Advertisement

Automated multi-criterial planning with beam angle optimization to establish non-coplanar VMAT class solutions for nasopharyngeal carcinoma

Open AccessPublished:July 16, 2022DOI:https://doi.org/10.1016/j.ejmp.2022.06.017

      Highlights

      • Automated planning was used to select non-coplanar IMRT beams to complement VMAT.
      • Developed arc class solutions (CS) had only 1 or 2 added non-coplanar VMAT arcs.
      • With CSs, OAR doses were overall superior compared to only coplanar VMAT.
      • OAR gains with CSs were highly patient-specific.

      Abstract

      Purpose

      Complexity in selecting optimal non-coplanar beam setups and prolonged delivery times may hamper the use of non-coplanar treatments for nasopharyngeal carcinoma (NPC). Automated multi-criterial planning with integrated beam angle optimization was used to define non-coplanar VMAT class solutions (CSs), each consisting of a coplanar arc and additional 1 or 2 fixed, non-coplanar partial arcs.

      Methods

      Automated planning was used to generate a coplanar VMAT plan with 5 complementary computer-optimized non-coplanar IMRT beams (VMAT+5) for each of the 20 included patients. Subsequently, the frequency distribution of the 100 patient-specific non-coplanar IMRT beam directions was used to select non-coplanar arcs for supplementing coplanar VMAT. A second investigated CS with only one non-coplanar arc consisted of coplanar VMAT plus a partial arc at 90° couch angle (VMATCS90). Plans generated with the two VMATCSs were compared to coplanar VMAT.

      Results

      VMAT+5 analysis resulted in VMATCS60: two partial non-coplanar arcs at couch angles 60° and −60° to complement coplanar VMAT. Compared to coplanar VMAT, the non-coplanar VMATCS60 and VMATCS90 yielded substantial average dose reductions in OARs associated with xerostomia and dysphagia, i.e., parotids, submandibular glands, oral cavity and swallowing muscles (p < 0.05) for the same PTV coverage and without violating hard constraints. Impact of non-coplanar treatment and superiority of either VMACS60 and VMATCS90 was highly patient dependent.

      Conclusions

      Compared to coplanar VMAT, dose to OARs was substantially reduced with a CS with one or two non-coplanar arcs. Preferences for coplanar or one or two additional arcs are highly patient-specific, balancing plan quality and treatment time.

