be deposited at shallower depths as is being seen in this scenario. Had the
field setup not been ideal, the hotspot would have manifested at the match.
Figure 22: Sagittal view of the hotspot in the plan summary
The plan was also assessed for full coverage and cold spots utilizing the dose wash tool
by turning on 100% of the prescription dose (green), 95% of the prescription dose (light blue),
and the “zzPTV_Total” (pink) structure made up of all PTVs in the plan summary (Figure 23 –
Figure 24). The majority of the target is receiving 100% of the prescribed dose, and the areas in
pink represent cold spots, or locations where full coverage could not be achieved. The DVH was
evaluated once again at the end to ensure ProKnow results were reasonable (Figure 25).
Figure 23: Color wash displaying 100% prescription dose coverage (green) with the
“zzPTV_Total” structure (pink) turned on to assess for cold areas
Figure 24: Color wash displaying 95% prescription dose coverage (light blue) with the
“zzPTV_Total” structure (pink) turned on to assess for cold areas
Figure 25: The DVH of the CSI plan displaying targets and OAR
Reflection
The use of 3D-CRT for the brain and IMRT for the spine proved to be successful in
meeting ProKnow constraints. The 3D-CRT provided helpful blocking capabilities to the lenses
and parotids while achieving optimal target coverage, while the IMRT provided greater
conformality of higher dose levels along with dose reduction capabilities to the OAR while
achieving optimal target coverage. My clinical site prefers 3D-CRT due to the quicker
turnaround and blocking capabilities. This project enhanced my understanding of CSI treatments
and allowed me to not only plan using 3D-CRT and IMRT techniques but also experiment with
our new VMAT procedure (even though that attempt wasn’t as successful as would’ve been
desired). More importantly, I learned the importance of setting up match lines correctly (Figure
26) after a bad first attempt where due to incorrect field setup the hotspots manifested at the
match.
Figure 26: Sagittal view display of my patient’s fields with emphasis placed on correct
matching.
References
1. Gibbons JP, Khan FM. Khan's the Physics of Radiation Therapy. Philadelphia: Wolters
Kluwer; 2020.
2. Biancia CD. CSI: Craniospinal Irradiation. [PowerPoint]. New York, NY: In-
service;2020.
3. Noble DJ, Ajithkumar T, Lambert J, Gleeson I, Williams MV, Jefferies SJ. Highly
Conformal Craniospinal Radiotherapy Techniques Can Underdose the Cranial Clinical
Target Volume if Leptomeningeal Extension through Skull Base Exit Foramina is not
Contoured. Clin Oncol (R Coll Radiol). 2017;29(7):439-447.
doi:10.1016/j.clon.2017.02.013
4. Wang Z, Jiang W, Feng Y, et al. A simple approach of three-isocenter IMRT planning for
craniospinal irradiation. Radiat Oncol. 2013;8:217. Published 2013 Sep 17.
doi:10.1186/1748-717X-8-217
