Developing Therapy for Non-Hodgkin’s Lymphoma with Supercomputer Simulations

he overall mortality rate of cancer in the United States is declining, but increasing numbers of patients are dying from non-Hodgkin’s lymphoma, the nation’s fifth leading cause of cancer. Successful clinical trials of therapies using monoclonal antibodies to deliver radioactive isotopes to tumors are offering hope to patients with non-Hodgkin’s lymphoma. Now, University of Michigan nuclear medicine researcher Yuni Dewaraja and her colleagues are developing a method to improve the effectiveness of that therapy with patient-specific supercomputer simulations that will more precisely estimate the amount and distribution of radiation absorbed. "Usually, we apply supercomputers to advance fundamental science that is some distance from practical applications," said SDSC’s Amitava Majumdar. "It’s not often that you can use a supercomputer in a way that could eventually directly help a cancer patient."

Figure 1. Simulation and Tomography To verify supercomputer simulations using the SIMIND code, researchers have compared them to a known radiation distribution within an experimental patient (top), the radioactivity distribution measured with SPECT gamma tomography (center), and the simulated distribution using the SIMIND code (bottom). They found good agreement among simulation, measured, and true values.

Non-Hodgkin’s lymphomas develop in the lymphatic system, a network of thin tubes that branch like blood vessels into tissues and organs throughout the body. The tumor cells often crowd out normal cells within bone marrow, lymph nodes, and the thyroid gland. The usual chemotherapies for these tumors are nonspecific, indiscriminately harming normal and cancerous tissues alike. However, new radioimmune therapies treat patients with a source of radiation such as the isotope iodine-131 (131I) in a way that selectively targets the cancer cells, largely sparing normal tissue. This provides more effective treatment with fewer side effects.

Radioimmune therapies rely on monoclonal antibodies, proteins made in the laboratory by immortal lines of antibody-producing cells. Many monoclonal antibodies are used in cancer therapy; each recognizes and binds to a different protein on the cancer cell such as those of non-Hodgkin’s lymphoma. Monoclonal antibodies can be coupled, or conjugated, in the laboratory to a toxin, in this case 131I.

Once absorbed by the tumor, 131I performs two important functions. The radioactive isotope emits short-range beta particles, or electrons, which are ionizing and kill nearby cells. The 131I isotope also emits gamma photons that can pass through the body. They can be detected by a gamma camera, which rotates 360º around the patient’s body, like an X-ray CT scan, detecting the gamma photons at many positions. Reconstructing the data using Single Photon Emission Computed Tomography (SPECT) produces a 3-D image showing the location within the patient of the radiation and its quantity.

Michigan physicians Mark Kaminski and Richard Wahl have conducted phase II clinical trials of monoclonal antibodies conjugated with 131I to test the safety of the new drug and evaluate how well it works. In these trials, tumors disappeared in 48 of 76 non-Hodgkin’s lymphoma patients, and shrank at least 50 percent in another 26 patients. 131I has also shown promise in treating the advanced metastatic stage of the deadly childhood cancer, neuroblastoma.

Treating with Least Harm

The success of the clinical trials has spurred research sponsored by the National Cancer Institute to develop imaging methods capable of predicting absorbed radiation dose and revealing dose-response relationships. "Our goal is to be able to more accurately correlate the amount of radionuclide administered with the absorbed dose in the tumor and how the patient responds," said Dewaraja. The potential benefits of reliable early prediction include better screening to identify which patients will benefit from these treatments, and earlier planning of subsequent therapy. In addition, accurate predictions of the dose of radioactivity absorbed by healthy tissue will improve estimates of the maximum dose of such therapy the patient can tolerate. "While often less toxic than chemotherapy, physicians still need to know the maximum dose they can give without harming the patient," said Dewaraja.

When a patient diagnosed with non-Hodgkin’s lymphoma comes for treatment, physicians first order a standard X-ray CT scan to image the size and location of the patient’s tumor and internal organs. Next, a small tracer dose of about five millicuries of 131I-labeled antibody is given to help assess how much of a larger therapeutic dose will be absorbed by the patient. Then, in a series of steps the researchers are developing, physicians will combine the X-ray CT scan and the SPECT tomography derived from the tracer dose to guide the building of a digital representation of the individual patient’s body and tumor–a detailed 3-D virtual patient, or "voxel phantom."