      Keywords

      1. Introduction

      Nasopharyngeal carcinoma (NPC) tumors have planning target volumes (PTV) surrounded by many organs-at-risk (OARs). The standard of care for NPC is coplanar volumetric arc therapy (VMAT) [
      • Chen Y.P.
      • Chan A.T.C.
      • Le Q.T.
      • Blanchard P.
      • Sun Y.
      • Ma J.
      Nasopharyngeal carcinoma.
      ], which is known to result in fast treatments with high coverage for irregular targets [
      • Rana S.
      Intensity modulated radiation therapy versus volumetric intensity modulated arc therapy.
      ,
      • Teoh M.
      • Clark C.H.
      • Wood K.
      • Whitaker S.
      • Nisbet A.
      Volumetric modulated arc therapy: A review of current literature and clinical use in practice.
      ], while limiting excessive dose to the surrounding OARs. The most common radiation induced complications are associated excessive dose in the parotid glans, submandibular glands, oral cavity or in the swallowing muscles, which are OARs associated with xerostomia and dysphagia. Studies on several treatment sites have shown that non-coplanar beam arrangements could lead to improved healthy tissue sparing [
      • Wild E.
      • Bangert M.
      • Nill S.
      • Oelfke U.
      Noncoplanar VMAT for nasopharyngeal tumors: Plan quality versus treatment time.
      ,
      • Sharbo G.
      • Hashemi B.
      • Bakhshandeh M.
      • Rakhsha A.
      Radiobiological assessment of nasopharyngeal cancer IMRT using various collimator angles and non-coplanar fields.
      ,
      • Bangert M.
      • Ziegenhein P.
      • Oelfke U.
      Characterizing the combinatorial beam angle selection problem.
      ,
      • Bangert M.
      • Ziegenhein P.
      • Oelfke U.
      Comparison of beam angle selection strategies for intracranial IMRT.
      ,
      • Rossi L.
      • Breedveld S.
      • Heijmen B.J.M.
      • Voet P.W.J.
      • Lanconelli N.
      • Aluwini S.
      On the beam direction search space in computerized non-coplanar beam angle optimization for IMRT - Prostate SBRT.
      ,
      • Rocha H.
      • Dias J.M.
      • Ferreira B.C.
      • Lopes M.C.
      Beam angle optimization for intensity-modulated radiation therapy using a guided pattern search method.
      ], possibly minimizing side-effects.
      Wild et al. [
      • Wild E.
      • Bangert M.
      • Nill S.
      • Oelfke U.
      Noncoplanar VMAT for nasopharyngeal tumors: Plan quality versus treatment time.
      ] compared different coplanar and non-coplanar treatment techniques for nasopharyngeal carcinoma. They found that optimized non-coplanar VMAT treatments reduced mean and maximum doses in the OARs analyzed (eyes, optic nerves and chiasm) compared to coplanar VMAT, while maintaining target coverage. Another method that resulted in improved OAR sparing for NPC was the combination of VMAT and intensity modulated radiotherapy (IMRT) in different fractions for the same patient. Akbas et al. [
      • Akbas U.
      • Koksal C.
      • Kesen N.D.
      • Ozkaya K.
      • Bilge H.
      • Altun M.
      Nasopharyngeal carcinoma radiotherapy with hybrid technique.
      ] compared IMRT, VMAT and a hybrid IMRT and VMAT technique (each used in different fractions). This hybrid solution resulted in superior conformity for the targets and improved sparing of the brainstem and spinal cord.
      Some studies have investigated the optimization of dynamic non-coplanar VMAT trajectories, where the gantry and couch rotate simultaneously, as a treatment option for nasopharyngeal carcinoma [
      • Wild E.
      • Bangert M.
      • Nill S.
      • Oelfke U.
      Noncoplanar VMAT for nasopharyngeal tumors: Plan quality versus treatment time.
      ,
      • Rocha H.
      • Dias J.M.
      • Ferreira B.C.
      • Lopes M.C.
      Beam angle optimization for intensity-modulated radiation therapy using a guided pattern search method.
      ]. In both these studies the non-coplanar plans outperformed the coplanar plans, but treatment optimization times were very long. Moreover, these dynamic treatments can currently not be clinically delivered on regular linacs. Instead, optimization times for fixed non-coplanar VMAT arcs are acceptable and clinical application is feasible. To limit treatment times, the use of only few non-coplanar arcs per patient is then preferred.
      Selection of optimal patient-specific couch angles for non-coplanar arcs is complex and may be highly workload intensive in case of manual planning. In principle, a fixed arc-class-solution (CS) to be used for all patients could possibly be a solution in clinical planning. However, development of such a CS with manual trial-and-error planning would again be arduous and time-consuming. As far as we are aware of, the use of automated planning for development of VMAT CSs, consisting of a regular coplanar arc supplemented with a few fixed non-coplanar arcs, has never been investigated for NPC.
      To improve dose distributions while keeping delivery times acceptable, we previously proposed the VMAT+ treatment approach, complementing coplanar VMAT with ≤ 5 static patient-specific non-coplanar IMRT beams, and tested it for liver SBRT [
      • Sharfo A.W.M.
      • Dirkx M.L.P.
      • Breedveld S.
      • Romero A.M.
      • Heijmen B.J.M.
      VMAT plus a few computer-optimized non-coplanar IMRT beams (VMAT+) tested for liver SBRT.
      ] and prostate SBRT [
      • Sharfo A.W.M.
      • Rossi L.
      • Dirkx M.L.P.
      • Breedveld S.
      • Aluwini S.
      • Heijmen B.J.M.
      Complementing prostate SBRT VMAT with a two-beam non-coplanar IMRT class solution to enhance rectum and bladder sparing with minimum increase in treatment time.
      ]. Adding the fully automatically selected non-coplanar IMRT beams improved OAR sparing compared to coplanar VMAT. For prostate, VMAT+5 plans were used to derive a fixed 2-beam non-coplanar CS for the whole patient population to complement coplanar VMAT (VMAT+CS). VMAT+CS resulted in similar plan quality as VMAT+5, while making both planning and delivery faster.
      In this study, our in-house developed algorithm for automated multi-criterial plan generation with integrated beam angle optimization (BAO) was used for NPC to explore development and assess added value of VMAT CSs with few fixed, non-coplanar arcs supplementing regular coplanar VMAT. VMAT+5 plans were used for selection of non-coplanar arcs. All generated plans are clinically deliverable on regular linacs.

      2. Materials and Methods

      2.1 Global study design

      First, our in-house automated treatment planning system (TPS) was configured for NPC, in line with our clinical NPC planning protocol. Then the configuration was validated by comparison of clinically applied coplanar VMAT plans of previously treated patients (ClinVMAT) with corresponding automatically generated coplanar VMAT plans (AutoVMAT), to ensure use of high-quality automatically generated plans in this study. Next, the system was used to automatically generate VMAT+5 plans for the included patients, each consisting of coplanar VMAT supplemented with 5 non-coplanar IMRT beams with optimized patient-specific beam directions. Based on the overall distribution of selected patient-specific beam angles for the VMAT+5 plans, we then defined a non-coplanar VMAT CS, aiming at a reduction in the number of couch rotations during treatment compared to VMAT+5 (less than four). A second CS was defined as coplanar VMAT with one additional partial non-coplanar VMAT arc at couch angle 90° (for symmetry reasons, as targets can be left-sided, central and right-sided). Partial non-coplanar arcs were used for linac-patient collision avoidance. Plans generated with the two CSs were compared with coplanar VMAT plans. The Rating guidelines for treatment planning studies [
      • Hansen C.R.
      • Crijns W.
      • Hussein M.
      • Rossi L.
      • Gallego P.
      • Verbakel W.
      • et al.
      Radiotherapy Treatment plannINg study Guidelines (RATING): A framework for setting up and reporting on scientific treatment planning studies.
      ] were used for study design and writing of the paper.

