Tumor Characterizing Motion Using 4-D Computed Tomography
Student : Huanmei Wu (NU) Contributors: Eike Rietzel (MGH), George Chen (MGH), David Kaeli (NU), Betty Salzberg (NU)
Abstract
The general objective of radiotherapy is to achieve tumor control by depositing a high, lethal dose in the target volume, while maximally sparing surrounding organs and tissue. Therefore, precise localization of the target volume is needed for treatment planning; treatment planning is usually based on computed tomography (CT) scans. Especially in the lung and abdominal region tumors are subject to respiratory motion. Currently such motion is accounted for during treatment planning by an expansion of the volume to be irradiated. In most cases such expansions are more tumor site specific than patient specific. A new imaging technique, time-resolved CT imaging (4DCT), provides several discrete, volumetric snapshots of the patient while breathing. Based on 4DCT data the motion of tumors due to respiration can be analyzed patient specifically. We have started to study the motion of lung tumors based on manual segmentations of 4DCT data. Temporal variations of tumors can be described by volumetric variations, composite volume, center of mass trajectory, time-dependent minimum bounding box, and probability density function.
3.1. Visualization of Tumor Motion
Based on segmented 4DCT data, motion of different organs as well as the tumor can be visualized. Figure 2 shows an example of 4D motions of the tumor (pink), left lung (green), right lung (yellow) and patient (blue). Each of the images shows a different phase of the patient's breathing cycle.
3.4. Minimum Bounding Box
To include volumetric changes as well as trajectories, the minimum bounding box (MMB) of the segmented tumor is calculated per phase. Figure 5 shows that the MMB is a good estimation of tumor trajectory as well as volumetric changes. Center of mass and MBB motion are comparable, volumetric changes are indicated fairly well. However, only rigid translations can be captured with the MBB, tumor rotations are not necessarily reproduced. To track individual voxels deformable models will have to be used.
Boundary Changes
80 70
Right-Left
345 343 341 339 337 335 333 331 329 327 325
GTV RL Center MBB RL Center
Anterior-Posterior
210 208 206 204 202 200 198 196 194 192 190
GTV AP Center MBB AP Center
Superior-Inferior
66 64
RL Centers (mm)
SI Centers (mm)
Max-Min (mm)
60 50 40 30 20 10 0
RL AP SI
AP Centers (mm)
62 60 58 56 54 52 50 1 2 3 4
GTV SI Center MBB SI Center
1
2
3
4
5
6
7
8
9
10
1
2
3
4
5
6
7
8
9
10
1
2
3
4
5
6
7
8
9
10
5
6
7
8
9
10
Phase
18 16
Phase
Phase
Phase
90 80
20 15 10
GTV Volume Changes MBB Volume Changes Center of Mass distance Changes
Relative Volume (%)
GTV / MBB Volume (%)
Individual/Combined%
14 12 10 8 6 4 2
70 60 50 40 30 20 10 0
1. State of the Art
Treatment planning in radiotherapy is based on CT data. For segmenting target volumes (gross tumor volume, GTV), sometimes other imaging modalities are used as well. The GTV is expanded to encompass microscopical spread of cancer cells (clinical target volume, CTV). To ensure adequate dose coverage of the clinical target during irradiations the CTV is expanded to the PTV (planning target volume). The PTV is designed to incorporate possible set-up uncertainties, daily anatomical variations, and respiratory motion. The PTV concept usually does not reflect patient specific temporal motion of the tumor [1,2].
Changes (%)
5 0 -5 -10 -15 -20 -25
Figure 2 Spatio-temporal motion
GTV Volume MBB Volume
3.2. Volumetric Changes
The volumes of each segmented structure are calculated for different phases of the respiratory cycle. Phases are evenly distributed in time over a full respiratory period. Figure 3 shows volumetric changes of the segmented tumor volume per breathing phase. Total volume (left), relative compared volume to a composite volume for all phases (center), and volumetric changes from phase to phase (right) are plotted. The most significant errors in the current analysis might be introduced by manual segmentation. Therefore, for a final conclusion on volumetric variability of tumors, further investigations are necessary.
12000
0 1 2 3 4 5 6 7 8 9 10
1
2
3
4
5
6
7
8
9
10
-30 1 2 3 4 5 6 7 8 9
Phase
Phase
Phase
Figure 5 Minimum bounding box (MBB)
4. Future Goals
Based on our current studies probability density functions (PDF) of tumors and organs can be calculated. PDFs summarize the spatiotemporal motion. In a next step we will use PDFs of tumors and organs at risk to optimize beam angles for patient treatments based on tumor motion. Angle optimization will be based on parallel projections of the patient geometry under all possible angles for all phases. The smallest changes in target projection for different phases should identify treatment angles which are least affected by the motion. We plan to design and apply database technologies for medical research. An important part will be complex multi-attribute or spatio-temporal subsequence similarity matching. Subsequence similarity matching over multiple streams, such as trajectories, volumes, and minimum bounding box can be used for inter-patient comparisons and studies related to tumor / organ motion at different sites.
Strategic Research Plan
Bio-Med S1 S2 S3 S4 Enviro-Civil S5 L3
2. Workflow
4DCT provides several discrete volumetric representations of a patient during a breathing cycle. In each of these volumes, the tumor is manually contoured. Based on the segmented data, temporal changes of tumors and organs can be analyzed. Different properties as described above are computed for each of the datasets. These properties are used to characterize tumor motion. The figure below shows the data processing workflow.
4DCT data acquisition
Volume (voxel)
8000 6000 4000 2000 0 1 2 3 4 5 6 7 8 9 10
1
2
3
4
5
6
7
8
9 10
Volumetric Changes (voxel)
Relative Volume %
10000
70 60 50 40 30 20 10 0
200000 150000 100000 50000 0 -50000 -100000 -150000 -200000 -250000 1 2 3 4 5 6 7 8 9 10
Phase
Phase
Phase
Figure 3 Tumor volumetric changes
manual segmentation
3.3. Center of Mass Trajectory
Overall tumor motion can be characterize...