June 1, 2009
Tumor Autocontouring — Efficiently Maximizing Dose and Minimizing Damage to Healthy Tissue
By Mark Klincewicz, BME
Vol. 10 No. 11 P. 22
Any method aimed at treating a tumor surgically or with radiation therapy requires a precise knowledge of the cancerous volume periphery and must adhere to the maxim that treatment should be amplified and the consequences on adjacent normal tissues should be minimized.
Radiation therapy seeks to effectively irradiate tumor cells, but irradiation to adjacent normal cells within anatomical structures is not the treatment path an oncologist wants to take. To minimize radiation dosage to adjacent normal cells, targeted tumors or volumes must be precisely identified. Tumor contouring attempts to achieve that goal with high accuracy and reliability by utilizing various automated segmentation processes.
In most cases, contouring is carried out manually by a specialist. Digital images, obtained from modalities such as CT or MRI, are used to view and locate the tumor. The doctor then marks the perimeter of the cancerous tissue on each image. However, the accuracy of the perimeter markings varies from doctor to doctor. This subjective variability is further exacerbated by the limits of the medical image. For instance, images containing many kinds of tissues (eg, dense breast tissue, ducts, blood vessels) other than tumors, as well as noise, make it difficult to mark the target using just simple edge manual techniques.
Finding the Edge
Along with contouring the targeted volume, identifying an adjacent normal structure and its periphery is a key to safe treatment. For example, it is vitally important to protect brain structures that control motor, sensory, and language functions. On a broader scale, every healthy structure in our bodies should be spared whenever possible.
“With manual contouring, you start with a blank CT scan, and you draw the shapes that you would treat or the organs that you want to avoid,” says Tom McGowan, MD, MBA, FRCPC, head of radiation oncology at Credit Valley Hospital in Mississauga, Ontario, Canada. “What autocontouring does then, it applies these contours into the CT scan, into the 3D data set, and then uses various intelligence tools to shape it to the structures that are there to try and match the contour set of the prostate or bladder to the actual bladder and the actual prostate that the patient has. That is only the first step. Then everything gets brought back into the planning system, and the physician needs to approve and finalize the contours.”
“We build what’s called a template, by tumor or by region, where all the various organs that we can outline are outlined,” says Diane Heaton, MD, a radiation oncologist and the medical director of Oklahoma CyberKnife. “The computer can then develop what we call ‘dose volume histograms.’ Once we have the volume of each normal structure, the computer tells us how much maximum dose can go to that area. For the lung region, I will contour the lungs, skin, esophagus, spinal cord, heart, major vessels and arteries, and a number of different areas.”
Manual contouring is both subjective and time consuming. Autocontouring, or computer-aided contouring, was introduced and developed to alleviate these problems. It uses algorithms to create a more objective and standardized contouring method. It can use computer-aided diagnosis applications to save some of the time required to manually sketch a precise contour via automatic segmentation.
“For patients with palliative treatments,” says McGowan, “we can get them on and off the scanner quickly, and then we can spend the time-to-field offline, without the patient laying on a hard bed, potentially in pain, as we are trying to shape these fields. The whole move toward electronic manipulation of treatment plans and images has been huge.
“Physicians are busy,” adds McGowan. “If we can fill in the work that has to do with filling in the structure, then it is much faster to just tweak it and modify it than it is to do slice after slice after slice of contours around a bladder or prostate that is well defined. When you put a prostate set on it, it is a perfect fit. Whereas, in other areas where the contrast isn’t as great, we would have to spend time modifying it. I think it’s going to be a real time-saver.”
The subjectivity of the doctor is not the only variable that affects the task of tumor contouring. Internal organ motion contributes to problems identifying and monitoring cancerous volumes, and images of a patient taken on different days may not match up due to such internal motion.
Efficiency Improving Accuracy
“Autocontouring, I think, will improve accuracy and the ability to look at more and more structures because it’s not as laborious to outline those structures, which, as before, had to be done by hand,” says Heaton. “You can think twice if [you don’t have to contour] the extra structures [because] you can do them automatically. We literally can get all the structures in the chest identified.
“Before autocontouring, all of that basically had to be done by hand, so that could be up to a three-hour process. With autocontouring, a lot of that can be done with the computer, and it makes that process as short as 30 minutes,” she adds.
The development of technology, however, has not just focused on contouring itself but providing information for many doctors on the oncology team. Medical News Today reported the following in an article titled “IPlan Neuroradiology: Advanced Image Analysis Software for Improved Diagnosis by BrainLAB”: “…From diagnosis through to the treatment stage, anatomical and functional data gathered by radiologists for analysis can now be used for a comprehensive range of treatment solutions, including surgery, chemotherapy, and radiation oncology. Neurosurgeons can load the analyzed 3D brain anatomy into their intraoperative image guidance systems, providing detailed information of the brain right at their fingertips, whereas radiation oncologists can use it for radiation dose planning and delivery.”