      2.2 Patients and clinical protocol

      In this study we used contoured planning CT-scans of 20 NPC patients that were recently treated at our center, all fulfilling the clinically used planning constraints, the prescribed PTV coverage, and treated with coplanar VMAT.
      All patients were planned with a simultaneous integrated boost scheme, prescribing 70 Gy to the primary tumor and pathological lymph nodes (PTV70), and 54.25 Gy to the elective nodal areas (PTV54.25), delivered in 35 fractions. PTV54.25 was created by uniform expansions of the primary CTVs and the elective lymph node CTVs by 0.5 cm (clipped at the patient surface by 0.5 cm). PTV70 consisted of the primary CTVs, each expanded with a 0.5 cm margin (clipped at the patient surface by 0.5 cm). Average PTV70 and PTV54.25 volumes were 349 cc (range: 86–812 cc) and 751 cc (range: 184–1281 cc), respectively. The two PTVs of all included patients are presented in Figure A1 of Electronic Supplement A.
      Dosimetric aims of the clinically applied planning protocol are presented in Table 1. The aim was to deliver 95% of the prescribed dose to at least 98% of the PTVs (V95%>98%), while maintaining PTV70 V107% below 2 cc (V107%<2 cc). Hard constraints were set for maximum doses of serial OARs: spinal cord, brainstem, brain, optical nerves, chiasm, retina, and mandible. While respecting the hard constraints, doses in parotid glands, submandibular glands (SMGs), oral cavity, swallowing muscles (organs associated with xerostomia and dysphagia), larynx, esophagus, cochleas and lenses were minimized, with parotid glands, SMGs, oral cavity, and swallowing muscles having highest priority.
      Table 1Clinically used planning aims for nasopharyngeal carcinoma RT.
      StructureGoals
      PTV70V95% > 98%
      V107% < 2 cc
      PTV54.25V95% > 98%
      Spinal cordD0.03cc < 50 Gy
      BrainstemD0.03cc < 60 Gy
      BrainD0.03cc < 70 Gy
      MandibleD0.03cc < 70 Gy
      Optical nervesD0.03cc < 55 Gy
      ChiasmD0.03cc < 55 Gy
      RetinaD0.03cc < 45 Gy
      ParotidsDmean < 26 Gy
      Submandibular glandsDmean < 39 Gy
      Oral cavityDmean < 50 Gy
      Swallowing musclesDmean < 55 Gy
      LarynxDmean < 45 Gy
      EsophagusDmean < 60 Gy
      CochleasDmean < 45 Gy
      LensesD0.03cc < 5 Gy
      Automated plan generation was validated by comparing the AutoVMAT plans with their corresponding ClinVMAT plans.

