RESEARCH Open Access
A 4D IMRT planning method using deformable
image registration to improve normal tissue
sparing with contemporary delivery techniques
Xiaoqiang Li, Xiaochun Wang, Yupeng Li and Xiaodong Zhang*
Abstract
We propose a planning method to design true 4-dimensional (4D) intensity-modulated radiotherapy (IMRT) plans,
called the t4Dplan method, in which the planning target volume (PTV) of the individual phases of the 4D
computed tomography (CT) and the conventional PTV receive non-uniform doses but the cumulative dose to the
PTV of each phase, computed using deformable image registration (DIR), are uniform. The non-uniform dose
prescription for the conventional PTV was obtained by solving linear equations that required motion-convolved 4D
dose to be uniform to the PTV for the end-exhalation phase (PTV50) and by constraining maximum inhomogeneity
to 20%. A plug-in code to the treatment planning system was developed to perform the IMRT optimization based
on this non-uniform PTV dose prescription. The 4D dose was obtained by summing the mapped doses from
individual phases of the 4D CT using DIR. This 4D dose distribution was compared with that of the internal target
volume (ITV) method. The robustness of the 4D plans over the course of radiotherapy was evaluated by computing
the 4D dose distributions on repeat 4D CT datasets. Three patients with lung tumors were selected to demonstrate
the advantages of the t4Dplan method compared with the commonly used ITV method. The 4D dose distribution
using the t4Dplan method resulted in greater normal tissue sparing (such as lung, stomach, liver and heart) than
did plans designed using the ITV method. The dose volume histograms of cumulative 4D doses to the PTV50,
clinical target volume, lung, spinal cord, liver, and heart on the 4D repeat CTs for the two patients were similar to
those for the 4D dose at the time of original planning.
Keywords: 4D CT, IMRT, treatment planning, respiratory motion, deform
1. Introduction
Implementations of four-dimensional (4D) radiotherapy
based on 4D computed tomography (CT) datasets have
been described by Rietzel et al [1] and Keall [2]. In 4D
radiotherapy, the treatment plan is designed on each 4D
CT image set (i.e., 4D treatment planning), and radia-
tion is delivered throughout the patient’s breathing cycle
(i.e., 4D treatment delivery), which ensures adequate
coverage of the tumor target without increasing the
treated volume. Because 4D treatment planning
accounts for temporal changes in anatomy, 4D radio-
therapy holds promise as the optimal method for treat-
ing patients. However, 4D radiotherapy currently
requires 4D treatment delivery, which necessitates
sophisticated device(s) to synchronize the treatment
delivery with the patient’s respiration. Most centers have
the ability to acquire 4D CT images, but they do not
have the ability to perform 4D radiation delivery.
Instead, 4D CT images are primarily used to define the
internal target volume (ITV), which is essentially the
envelope needed to enclose the target as it moves
throughout the breathing cycle. 4D CT [3-9] provides a
more accurate tumor volume definition since it limits
motion artifacts during CT acquisition, displays the ana-
tomically correct shape and size of the tumor, and
demonstrates respiration-induced motion of the tumor
and organs at risk. Previous studies using 4D CT data-
sets have mostly been focused on dosimetric verification
to determine if dose distribution planned on one or part
of the 4D CT datasets is adequate to estimate the cumu-
lative dose from all 4D CT datasets [1,10]. Few studies
* Correspondence: xizhang@mdanderson.org
Department of Radiation Physics, The University of Texas, MD Anderson
Cancer Center, Houston, Texas 77030, USA
Li et al. Radiation Oncology 2011, 6:83
http://www.ro-journal.com/content/6/1/83
© 2011 Li et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons
Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in
any medium, provided the original work is properly cited.
have investigated whether the information on anatomic
motion provided by 4D CT can be used to design treat-
ment plans that confer the advantages of 4D treatment
delivery without requiring additional equipment.
