Radiation therapy of GBM is usually delivered using a uniform prescription dose with large clinical target margins. However, this doesn't account for patient-specific factors such as diffuse disease distribution or tumour heterogeneity. To address this shortfall, a research team from the University of Washington Medical Center (Seattle, WA) has developed an intensity-modulated radiotherapy (IMRT) optimization algorithm that incorporates patient-specific radiobiological information (Phys. Med. Biol. 57 8271).

Optimization algorithm

To perform IMRT optimization, the researchers used a multi-objective evolutionary algorithm (MOEA), which results in a set of IMRT plans that represent optimal trade-offs among all clinical objectives. They combined this algorithm with a GBM model describing spatial and temporal changes in tumour cell density. The GBM model quantifies rates of proliferation and invasion using patient-specific parameters derived from MR images.

"Patient-specific factors for the GBM model can be obtained using the initial patient MRI and an MRI at the beginning of treatment, due to the correlation between rate of growth and radio-sensitivity found in previous work,"explained lead author Clay Holdsworth (now at Brigham and Women's Hospital in Boston, MA).

The resulting algorithm was used to perform weekly IMRT optimizations based on predicted changes in the tumour arising from radiation-induced cell kill, tumour diffusion and proliferation.

"Although it requires extra computation time, there is actually less manual labour in developing each plan," Holdsworth noted. "After the treatment planner does the contouring and defines the decision criteria, the algorithm will determine the optimal weights and penalty objectives without human interaction. Where work might be added is in the adaptive part of this project. Since a new plan is developed for each week, this means that six or seven plans must be developed instead of one."

Plan comparison

The researchers simulated seven weeks of IMRT for a patient with relatively radio-resistant GBM and a patient with relatively radio-sensitive disease. They examined five different MOEA optimizations, each based on three decision criteria. All simulations were optimized to minimize the maximum dose per fraction to any voxel and the equivalent uniform dose (EUD) to normal tissue.

Three of the plans also minimized tumour cell survival after one week of treatment, while restricting the maximum dose per fraction to any voxel to 3 Gy, 8 Gy or no limit. The other two were optimized for maximum tumour cell kill or minimum tumour survival 11 weeks after one week of treatment, both using an 8 Gy dose restriction.

The researchers first assessed the predicted tumour radius (as visible on T2 MRI) over time for the MOEA optimizations, as well as for a standard clinical plan that delivered 61.2 Gy in 1.8 Gy daily fractions over seven weeks. They also examined the "treatment gain", defined as the added number of days for the tumour radius to reach 4 cm following radiotherapy, compared with an untreated control.

For both patients, plans with a 3 Gy dose limitation showed worse performance than the standard plan, although with greatly reduced EUD to normal tissue. Increasing the maximum dose restriction boosted the tumour control. For the radio-resistant patient, the other four optimized plans improved treatment gain from 27 days for the standard plan to between 77 and 90 days.

In the more radio-sensitive patient, tumour response depended upon both the maximum dose and the decision criterion used. With no dose limit, treatment gain improved from 68 days for the standard plan to 157 days, though this criterion is not suitable for clinical use. Optimization based on tumour survival 11 weeks after a week of treatment gave a treatment gain of 153 days, while the other two plans provided gains of 111 and 113 days.

The finding that the plan based on tumour survival 11 weeks after a week of treatment exhibited similar tumour control to the plan with no maximum dose, but with a much lower dose, suggests that this represents the best target objective for use in the MOEA. This result also implies that the location of tumour cells targeted for radiotherapy is important for long-term control and that targeting glioblastoma cells at or near the tumour edge could prove as effective as delivering high doses to the tumour bulk.

Dose-volume histograms of the normal brain cell population revealed a reduced dose to normal tissue for all MOEA plans compared with the standard plan.

Based on results generated by the mathematical model, the authors conclude that the standard delivery of uniform dose to the bulk of the tumour with a 2.5 cm margin could potentially be improved, both in terms of tumour control and normal tissue dose. MOEA-optimized plans demonstrated a potential for improved tumour control and lower normal tissue EUD over current standard clinical practice.

Related articles in PMB
Adaptive IMRT using a multiobjective evolutionary algorithm integrated with a diffusion–invasion model of glioblastoma
C H Holdsworth et al Phys. Med. Biol. 57 8271
Predicting the efficacy of radiotherapy in individual glioblastoma patients in vivo: a mathematical modeling approach R Rockne et al Phys. Med. Biol. 55 3271