      2.3 System for automated plan generation

      All plans used for this study were automatically generated with Erasmus-iCycle [
      • Heijmen B.
      • Voet P.
      • Fransen D.
      • Penninkhof J.
      • Milder M.
      • Akhiat H.
      • et al.
      Fully automated, multi-criterial planning for Volumetric Modulated Arc Therapy – An international multi-center validation for prostate cancer.
      ,
      • Breedveld S.
      • Storchi P.R.M.
      • Voet P.W.J.
      • Heijmen B.J.M.
      ICycle: Integrated, multicriterial beam angle, and profile optimization for generation of coplanar and noncoplanar IMRT plans.
      ]. For many tumor sites, this system has demonstrated superiority of automatically generated plans, over manually generated plans [
      • Heijmen B.
      • Voet P.
      • Fransen D.
      • Penninkhof J.
      • Milder M.
      • Akhiat H.
      • et al.
      Fully automated, multi-criterial planning for Volumetric Modulated Arc Therapy – An international multi-center validation for prostate cancer.
      ,
      • Voet P.W.J.
      • Dirkx M.L.P.
      • Breedveld S.
      • Fransen D.
      • Levendag P.C.
      • Heijmen B.J.M.
      Toward fully automated multicriterial plan generation: A prospective clinical study.
      ,
      • Voet P.W.J.
      • Dirkx M.L.P.
      • Breedveld S.
      • Al-Mamgani A.
      • Incrocci L.
      • Heijmen B.J.M.
      Fully automated volumetric modulated arc therapy plan generation for prostate cancer patients.
      ,
      • Sharfo A.W.M.
      • Voet P.W.J.
      • Breedveld S.
      • Mens J.W.M.
      • Hoogeman M.S.
      • Heijmen B.J.M.
      Comparison of VMAT and IMRT strategies for cervical cancer patients using automated planning.
      ,
      • Sharfo A.W.M.
      • Breedveld S.
      • Voet P.W.J.
      • Heijkoop S.T.
      • Mens J.M.
      • Hoogeman M.S.
      Validation of fully automated VMAT plan generation for library-based plan-of-the-day cervical cancer radiotherapy.
      ,
      • Della Gala G.
      • Dirkx M.L.P.
      • Hoekstra N.
      • Fransen D.
      • Lanconelli N.
      • van de Pol M.
      • et al.
      Fully automated VMAT treatment planning for advanced-stage NSCLC patientsVollautomatische VMAT-Behandlungsplanung für Patienten mit fortgeschrittenem NSCLC.
      ,
      • Buergy D.
      • Sharfo A.W.M.
      • Heijmen B.J.M.
      • Voet P.W.J.
      • Breedveld S.
      • Wenz F.
      • et al.
      Fully automated treatment planning of spinal metastases - A comparison to manual planning of Volumetric Modulated Arc Therapy for conventionally fractionated irradiation.
      ,
      • Sharfo A.W.M.
      • Dirkx M.L.P.
      • Bijman R.G.
      • Schillemans W.
      • Breedveld S.
      • Aluwini S.
      • et al.
      Late toxicity in the randomized multicenter HYPRO trial for prostate cancer analyzed with automated treatment planning.
      ,
      • Sharfo A.W.M.
      • Stieler F.
      • Kupfer O.
      • Heijmen B.J.M.
      • Dirkx M.L.P.
      • Breedveld S.
      • et al.
      Automated VMAT planning for postoperative adjuvant treatment of advanced gastric cancer.
      ,
      • Buschmann M.
      • Sharfo A.W.M.
      • Penninkhof J.
      • Seppenwoolde Y.
      • Goldner G.
      • Georg D.
      • et al.
      Automated volumetric modulated arc therapy planning for whole pelvic prostate radiotherapyAutomatisierte volumenmodulierte Arc-Therapieplanung für Ganzbecken-Prostatabestrahlung.
      ,
      • Bijman R.
      • Rossi L.
      • Sharfo A.W.
      • Heemsbergen W.
      • Incrocci L.
      • Breedveld S.
      • et al.
      Automated Radiotherapy planning for patient-specific exploration of the trade-off between tumor dose coverage and predicted radiation-induced toxicity—A proof of principle study for prostate cancer..
      ,
      • Rossi L.
      • Cambraia Lopes P.
      • Marques Leitão J.
      • Janus C.
      • van de Pol M.
      • Breedveld S.
      • et al.
      On the importance of individualized, non-coplanar beam configurations in mediastinal lymphoma radiotherapy, optimized with automated planning.
      ,
      • Fjellanger K.
      • Hysing L.B.
      • Heijmen B.J.M.
      • Pettersen H.E.S.
      • Sandvik I.M.
      • Sulen T.H.
      • et al.
      Enhancing radiotherapy for locally advanced non-small cell lung cancer patients with ice, a novel system for automated multi-criterial treatment planning including beam angle optimization.
      ].
      The system has an option for integrated optimization of beam intensity profiles and (non-coplanar) beam angle directions [
      • Sharfo A.W.M.
      • Dirkx M.L.P.
      • Breedveld S.
      • Romero A.M.
      • Heijmen B.J.M.
      VMAT plus a few computer-optimized non-coplanar IMRT beams (VMAT+) tested for liver SBRT.
      ,
      • Sharfo A.W.M.
      • Rossi L.
      • Dirkx M.L.P.
      • Breedveld S.
      • Aluwini S.
      • Heijmen B.J.M.
