Automatic IMRT Treatment Planning Using Templates with Automated Optimization Tuning Methods

One rate-limiting step from CT simulation to treatment delivery is treatment planning, particularly labor-intensive dosimetric design and calculation. This study compares an automated treatment planning module versus physicist-optimized clinical plans for lung cancer radiation therapy. For 15 patients with locally advanced lung cancer, clinical and automatic IMRT (11) and VMAT (4) plans were analyzed. Manual contours - gross, clinical and planning target volumes (GTV, CTV and PTV) as well as organs at risk (OAR) and planning goals, such as PTV coverage and tolerance doses of OARs, were defined by one physician according to department guidelines. Clinical plans were generated by experienced medical physicists. Automatic plans were created independently utilizing a template configured with auto-planning optimization goals based on both dose-volume and equivalent uniform dose (EUD)-based objectives. The technical parameters of the template included beam, dose, and visualization settings except for beam angles. The latter were unique to each individual patient and were dependent on tumor location. The auto-planning algorithm performed multiple optimization loops that iteratively adjusted parameters to meet constraints and spare OARs with minimal compromise to target coverage conformality and uniformity. The first 10 cases were used to create and “train” the template. Once “trained”, the template was used to create automatic plans for 5 additional patients. An independent physician compared clinical and automatic plans using: CTV/PTV coverage and dose conformality; Dmax of Cord and Brachial plexus; Lung mean dose,V20 Gy and V30 Gy; Esophagus mean dose and V60 Gy; Heart mean dose and V40 Gy. All dose parameters revealed only minor, non-significant differences (p > 0.05 on t-test) between clinical and automatic plans for both post “training” cases and the entire cohort. No significant difference in PTV dose conformality was identified (p value = 0.54). Both clinical and automatic plans met the majority of dose constraints in the 5 post “training” cases, the automated plans did not meet only 2 constraints that were achieved by the clinical plans (Brachial Plexus and CTV). The average time for automatic planning (Enterprise Server) once the template was trained was 23 minutes (range, 17 to 34) compared to an average of 4 hours (range, 3-8 hr) for the clinical plans. Automatic planning simplifies the planning process through the use of templates that can be “trained” according to department policies. Automatic planning provided a clinically important reduction in plan optimization time while producing plans of similar quality as that of clinical plans.