2012 Progress Report DBP Adaptive Radiotherapy

From NAMIC Wiki
Jump to: navigation, search
Home < 2012 Progress Report DBP Adaptive Radiotherapy

Back to NAMIC_2012_Progress_Report

4.1.3 Adaptive Radiotherapy for Head and Neck Cancer

Key Investigators Gregory C. Sharp, PI, Massachusetts General Hospital Polina Golland, NA-MIC Algorithms, MIT Allen Tannenbaum, NA-MIC Algorithms, Georgia Tech Steve Pieper: NA-MIC Engineering, Isomics Host Institution: Massachusetts General Hospital

<< DRAFT >>

BACKGROUND

Head and neck cancers account for about 60,000 new cancer cases per year and represent 4-6% of all cancers in the United States. Sixty percent of patients present with advanced disease. The five-year survival is approximately 50%. These cancers are treated by a combination of chemotherapy, radiotherapy, and surgery. During a six-week regimen of radiotherapy, head and neck cancer patients often exhibit anatomic changes that affect their treatment. These changes include tumor regression or growth, changes in lymph node size, and changes in air cavities. Uncorrected, these changes can increase the risk of treatment complications or reduce treatment efficacy.

Adaptive radiotherapy addresses the problem of anatomic change by incrementally adjusting the radiotherapy plan, and is a prime example of personalized medicine. A mid-treatment adjustment is complex: it requires a new CT image, image segmentation, deformable registration, and mapping of the previously delivered dose onto the new image. This project proposes to use the NA-MIC Kit to develop a simple, practical workflow for achieving adaptive radiotherapy which can be applied on a case-by-case basis.

Role of Imaging in Adaptive Radiotherapy

Biology of Radiation Therapy Ionizing radiation, such as photons or charged particles, creates ionized atoms and molecules as it travels through matter. Energy is transferred from the incident radiation beam to the aqueous solution within a cell, which creates free radicals. The most important of these radicals, the hydroxyl radical (OH*), indiscriminately attacks neighboring molecules, and binds with them in a chemical reaction. A radiation beam of high intensity will create a large number of these radicals, which will bind with nuclear DNA in a cell, and break the deoxyribose backbone of the DNA molecule. When both DNA strands are broken, the cell can no longer reproduce itself, and therefore can no longer proliferate as cancer.

Proton Therapy Proton-beam radiotherapy operates on the same biological principles as traditional photon-beam treatments, but their physical properties are quite different. When a photon interacts with matter, it delivers most of its energy locally, and the beam is attenuated exponentially as the number of photons is reduced with depth. In contrast, each proton has many interactions as it travels through matter, and loses a little energy with each interaction. When it has lost almost all of its energy, the proton stops, and delivers its remaining dose. This means that protons are an effective technique for sparing healthy tissues distal to the tumor, which reduces radiation-related side effects.

RESEARCH PROGRESS REPORT

Clinical Analysis was performed to determine if adaptive replanning is needed to compensate for anatomic change during proton-beam radiotherapy. CT images of eight patients treated with proton therapy in the base of skull were acquired prior to radiotherapy, and mid-treatment. The physician delineated the tumor volumes and critical structure volumes on the pre-treatment scan. The CT scan acquired at mid-course was registered rigidly to the bony structures of the skull to remove the setup error. A deformable B-spline based registration was then performed to transfer structure contours from the planning CT to the mid-treatment CT. The original treatment plan was the applied to the mid-course CT, and proton dose was recalculated. Anatomic differences between the two CT scans were analyzed interactively, and dose distributions were evaluated by comparing isodose lines and distance-to-agreement analysis. Dose-volume histograms were compared using transferred contours on the mid-course CT and original contours on the pre-treatment CT.

Anatomic change was noted in all eight cases. The most prominent patient response to the treatment was tumor shrinkage in the nasal cavity and the paranasal sinuses. Three examples of dramatic anatomic change are shown in Figure 1. In the first case (left), the tumor volume (outlined in green) has become calcified as a result of therapy. The calcified tissue is of higher density, which affects the range of a proton beam. The beam loses energy more rapidly in the dense region, which might lead to underdosing at the most distal edge of the beam. In the other two cases (center, right), the tumor has shown a good response to therapy, and the tumor has shrank. Areas within the volume which were tissue density are now replaced with air. In contrast to the calcified tissue, which absorbs more energy per unit length than tissue, the air absorbs little energy, but does scatter the proton beam. The expected effect of this change is to produce overdosing at the distal edge of the beam.