      Complementing prostate SBRT VMAT with a two-beam non-coplanar IMRT class solution to enhance rectum and bladder sparing with minimum increase in treatment time.
      ,
      • Breedveld S.
      • Storchi P.R.M.
      • Voet P.W.J.
      • Heijmen B.J.M.
      ICycle: Integrated, multicriterial beam angle, and profile optimization for generation of coplanar and noncoplanar IMRT plans.
      ,
      • Rossi L.
      • Cambraia Lopes P.
      • Marques Leitão J.
      • Janus C.
      • van de Pol M.
      • Breedveld S.
      • et al.
      On the importance of individualized, non-coplanar beam configurations in mediastinal lymphoma radiotherapy, optimized with automated planning.
      ,
      • Fjellanger K.
      • Hysing L.B.
      • Heijmen B.J.M.
      • Pettersen H.E.S.
      • Sandvik I.M.
      • Sulen T.H.
      • et al.
      Enhancing radiotherapy for locally advanced non-small cell lung cancer patients with ice, a novel system for automated multi-criterial treatment planning including beam angle optimization.
      ,
      • Rossi L.
      • Breedveld S.
      • Heijmen B.J.M.
      • Voet P.W.J.
      • Lanconelli N.
      • Aluwini S.
      On the beam direction search space in computerized non-coplanar beam angle optimization for IMRT - Prostate SBRT.
      ,
      • Voet P.W.J.
      • Breedveld S.
      • Dirkx M.L.P.
      • Levendag P.C.
      • Heijmen B.J.M.
      Integrated multicriterial optimization of beam angles and intensity profiles for coplanar and noncoplanar head and neck IMRT and implications for VMAT.
      ,
      • Rossi L.
      • Sharfo A.W.
      • Aluwini S.
      • Dirkx M.
      • Breedveld S.
      • Heijmen B.
      First fully automated planning solution for robotic radiosurgery–comparison with automatically planned volumetric arc therapy for prostate cancer.
      ,
      • Schipaanboord B.W.K.
      • Breedveld S.
      • Rossi L.
      • Keijzer M.
      • Heijmen B.
      Automated prioritised 3D dose-based MLC segment generation for step-and-shoot IMRT.
      ,
      • Bijman R.
      • Rossi L.
      • Janssen T.
      • de Ruiter P.
      • van Triest B.
      • Breedveld S.
      • et al.
      MR-linac radiotherapy – the beam angle selection problem.
      ,
      • Schipaanboord B.W.K.
      • Giżyńska M.K.
      • Rossi L.
      • de Vries K.C.
      • Heijmen B.J.M.
      • Breedveld S.
      Fully automated treatment planning for MLC-based robotic radiotherapy.
      ,
      • Schipaanboord B.W.K.
      • Heijmen B.J.M.
      • Breedveld S.
      TBS-BAO: fully automated beam angle optimization for IMRT guided by a total-beam-space reference plan.
      ], which was used in this study. Erasmus-iCycle generates Pareto-optimal and clinically favorable plans by applying an appropriate treatment site specific configuration (’wish-list’) [
      • Breedveld S.
      • Storchi P.R.M.
      • Voet P.W.J.
      • Heijmen B.J.M.
      ICycle: Integrated, multicriterial beam angle, and profile optimization for generation of coplanar and noncoplanar IMRT plans.
      ,
      • Voet P.W.J.
      • Dirkx M.L.P.
      • Breedveld S.
      • Al-Mamgani A.
      • Incrocci L.
      • Heijmen B.J.M.
      Fully automated volumetric modulated arc therapy plan generation for prostate cancer patients.
      ,
      • Hussein M.
      • Heijmen B.J.M.
      • Verellen D.
      • Nisbet A.
      Automation in intensity modulated radiotherapy treatment planning-a review of recent innovations.
      ], containing hard constraints that can never be violated, and objectives that are optimized in order of priority. Wish-lists are generated through an iterative process where the initial aim is to mimic the clinical plans’ quality, respecting the same guidelines, constraints and trade-offs. In later iterations, the wish-list is improved maximally, with a drive to surpass clinical plan quality [
      • Heijmen B.
      • Voet P.
      • Fransen D.
      • Penninkhof J.
      • Milder M.
      • Akhiat H.
      • et al.
      Fully automated, multi-criterial planning for Volumetric Modulated Arc Therapy – An international multi-center validation for prostate cancer.
      ].
      When beam angle optimization (BAO) is combined with beam profile optimization, a candidate beam set has to be defined. Implicitly, the candidate beam set also defines the desired type of treatment technique, i.e., coplanar or non-coplanar.
      Erasmus-iCycle uses fluence map optimization (FMO) for automated plan generation. Subsequently, a commercial treatment planning system is used for automated segmentation of the generated FMO plan, mimicking the FMO dose distribution [
      • Sharfo A.W.M.
      • Dirkx M.L.P.
      • Breedveld S.
      • Romero A.M.
      • Heijmen B.J.M.
      VMAT plus a few computer-optimized non-coplanar IMRT beams (VMAT+) tested for liver SBRT.
      ,
      • Heijmen B.
      • Voet P.
      • Fransen D.
