GOVERNMENT

Remarkable collaboration with M.D. Anderson Cancer Center enables Lonestar to carry out real-time, data-driven treatment: With the dog in the MRI machine, the supercomputer prepared for surgery. The procedure was the culmination of three years of research and development into the algorithms, computer codes, imaging technology, and cyberinfrastructure that would allow a supercomputer in Austin to perform a minimally invasive laser treatment on a canine in Houston, without the intervention of a surgeon. The scientists took a collective breath. “We had a fifteen minute window in which a million things had to go right for this treatment to be successful,” explained David Fuentes, a post-doctoral student at The University of Texas at Austin’s Institute for Computational Engineering and Sciences (ICES), and the central developer of the project. “There had to be no flaw, no silly bug, everything had to go perfectly. And if that wasn’t complicated enough, you add the complexity of a living animal. This is a pretty formidable problem.” And yet, in April 2008, when the researchers performed the first full run of the system on a canine subject, the coordination went off without a hitch, proving the potential of supercomputers for patient-specific treatments and blazing a path to next-generation cyber-surgical methods. Sponsored by a National Science Foundation "Dynamic Data Driven Application Systems” grant to ICES, the project brought together computational experts from UT Austin, cyberinfrastructure specialists and systems from the Texas Advanced Computing Center (TACC), and leading technologists from the M.D. Anderson Cancer Center in Houston. Together, they created a mechanism for real-time laser surgery, using the Lonestar supercomputer, which can sense how the patient is responding to the treatment and make adjustments accordingly. Depicted is the real-time visualization provided during the laser treatment. The color scale shown at the left is from 35-65 degrees Celsius. To watch the evolving visualization as it appeared to the researchers, click here. Visualizations courtesy of David Fuentes. “It’s been an extremely challenging problem that’s met one unresolved open problem after another, solved it and pushed forward,” J. Tinsley Oden, director of ICES and principal investigator of the project, said. “And now we have a system that’s working.” Improving on Real-Time Treatment Methods Prostate cancer is a leading killer of men in the United States. When caught early, the disease is treatable, but current surgical methods lead to long, painful recoveries and sometimes, impotence. “The current surgical methods are frequently ad hoc,” Oden said. “Computer modeling and simulation, on the other hand, can provide a method of predicting the effects of the physical environment on physical events in real-time. That’s something that’s completely new in the medical profession.” The laser cancer treatment project uses the massive parallel processing power of supercomputers like Lonestar to perform real-time, patient-specific surgery remotely, in a way that responds to data-intensive monitoring methods. Using precise lasers, state-of-the-art thermal imaging technology, and computational methods that synthesize complex information in a fraction of a second, dynamic, data-driven laser treatments are being pursued as a minimally invasive alternative to the standard treatment of cancer. “We’re basically bringing engineering tools into medicine,” Oden said. “We’re making surgery an engineering or mathematical process.” Killing Cancer Cells When a cancerous cell is heated or cooled beyond a certain threshold, it dies. With the new data-driven surgical treatment, an interstitial laser fiber is inserted into cancerous tissue where it raises the temperature within a tumor in a controlled way. Schematic of the peer to peer communication architecture used to control the laser treatment process. Feedback control is achieved through the continual interaction of the data, compute, and visualization modules. Image courtesy of David Fuentes. However, to do this successfully, scientists needed to develop precise models that showed how a laser conducts heat across tissue and blood, and also assess how well it can be controlled. Fuentes translated Pennes’ bio-heat transfer model (published in 1948) into a parallelized computer code that simulates how a temperature field evolves through tissue with the application of a laser heat source in time. “Real-time control and real-time feedback based on imaging acquisition are really hot topics in clinical use,” UT Austin’s Fuentes said. “The current state-of-the-art is real-time monitoring using thermal images where you manually control the laser, look at the real-time results and decide from there how to control the laser. The next step is adding computational prediction.” The team’s methods do just that, providing a controlled, testable framework for planning the future development of the