In the case of a brain tumor, the process starts with the evaluation by a physician. The imaging is done to determine that the patient indeed has a brain tumor. Then it is imported into planning software to define the tumor and all the organs at risk, says Ramona Beasley, a product manager for BrainLAB AG-Oncology North America.
The tumor’s contour is then obtained via an automatic fusion process that allows the fusion of multiple data sets, such as PET, MR, and CT, according to Beasley. This fusion process allows the physician to get a clearer idea of the exact location of the tumor. For example, if the peripheral borders of the active tumor are hard to distinguish, perhaps located behind bone in an MRI scan, the clinician can fuse the CT and MR together. That overlay can be utilized to determine exactly where the tumor borders are. If the clinician is concerned that an uptake of radioactive isotopes introduced by an MR contrast agent could cause activity outside of the originally contoured border, a PET image can be introduced to define the contour more precisely. Functional data analysis is further enhanced with advanced functional MRI and diffusion tensor imaging algorithm capabilities.
BrainLab’s IPlan has tools called SmartBrush and SmartShaper to help define the tumor. Keeping in mind that a CT scan produces hundreds of images and highlighting the tumor in each image would be necessary to complete the volume in 3D, SmartBrush allows the user to select the first slice and the last slice. A linear interpolation includes everything inside that volume to give a 3D display of the tumor volume. Then, “The SmartShaper allows the user to manipulate that contour in 3D, so all three planes—coronal, sagittal, and axial—can manipulate the contouring all at once,” Beasley says.
The ultimate objective is to increase the therapeutic ratio by increasing tumor control probability while decreasing normal tissue complication probability.
Paul Yokoyama, a product manager for the user experience team with Varian Medical Systems, says Varian’s Eclipse treatment planning system supports this goal. “Eclipse is a treatment planning system, a computer program that will generate a virtual 3D model of the patient based on CT and other anatomical images and then simulate different types of treatment plans on that 3D model.”
The simulation of radiation treatment on that model can consider different modalities, types of treatment machines, types of angles, and types of beam arrangements depending on the physician’s intentions. Using those various simulations, the system calculates the effect of the radiation dose inside the patient without actually delivering it. The Eclipse can model the best solution for treating the targeted lesion, sparing as much of the critical structures as feasible.
“Once treatment is dispensed, it cannot be taken back,” says Yokoyama. “Extreme care must be exercised when radiation therapy is used, so every precaution is made to properly target the area and treat with proper dosage.”
A significant part of autocontouring, or autosegmentation, says Michele Abraham, a product manager with Accuray Incorporated, relies on a library of different CT images acquired from other people for the same anatomical structures that are to be contoured in the current patient. With prostate autocountouring, for example, prostate CTs from 50 to 100 different men are acquired. Targets and critical structures are contoured on each CT slice. These contours are then stored in a library for later retrieval. When a prostate patient's images are imported for planning, the Accuray autosegmentation tool will search the library for the best fit.
“The autosegmentation tool looks for differences in contrast and shape and automatically populates a ‘best fit’ contour onto the current patient's anatomy,” Abraham says. By starting with the best match from the database, clinicians can save time because they only need to adjust the delineation of the best fit to match the current patient’s anatomy.
As archaic as relying on existing databases to contour anatomical structures seems, the technology is effective. Autocontouring has attained accuracy levels reaching error limits of no more than 1 mm.
But that accuracy goes for naught if one thing is ignored during radiation therapy—the position of the patient. Whether using a positioning device, a tracking device, or taking x-ray images of the patient during treatment, Yokoyama says that “making sure the patient is in exactly the same position as when the initial computer modeling is made is extremely important.”
One method for precise positioning requires the patient to be placed within a certain range of positioning. Once the patient is in this general position, x-ray images are taken throughout the treatment and compared to where the patient’s anatomy is expected to be. The machinery moves the patient, adjusting and aligning the anatomy in the correct position, delivering beams to the patient based on where their anatomy is at that particular point in time.
“When we do computer modeling with the patient in a certain position, we have to make sure that when we deliver the treatment, the patient is in exactly the same position; otherwise, the plan that we have done beforehand is not going to be worth anything,” says Yokoyama. “If the patient is in a different position, it could mean that we are not targeting the area that we want to treat.”
As cancer treatment planning and targeted radiation therapy become more sophisticated, tumor contouring makes it easier for users to define tumor volumes with speed and accuracy and then deliver as much dosage as possible to the target while minimizing the damage to normal anatomy surrounding it.
“I believe it is quicker and more accurate in many instances,” says Heaton. “This is an area in medicine that is literally exploding. I think we’re going to see computerized mapping, computerized identification structures, and [it will probably be used in] both the diagnostic and therapeutic realms.”
— Mark Klincewicz, BME, is a freelance writer based in Chicago