      • Penninkhof J.
      • Milder M.
      • Akhiat H.
      • et al.
      Fully automated, multi-criterial planning for Volumetric Modulated Arc Therapy – An international multi-center validation for prostate cancer.
      ,
      • Fjellanger K.
      • Hysing L.B.
      • Heijmen B.J.M.
      • Pettersen H.E.S.
      • Sandvik I.M.
      • Sulen T.H.
      • et al.
      Enhancing radiotherapy for locally advanced non-small cell lung cancer patients with ice, a novel system for automated multi-criterial treatment planning including beam angle optimization.
      ,
      • Voet P.W.J.
      • Breedveld S.
      • Dirkx M.L.P.
      • Levendag P.C.
      • Heijmen B.J.M.
      Integrated multicriterial optimization of beam angles and intensity profiles for coplanar and noncoplanar head and neck IMRT and implications for VMAT.
      ,
      • Rossi L.
      • Sharfo A.W.
      • Aluwini S.
      • Dirkx M.
      • Breedveld S.
      • Heijmen B.
      First fully automated planning solution for robotic radiosurgery–comparison with automatically planned volumetric arc therapy for prostate cancer.
      ]. For VMAT, an FMO plan with equi-angular IMRT beams is generated with Erasmus-iCycle, which is then automatically segmented in the commercial TPS for VMAT delivery [
      • Heijmen B.
      • Voet P.
      • Fransen D.
      • Penninkhof J.
      • Milder M.
      • Akhiat H.
      • et al.
      Fully automated, multi-criterial planning for Volumetric Modulated Arc Therapy – An international multi-center validation for prostate cancer.
      ,
      • Voet P.W.J.
      • Dirkx M.L.P.
      • Breedveld S.
      • Fransen D.
      • Levendag P.C.
      • Heijmen B.J.M.
      Toward fully automated multicriterial plan generation: A prospective clinical study.
      ,
      • Voet P.W.J.
      • Dirkx M.L.P.
      • Breedveld S.
      • Al-Mamgani A.
      • Incrocci L.
      • Heijmen B.J.M.
      Fully automated volumetric modulated arc therapy plan generation for prostate cancer patients.
      ,
      • Sharfo A.W.M.
      • Voet P.W.J.
      • Breedveld S.
      • Mens J.W.M.
      • Hoogeman M.S.
      • Heijmen B.J.M.
      Comparison of VMAT and IMRT strategies for cervical cancer patients using automated planning.
      ,
      • Sharfo A.W.M.
      • Breedveld S.
      • Voet P.W.J.
      • Heijkoop S.T.
      • Mens J.M.
      • Hoogeman M.S.
      Validation of fully automated VMAT plan generation for library-based plan-of-the-day cervical cancer radiotherapy.
      ,
      • Della Gala G.
      • Dirkx M.L.P.
      • Hoekstra N.
      • Fransen D.
      • Lanconelli N.
      • van de Pol M.
      • et al.
      Fully automated VMAT treatment planning for advanced-stage NSCLC patientsVollautomatische VMAT-Behandlungsplanung für Patienten mit fortgeschrittenem NSCLC.
      ,
      • Buergy D.
      • Sharfo A.W.M.
      • Heijmen B.J.M.
      • Voet P.W.J.
      • Breedveld S.
      • Wenz F.
      • et al.
      Fully automated treatment planning of spinal metastases - A comparison to manual planning of Volumetric Modulated Arc Therapy for conventionally fractionated irradiation.
      ,
      • Sharfo A.W.M.
      • Dirkx M.L.P.
      • Bijman R.G.
      • Schillemans W.
      • Breedveld S.
      • Aluwini S.
      • et al.
      Late toxicity in the randomized multicenter HYPRO trial for prostate cancer analyzed with automated treatment planning.
      ,
      • Sharfo A.W.M.
      • Stieler F.
      • Kupfer O.
      • Heijmen B.J.M.
      • Dirkx M.L.P.
      • Breedveld S.
      • et al.
      Automated VMAT planning for postoperative adjuvant treatment of advanced gastric cancer.
      ,
      • Bijman R.
      • Rossi L.
      • Sharfo A.W.
      • Heemsbergen W.
      • Incrocci L.
      • Breedveld S.
      • et al.
      Automated Radiotherapy planning for patient-specific exploration of the trade-off between tumor dose coverage and predicted radiation-induced toxicity—A proof of principle study for prostate cancer..
      ,
      • Rossi L.
      • Cambraia Lopes P.
      • Marques Leitão J.
      • Janus C.
      • van de Pol M.
      • Breedveld S.
      • et al.
      On the importance of individualized, non-coplanar beam configurations in mediastinal lymphoma radiotherapy, optimized with automated planning.
      ,
      • Fjellanger K.
      • Hysing L.B.
      • Heijmen B.J.M.
      • Pettersen H.E.S.
      • Sandvik I.M.
      • Sulen T.H.
      • et al.
      Enhancing radiotherapy for locally advanced non-small cell lung cancer patients with ice, a novel system for automated multi-criterial treatment planning including beam angle optimization.
      ,
      • Rossi L.
      • Sharfo A.W.
      • Aluwini S.
      • Dirkx M.
      • Breedveld S.
      • Heijmen B.
      First fully automated planning solution for robotic radiosurgery–comparison with automatically planned volumetric arc therapy for prostate cancer.
      ].