Using the voxel phantom in a supercomputer simulation of a full therapeutic dose, the researchers will then be able to estimate the true radioactivity distribution in the real patient. Currently, the researchers are validating this simulation method, including developing better quantification of the SPECT gamma data. "Eventually, our research will also use simulations of the transport of electrons in the tumor to determine patient-specific radiation dose distributions," said Dewaraja. "This should be much more accurate because we will simulate the actual energy deposition in the tumor of each patient."

Figure 2. Virtual Patient This "voxel man" is a high-resolution, 3-D digital representation of a patient. The Zubal Phantom, obtained from researchers at Yale University and based on X-ray CT and MRI images, is being used by Michigan researchers to verify quantification of radioactivity uptake in tumor and organs by Single Photon Emission Computed Tomography (SPECT). The gray area is the collimator for the SPECT gamma camera.

Bigger Computers Are Best

To simulate the radioactivity distribution and gamma photon paths, the researchers began with the SIMIND serial Monte Carlo code, which runs on a single computer processor. Michael Ljungberg of Lund University in Sweden developed the code. The simulations model the path of each gamma photon–from its emission from an 131I isotope to its entry into the camera. In computing each photon’s path, the simulation must model absorption and scattering within the patient’s tissues. The Monte Carlo statistical method has proven highly accurate in modeling these interactions. It randomly assigns photon absorption and scattering events based on known probability density functions for how photons interact with the various types of tissue. In addition to requiring a highly accurate parallel random number generator, a major challenge in using the Monte Carlo method has been the long computation times required to simulate large numbers of photons.

"We found there were large errors even when using simple shapes like cylinders to approximate the tumor, especially for smaller tumors," said Dewaraja. That is, the radioactivity distribution depends sensitively on the specific geometry of the patient and tumor. For large, spherical tumors the quantification error of the estimate is less than 3 percent, but for a smaller or irregular tumor the error can reach 50 percent. Thus, accurate simulations require the use of a more realistic, individualized 3-D virtual patient.

Accurate simulations of the 131I radioactivity distribution are computationally demanding not only because of the need to use a high-resolution 3-D virtual patient, but also because of the need to use Monte Carlo methods to simulate millions of individual photons. Because of these factors, CPU time can exceed one month using the single-processor SIMIND code. To cut the time required, Dewaraja and Ljungberg collaborated with Abhijit Bose, also at NPACI partner Michigan, in developing a parallel version of the SIMIND code that initially ran on NPACI’s IBM SP2 at Michigan.

"The problem is embarrassingly parallel," said Bose. "Physically, it’s a number of independent events, so to run a large number of photon simulations you just divide them among the processors." Each processor performs the entire simulation for all photons assigned to it and reports the results to the host processor, which calculates the final result.

Even with an efficient parallel code, it’s necessary to simulate millions of photons to achieve sufficient accuracy, and the researchers found a major speed increase from running the parallel code. A simulation that would have taken a month with one processor, takes four to five hours on 512 processors of NPACI’s Blue Horizon. The researchers expect full-machine runs on the 1,152-processor Blue Horizon to take as little as two hours. "We anticipate it will easily scale to 1,000 processors, and more–it looks like a natural migration to the clusters of the TeraGrid," said Majumdar.

NPACI has also contributed to the research through a Strategic Applications Collaboration. In addition to integrating the Scalable Parallel Random Number Generator to enable accurate parallel Monte Carlo simulations, another important goal has been to bring the benefits of NPACI’s advanced parallel supercomputers to the medical imaging community, which until now has made little use of them.

Toward Clinical Application

"Previously, people in our field thought it was impossible to simulate an individual patient because it would simply take too much computer time," said Dewaraja. "But through the faster parallel code and having access to large NPACI machines we’re showing that it’s within sight to do realistic simulations for individual patients."

One of the researchers’ goals is to produce a tool that clinicians can use to accurately quantify the uptake and absorption of radioactivity in the tumor and healthy tissues of each patient. Other simulations will model the activity and dose of ionizing beta particles within the tumor. "But the most rewarding part is creating an application that may eventually have clinical use in helping cancer patients," said Dewaraja.
–PT