Atlas-Based Dose Optimization Via Linear Programming For Imrt Using A High-Throughput Computing Environment

To develop and evaluate an atlas-based dose optimization (DO) scheme using linear programming (LP) for disease site specific IMRT treatment planning using a high-throughput computing environment (HTC). Ten locally advanced head and neck cases were retrospectively selected for this study. The treatment targets involved primary tumor, high risk and low risk nodal volumes. Organs-at-risk (OARs) included both parotids and the spinal cord. Seven-beam IMRT plans were generated using the Pinnacle3 planning system. More than 2000 apertures from these 10 cases were extracted (on average 30 apertures per beam per case). These apertures served as the basis for the atlas from which prospective case-specific apertures were derived. Aperture-based 3D dose distributions were calculated in parallel using Monte Carlo-based Kernel Superposition (MCKS) and facilitated using Condor in the HTC environment consisting of 48 2.26GHz CPUs. Condor provided the job queuing mechanism, priority scheme, resource monitoring, and resource management. Aperture weights of each case were optimized via an in-house developed sequential linear program (SLP) DO approach. The SLP approach involved solving the DO problem by dividing the problem into sub-problems (due to the size of the DO problem), each of which was solved using LP while maintaining a global view of the full problem. Our approach was compared with three approaches: conventional IMRT, two-arc intensity-modulated arc therapy (IMAT) (generated using Pinnacle SmartArc 9.0) and SLP corresponding to apertures derived from the conventional IMRT plans for the specific case. Approximately 200 apertures per case were returned with non-zero weight assignments via the atlas-based SLP scheme and 40% of the selected atlas apertures were from other cases. Superior plan quality was achieved with atlas-based plans. On average, at dose constraint levels parotid gland sparing improved 30% (±7%), 31% (±9%) and 18% (±5%), respectively using the atlas-based DO scheme compared with conventional IMRT, two-arc IMAT and the SLP scheme using apertures derived from the conventional IMRT plans. The improvement in the maximum dose to the spinal cord was 10% (±4%), 20% (±6%) and 6% (±2%), respectively when compared with the other three approaches. t-test results showed that these improvements were significant (p < 0.001). Dose homogeneity for all PTVs (tumor and nodes) was higher for the atlas-based plans compared with the other approaches. The atlas-based approach also resulted in lower integral dose to non-PTV and non-OAR tissues. Our results suggest that atlas-based DO using the SLP approach results in superior plan quality compared with conventional IMRT and IMAT.