In this paper, we describe an effective and practical 4D
treatment planning method, which we refer to true 4D
planning (t4Dplan) method, for intensity-modulated
radiotherapy (IMRT) using 4D CT datasets to maximize
critical structure sparing. In traditional treatment plan-
ning, the prescribed dose is planned to be distributed
uniformly to the target while minimal dose is delivered
to the surrounding normal structures on the planning
CT under the assumption that the planning CT truly
represents the patient anatomy that will be present dur-
ing treatment. In our t4Dplan method, however, plan-
ning deliberately creates non-uniform dose distribution
in the target (i.e., it creates hot regions along the target’s
direction of motion on the planning CT) to achieve a
uniform dose distribution in the target and minimal
dose to the surrounding normal structures on the final
4D dose distribution. The difference between the
t4Dplan method and the traditional ITV method is illu-
strated in figure 1. The t4Dplan method does not
require 4D treatment delivery and is solely dependent
on the 4D datasets acquired during the planning pro-
cess. Compared to some other techniques such as
respiratory gating [11], breath hold [12,13] and dynamic
MLC tumor tracking [14-16], the t4Dplan method is
easier to implement in the clinic because it uses the cur-
rent treatment planning and delivery systems.
2. Materials and methods
2.1. t4Dplan
The t4Dplan method, which uses 4D CT datasets,
designs treatment plans as follows:
1. A reference CT dataset is selected from all the 4D
CT datasets. Usually, an end-of-exhalation phase CT
(i.e., the 50% phase [T50]) is selected as the refer-
ence CT dataset [17] since patients spend more time
at the end of exhalation [18].
2. The target volume (TV) is outlined based on the
reference CT.
3. The motion TV (MTV) is outlined on the refer-
ence CT as the combined volume of the target at all
phases of the 4D CT datasets (i.e., the MTV is an
envelope enclosing the target as it moves throughout
the breathing cycle).
The t4Dplan method calculates a deliverable non-uni-
form dose distribution (i.e., the apparent dose distri-
bution [AppD]) to the MTV. The final 4D dose
distribution is determined by recalculating the t4Dplan
on each phase of the 4D CT dataset and creating a
time-averaged cumulative dose distribution based on
deformable image registration (DIR).
For each voxel on the reference CT, the corresponding
voxel on another phase of the CT dataset can be derived
through DIR by transforming the source image (i.e., the
reference CT) to the target image (i.e., another phase of
the CT dataset), such that
υ
j
i = T
j,ref × υrefi , (1)
where υ refi is the position vector for the ith voxel on
the reference CT, Tj,ref represents the transform matrix
from the reference CT to the jth phase of the CT data-
set, and υ ji is the position vector for the corresponding
voxel on the jth phase of the CT dataset for the ith
voxel on the reference CT.
In the current study, to derive the non-uniform dose,
we first assumed that the dose on each phase of the 4D
CT was approximately the same as the AppD on the
reference CT. This approximation assumes the internal
movement of anatomy will not impact the dose distri-
bution and is a good approximation for photon dose
calculation. It should be noted that this approximation
is only used in the derivation of a non-uniform dose
prescription. For the final designed plan, we used the
exact 4D dose calculation without this approximation.
The 4D dose for each voxel on the reference CT can be
approximated as the time-averaged cumulative dose of
the corresponding voxel on all phases in the CT dataset,
such that
D4D(υrefi ) =
1
K
K∑
j=1
DAppD(υ ji), (2)
where K represents the number of phases of the CT
datasets, D4D(υrefi ) is the 4D dose for the ith voxel on
the reference CT, and DAppD(υ ji) is the AppD for the
corresponding voxel on the jth phase of the dataset.
Assuming the MTV and TV on the reference CT have
n and m (n >m) voxels, respectively, and the AppD
values for the n voxels of the MTV are D1, D2, ..., Dn,
the 4D dose distribution for the TV with m voxels can
be determined using the following linear equations
derived from equation (2):
D4D(υref1 ) =
1
K
(DAppD(υ11) +D
AppD(υ21) + ...... +D
AppD(υK1 )) = D0, 1
st voxel;
D4D(υref2 ) =
1
K
(DAppD(υ12) +D
AppD(υ22) + ...... +D
AppD(υK2 )) = D0, 2
nd voxel;
D4D(υrefm ) =
1
K
(DAppD(υ1m) +D
AppD(υ2m) + ...... +D
AppD(υKm)) = D0, mth voxel,
(3)
where DAppD(υ ji) = D1,D2, . . ., or Dn, are the
unknown parameters, and D0 is the uniform dose pre-
scribed to the TV (i.e., the final 4D dose distribution on
the TV). Here, we have n unknown parameters (i.e., D1,
Li et al. Radiation Oncology 2011, 6:83
http://www.ro-journal.com/content/6/1/83
Page 2 of 14
D2, ..., Dn) that need to be derived from m equations,
with m