MIT software improves accuracy of robot used in cancer therapy


A robotic system that will maneuver patients into position for a beam of cancer-blasting particles is now about 15 times more accurate, thanks to software developed by MIT engineers.

Original tests of the system, which is a crucial part of an upcoming cancer treatment center at Massachusetts General Hospital, showed that it could maneuver patients to within seven or eight millimeters of a given position -- but this was far from the desired accuracy of half a millimeter. The MIT software automatically corrects for the inherent errors responsible for the problem, refining the system's accuracy to less than 0.33 mm, which is even better than the original goal.

The software could also have applications for other large robotic systems that must be very accurate. One example: the robots that perform inspection and repair work inside nuclear power facilities.

The new robotic system will be part of MGH's Northeast Proton Therapy Center, scheduled to open in 1999, in which subatomic particles -- protons -- obliterate cancerous tissue. According to MGH literature, protons "can concentrate their dose in a tumor better than conventional (high-energy X-ray) radiation."

Proton beam therapy is available at some 13 centers around the world, including two in the United States. MGH's upcoming center will feature advances including a more precise proton beam. "They wanted to bring the accuracy down to half a millimeter. That's important for certain types of cancers, especially inoperable brain tumors," said Professor Steven Dubowsky of MIT's Department of Mechanical Engineering, who served on the hospital's advisory board for the center.

About three years ago, however, when Professor Dubowsky reviewed the robotic patient positioning system (PPS) that would be key to this accuracy, "it was clear that it wouldn't be precise enough," he said. It is difficult to create a large, powerful robotic system that is also very accurate, he explained. Tiny errors in the construction of such a robot or deformations introduced during its use are magnified, resulting in large "end-effector" errors leading to a low precision.

The PPS was no different. The system involves placing a patient at the end of a robotic arm that's approximately 12 feet long. The arm then moves the patient into the treatment area and orients him or her in the best possible position for treatment. The accuracy of this positioning is influenced not only by tiny construction errors but also by "the bending of the robotic arm due to the weight of the patient," Professor Dubowsky said.

He and colleagues developed software that automatically corrects for these errors. "A computer monitors the position of the robot, the patient's weight, and how that weight is applied to the robotic arm," Professor Dubowsky said. With that information in hand, "it goes back to the software [we developed] and says, 'given these parameters, what are the errors introduced, and what should I do to move the patient back to the correct position?'"

To develop the software, the team first took a series of measurements of the system to characterize it. This sounds straightforward, but actually involved a separate challenge.

"Usually a system this complicated [the robot can move in six degrees of freedom] requires millions and millions of measurements that would take millions of years to make," Professor Dubowsky explained. The MIT team solved the problem by developing mathematical methods that allow them to characterize the system with less than 400 measurements made in just a few hours.

OTHER APPLICATIONS

Beyond the current application for the MGH positioning system, the new software could also be useful for a variety of other large robotic systems that must be very accurate. For example, Professor Dubowsky said, the French are interested in applying the software tothe robots that perform inspection and repair work inside nuclear power facilities.

Professor Dubowsky's colleagues on the project include Constantinos Mavroidis (now at Rutgers University) and graduate students Marco A. Meggiolaro and Phillip Drouet. They have reported the work in the Proceedings of the Sixth International Symposium on Advances in Robot Kinematics (June 1998), the Proceedings of the ASME Design Engineering Technical Conferences (September 1998), a recent issue of Robotics Computer Integrated Manufacturing and other publications.

Dr. Michael Goitein of MGH is project director for the Northeast Proton Therapy Center.

The work was funded by Massachusetts General Hospital with support from the NIH. Additional funding came from the Brazilian and French governments via fellowships to Mr. Meggiolaro and Mr. Drouet, respectively.

A version of this article appeared in MIT Tech Talk on February 10, 1999.


Topics: Cancer, Computer science and technology, Health sciences and technology, Mechanical engineering, Artificial intelligence

Back to the top