Combined metabolic and therapeutic imaging improves tumor therapy

Advanced computer modeling and artificial intelligence algorithms allow physicians treating prostate cancer to zero in on active tumor cells by mapping MR spectroscopy data onto ultrasound therapeutic images.

Standard treatment for prostate cancer manages the prostate as a single mass, using a single dose for the entire diseased organ. Experts have hypothesized, however, that elevated levels of choline are indicative of aggressive tumor activity, which would benefit from higher doses of radiotherapy.

"In our novel treatment, in addition to specifying a minimum dose that the entire prostate must receive, it is also possible to escalate the radiation dosage to areas containing more cancer cells using the biological cancer cell density information obtained via the MRS images," said Eva K. Lee, Ph.D., director of the Center for Operations Research in Medicine at the Georgia Institute of Technology.

Getting the biological information from an MRS scan and mapping it onto an ultrasound image is not a simple task. The endorectal probe used for gathering metabolic data is inflated with about 100 cc of air to ensure proper operation of the MR unit. Under these conditions, the prostate gland is pushed into the anterior position and assumes a flattened shape, said study collaborator Marco Zaider, Ph.D., a professor of physics in radiology at the Memorial Sloan-Kettering Cancer Center. Tumor treatment is then performed using ultrasound images obtained with the prostate in its normal uncompressed state.

"Assume you have a rubber ball inside of which you place a small metal sphere at a known location. Now change the shape of the ball by squeezing it and try to discover - by looking only at is new shape and not inside the ball - where the small metal sphere went. The algorithm we developed provides the answer to this question," Zaider said.

The algorithm and treatment system Lee and Zaider created was described at the 2004 American Society for Therapeutic Radiology and Oncology meeting. The algorithm takes information from the MRS prostate contour with outlined areas of high tumor cell proliferation, as well as from the ultrasound images of the prostate used for treatment.

Melding the MRS and ultrasound data together involves the use of a nonrigid body point matching algorithm and artificial intelligence routines to transform MRS contours onto the ultrasound contours. The transformation is quick, requiring less than one CPU minute to perform. The researchers reported less than 0.1% error in morphing shape.

The researchers used a planned target volume dose of 144 Gy using iodine-125 seeds for 15 patients. They placed strict dose bounds of 100% to 120% of the prescribed dose on healthy organs such as the urethra. Using the MRS imaging information, they were able to provide an escalation of dose of at least 120% to be placed on identified tumor pockets.

Even when a tumor pocket receiving higher radiation therapy dosage was close to the urethra, the dose received by the urethra was kept within strict predetermined dosage bounds. The technique allowed a gain in tumor control probability from 65% to 95%.

The technique has been used in 200 patients at MSKCC. Future clinical studies are needed to validate its importance and to measure the potential gain in clinical outcome. The technique could be ideal for mapping images of lung tumor volumes to account for respiratory motion or shrinking of tumor volume between day-to-day treatment fractions. It could also be used for registering MR images in brain mappings,and recovering dynamic heart motion in cardiac image analysis, Lee said.