      2.4 Automatically generated treatment plans

      All plans in this study were generated for an Elekta Synergy treatment machine (Elekta AB, Stockholm, Sweden) equipped with a VersaHD multi-leaf collimator, using 10 MV photon beams. In Erasmus-iCycle, VMAT arcs were simulated with 23 equi-angular beams. Automatically generated Erasmus-iCycle plans were converted into clinically deliverable segmented plans using the Monaco treatment planning system (TPS), version 5.11 (Elekta AB, Stockholm, Sweden). These conversions were fully automatically performed, using patient-specific Monaco templates, automatically derived from the patient-specific Erasmus-iCycle FMO dose distributions [
      • Rossi L.
      • Breedveld S.
      • Heijmen B.J.M.
      • Voet P.W.J.
      • Lanconelli N.
      • Aluwini S.
      On the beam direction search space in computerized non-coplanar beam angle optimization for IMRT - Prostate SBRT.
      ,
      • Sharfo A.W.M.
      • Rossi L.
      • Dirkx M.L.P.
      • Breedveld S.
      • Aluwini S.
      • Heijmen B.J.M.
      Complementing prostate SBRT VMAT with a two-beam non-coplanar IMRT class solution to enhance rectum and bladder sparing with minimum increase in treatment time.
      ,
      • Breedveld S.
      • Storchi P.R.M.
      • Voet P.W.J.
      • Heijmen B.J.M.
      ICycle: Integrated, multicriterial beam angle, and profile optimization for generation of coplanar and noncoplanar IMRT plans.
      ,
      • Buergy D.
      • Sharfo A.W.M.
      • Heijmen B.J.M.
      • Voet P.W.J.
      • Breedveld S.
      • Wenz F.
      • et al.
      Fully automated treatment planning of spinal metastases - A comparison to manual planning of Volumetric Modulated Arc Therapy for conventionally fractionated irradiation.
      ,
      • Rossi L.
      • Cambraia Lopes P.
      • Marques Leitão J.
      • Janus C.
      • van de Pol M.
      • Breedveld S.
      • et al.
      On the importance of individualized, non-coplanar beam configurations in mediastinal lymphoma radiotherapy, optimized with automated planning.
      ,
      • Fjellanger K.
      • Hysing L.B.
      • Heijmen B.J.M.
      • Pettersen H.E.S.
      • Sandvik I.M.
      • Sulen T.H.
      • et al.
      Enhancing radiotherapy for locally advanced non-small cell lung cancer patients with ice, a novel system for automated multi-criterial treatment planning including beam angle optimization.
      ,
      • Voet P.W.J.
      • Breedveld S.
      • Dirkx M.L.P.
      • Levendag P.C.
      • Heijmen B.J.M.
      Integrated multicriterial optimization of beam angles and intensity profiles for coplanar and noncoplanar head and neck IMRT and implications for VMAT.
      ,
      • Rossi L.
      • Sharfo A.W.
      • Aluwini S.
      • Dirkx M.
      • Breedveld S.
      • Heijmen B.
      First fully automated planning solution for robotic radiosurgery–comparison with automatically planned volumetric arc therapy for prostate cancer.
      ,
      • Schipaanboord B.W.K.
      • Breedveld S.
      • Rossi L.
      • Keijzer M.
      • Heijmen B.
      Automated prioritised 3D dose-based MLC segment generation for step-and-shoot IMRT.
      ,
      • Bijman R.
      • Rossi L.
      • Janssen T.
      • de Ruiter P.
      • van Triest B.
      • Breedveld S.
      • et al.
      MR-linac radiotherapy – the beam angle selection problem.
      ,
      • Schipaanboord B.W.K.
      • Giżyńska M.K.
      • Rossi L.
      • de Vries K.C.
      • Heijmen B.J.M.
      • Breedveld S.
      Fully automated treatment planning for MLC-based robotic radiotherapy.
      ,
      • Schipaanboord B.W.K.
      • Heijmen B.J.M.
      • Breedveld S.
      TBS-BAO: fully automated beam angle optimization for IMRT guided by a total-beam-space reference plan.
      ,
      • Hussein M.
      • Heijmen B.J.M.
      • Verellen D.
      • Nisbet A.
      Automation in intensity modulated radiotherapy treatment planning-a review of recent innovations.
      ,
      • Bijman R.
      • Sharfo A.W.
      • Rossi L.
      • Breedveld S.
      • Heijmen B.
      Pre-clinical validation of a novel system for fully-automated treatment planning.
      ,
      • Ma T.M.
      • Emami B.
      • Grimm J.
      • Xue J.
      • Asbell S.O.
      • Kubicek G.J.
      • et al.
      Volume effects in radiosurgical spinal cord dose tolerance: how small is too small?.
      ].
      The wish-list was developed based on the clinical protocol, using in total five training patients in two steps: an initial wish-list was created by extensive tuning with 2 patients (not included in the study cohort, to avoid bias), which was then followed by minor fine-tuning using 3 other patients (included in the study cohort as plan quality changes with the fine-tuning were very small). The final wish-list was then used to generate automated treatment plans for all other study patients. For more information on wish-list development for this study, see Electronic Supplement A, section A2.
      For BAO in VMAT+5 plan generation, the full non-coplanar search space consisted of 437 candidate beams, separated by 10°, excluding beams that could result in collisions between the patient situated at the treatment couch and the gantry, as verified at the treatment unit.