protocol. How It Works Several days before the surgery, the patient — in the test case, a canine — receives an initial MRI that provides the topography of the medical region of interest (here, the prostate). Using the data from the MRI and software available at TACC and ICES, a hexahedral mesh representing the biological domain as a three-dimensional model is created and laser parameter pre-optimization begins. In cancer treatment, optimization means more than just determining where to point the laser and for how long. Doing maximum damage to the tumor must be balanced with protecting healthy tissue, while simultaneously minimizing heat-shock proteins, whose expression can prevent tumor eradication. While Lonestar is crunching all the possibilities and variables for a successful treatment, Fuentes reserves computing time on the system for the day of the surgery and tests the optimization code repeatedly to make sure that everything will run smoothly within the fifteen-minute window of the experimental surgery. During this time period, a suite of visualization software is also repeatedly tested. This software, run on a remote visualization system at TACC, enables the researchers to monitor the progress of the surgery from multiple perspectives and in real time. On the day of the surgery the subject is re-scanned, and the optimal treatment is recalibrated based on the newest images. The treatment itself is broken into four stages: 1) Lonestar instructs the laser to heat the domain with a non-damaging calibration pulse; 2) the thermal MRI acquires baseline images of the heating and cooling of the patient’s tissue for model calibration; 3) Lonestar inputs this patient-specific information and recomputes the optimal power profile for the rest of the treatments; and 4) surgery begins, with remote visualizations and evolving predictions continuing throughout the procedure. “This is a laser treatment where the surgery is ongoing in Houston on the canine, and the laser is being controlled by Lonestar in Austin for the entire duration of the treatment,” Fuentes explained. “The data is orchestrated so every time a new set of thermal images is sent from Houston to Austin, the power control for the next five seconds is sent from Austin to Houston, and it’s done that way for the duration of the treatment.” Patient-specific medicine means reacting to real conditions that can only be discovered during surgery. The original, optimal profile might be quite accurate, but it was important to the team to develop methods that can adjust to the effects of each unique treatment. “We’ve established the feasibility of using computer models to predict the outcome of physical experiments on living tissue, and that in itself is a great triumph that will have a tremendous impact on cancer research and on surgery in general,” Oden said. The Future of Surgery According to everyone involved, the procedure was a medical, technical, and computing breakthrough. Although the dog gave his life to the research, his sacrifice furthers science by allowing researchers to assess the success of the treatment and plan improvements. “It’s a long process before these protocols are made robust and have wide-spread use in human subjects. But this is a step along a path that will be followed,” Oden said. “I’m confident that this process can lead to specific treatments in five to ten years.” As MRI devices, methods for mesh generation, and temperature modeling become more sophisticated over the comings years, the quality of the procedure will inevitably improve. But without high-performance computing centers like TACC, data-driven medical procedures like this one would be impossible. “A very important component of the project was having access to the computers at TACC,” Oden said. “The whole process requires that it be done in near real-time and that depends significantly on the use of parallel computing.” Laser surgery of prostate cancer isn’t the only treatment to which these computational methods and medical protocols can be applied, Fuentes explained. “From a computational point of view, if you want to switch to a variety of heating or cooling sources with a variety of imaging, it would just involve changing the computational model slightly,” he said. Cryotherapy, microwave, radio frequency and ultrasound treatments could all be adapted to work in a similar way, broadening the project’s reach significantly and extending computer-enabled medicine. “Years from now, if this technique were to reduce the number of cancer-related deaths, and improve the quality of life of patients, it will be a significant achievement for computational science,” Fuentes said. ************************************************************** Further reading: "Nanoshell-Mediated Laser Surgery Simulation for Prostate Cancer Treatment " "Computational Infrastructure for the Real-Time Patient-Specific Treatment of Cancer" Aaron Dubrow Texas Advanced Computing Center Science and Technology Writer