The effect of anatomic change on dose distribution was computed, and assessed. Our analysis of the dosimetry concentrated on changes in dose to the tumor volume, and changes in dose to the normal tissues. An example of changes in the dose to the tumor volume is shown in Fig 2, which demonstrates hot spots (left) and cold spots (right) in the planned dose distribution. We are still investigating the origin of these findings, but the most likely cause is anatomic change. From pre-treatment to mid-treatment CT, we identified a change in the average gross tumor volume of -12% (range 0%-36%). In most cases, the lost tumor volume is replaced with air. This is verified by noting a median decrease in tumor density of 43 Hounsfield Units (=4.3% of the density of water). In addition to tumor volume, we analyzed normal tissues and found a median increase of mean dose to the brainstem of 1.5%, and median increase of maximum dose to the brainstem of 8%. The details of the study will be presented at the AAPM annual meeting.

Algorithms Image segmentation is a key technology needed to make adaptive radiotherapy a practical option for head and neck cancer patients. Through our collaboration with MIT, we have performed a thorough evaluation of a non-parametric approach to image segmentation. This method is a multi-atlas based approach in which multiple labeled images are registered to the unlabeled target image, and then a weighted voting method is applied to transfer the labels. Analysis was performed on the left and right parotids, and the brainstem, as these structures are particularly challenging to contour due to their inter-patient variability. The algorithm compensates for patient variability by emphasizing contributions from training images that are more locally similar to the target image. With a database of sixteen images, we achieved mean Hausdorff distances of 2.8 mm for the brainstem, and between 3.4 and 4.4 mm for the parotid glands. The details of this study will be presented at the AAPM annual meeting.

Deformable image registration is another key component of adaptive radiotherapy. In this year, we have addressed the problem of interactive registration, which is needed for cases where automatic registration fails. The method used is an analytically regularized landmark spline, which takes point landmarks from the user, and interpolates a vector field as a sum of Gaussian kernels. Unlike previous approaches, this method is both local and regularized. The locality property is essential for making fine corrections to a deformation without disturbing distant areas, and the regularization is needed to preserve invertibility. This work has been statistically validated, and was recently published in Physics in Medicine and Biology as a featured article.

Engineering The engineering plan for the DBP is proceeding according to schedule, and we have achieved all project milestones. The DICOM-RT import/export modules were available early in 2011, and we released the first version of the adaptive dose warping module at the project week in June 2011. These modules are fully documented, and include end-user tutorials. In the third quarter of 2011, we upgraded the B-spline registration module to include analytic regularization. Finally, a working prototype of a dose review tool was introduced at the winter project week in January 2012. A screenshot for this module is shown Figure 3, which demonstrates the use of the 3D gamma method for dose comparison. The gamma method is a well established method in the medical physics community for comparing two spatial distributions of radiation, because it lets the user define action thresholds on both the absolute dose difference and the spatial similarity.

Plans for the Coming Year In the following year, we plan to address the following goals: (1) A 3D Slicer module for atlas-based segmentation of head and neck cancer; (2) Improved support for radiotherapy structure sets in 3D Slicer (3) Improve and document dose review tools; (4) Review and improve interactive segmentation tools.

Papers that Acknowledge NA-MIC Nadezhda Shusharina and Gregory Sharp, “Analytic regularization for landmark-based image registration,” Phys. Med. Biol. 57:6 1477-1498, 2012. doi:10.1088/0031-9155/57/6/1477

Marta’s paper???

Additional information is available on the NA-MIC wiki

http://www.na-mic.org/pages/DBP:Head_and_Neck_Cancer



Figure 1. Anatomic change was identified inside gross tumor volume on eight patients with head and neck cancer. The top row shows pre-treatment anatomy, while the bottom row shows mid-treatment anatomy.



Figure 2. In some cases, hot spots (left) and cold spots (right) of greater than 5% were identified in the planned dose distribution. These effects are likely caused by anatomic change.


Figure 3. A new dose comparison tool developed for Slicer uses the 3D Gamma index to compared two different doses and highlight changes.