      2.5 Plan evaluations and comparisons

      After confirming that adequate PTV70 dose coverage (V95 > 98%) was obtained, all plans were normalized to the same PTV70 dose coverage (V95% = 98%, the clinical goal), in order to facilitate fair comparisons of all generated plans.
      Plan evaluations and comparisons were mainly based on the clinically used parameters. For the parallel organs, the mean dose (Dmean) was evaluated; for serial organs, the D0.03cc was evaluated, as a more robust replacement of the Dmax [
      • Ma T.M.
      • Emami B.
      • Grimm J.
      • Xue J.
      • Asbell S.O.
      • Kubicek G.J.
      • et al.
      Volume effects in radiosurgical spinal cord dose tolerance: how small is too small?.
      ]. In addition to the clinically applied parameters, the dose bath was evaluated through the patient volumes for V10Gy, V30Gy and V50Gy.
      Statistical analyses were performed using the two-sided Wilcoxon signed-rank tests. Differences with p-value < 0.05 were considered statistically significant.

      3. Results

      3.1 Comparison of automatically and manually generated coplanar VMAT plans

      Development of the applied wish-list for automated plan generation for NPC is described in section A2 of Electronic Supplement A. The final wish-list can be found in Table A1 of Electronic Supplement A.
      All automatically generated plans rescaled to 98% coverage for PTV70 (section 2.5) fulfilled all clinical constraints. Differences between ClinVMAT and AutoVMAT for healthy tissues are summarized in Fig. 1 and Table B1 (Electronic Supplement B) and patient-specific comparisons in Figure B1.
      Figure thumbnail gr1
      Fig. 1Differences between ClinVMAT and AutoVMAT in healthy tissue plan parameters; positive values in case of an advantage for AutoVMAT. For plan parameters with an * the difference is statistically significant. All plans were normalized for equal PTV70 coverage. Green boxes show interquartile ranges (IQR), separated by blue lines as median. The whiskers extend to a maximum of 1.5 × IQR beyond the box, to each side
      [
      • Krzywinski M.
      • Altman N.
      Visualizing samples with box plots.
      ]
      . Values outside this range are plotted individually as outliers (+, in red). Subscripts L and R: Left and Right. Muscle S, Muscle M, Muscle I: superior, middle, and inferior swallowing muscles, respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
      Automatically generated VMAT plans had overall favourable plan parameters compared to ClinVMAT. For OARs related to xerostomia and dysphagia we observed statistically significant differences for right parotid Dmean (reduction by 2.2 ± 4.8 Gy), left parotid Dmean (reduction by 3.9 ± 5.9 Gy), left submandibular gland Dmean (reduction by 3.8 ± 6.2 Gy), oral cavity Dmean (reduction by 2.6 ± 4.1 Gy), superior, middle and inferior swallowing muscles Dmean (reductions by 2.1 ± 2.3 Gy, 4.5 ± 3.8 Gy and 8.2 ± 7.4 Gy). Moreover, dose reductions with AutoVMAT for the larynx, mandible, left optical nerve and chiasm were statistically significant. ClinVMAT only performed statistically significantly better for the spinal cord, but in the AutoVMAT plans the spinal cord D0.03cc was always below the constraint level. Figure B1 shows for each of the 20 study patients separately, differences in plan parameter values. Figs B1a and B1b show consistent advantages of AutoVMAT for parotids, SMGs, oral cavity and swallowing muscles for all study patients, while the level of sparing is highly patient-specific.
      The comparisons between AutoVMAT and ClinVMAT demonstrate high plan quality for the automated treatment planning and supports application of the wish-list in this study for development of non-coplanar VMAT CS for NPC.

      3.2 Establishment of non-coplanar class solutions

      All VMAT+5 plans resulted in sufficient PTV coverage and respected the clinically applied dose constraints. AutoVMAT and VMAT+5 plans are compared in Table 2 and Fig. 2. Compared to AutoVMAT, VMAT+5 improved doses in several clinically important structures at the cost of some losses that were considered clinically less relevant, e.g. because the related structures only had a D0.03cc constraint that had to be respected (which was always the case) with minor clinical interest in further dose reduction.
      Table 2Comparison of dosimetric plan parameters for AutoVMAT and differences with VMAT+5, VMATCS60 and VMATCS90. Mean values, standard deviations (SD) and ranges refer to the 20 patients in the study. Data related to statistically significant differences (p < 0.05) are in bold. NS: non-significant. Parallel OARs: parotid glands, SMGs, oral cavity, swallowing muscles, larynx, esophagus, cochleas and lenses. Serial OARs: spinal cord, brainstem, brain, optical nerves, chiasm, retina, and mandible.
      StructureParameterAutoVMATAutoVMAT - VMAT+5AutoVMAT - VMATCS60AutoVMAT - VMATCS90
      Mean ± SD[Min, Max]Mean ± SD[Min, Max]p-valueMean ± SD[Min, Max]p-valueMean ± SD[Min, Max]p-value
      PTV70V95% [%]98.1 ± 0[98, 98.1]0 ± 0[−0.1, 0.1]NS0 ± 0[−0.1, 0.1]NS0 ± 0[0, 0.1]NS
      V107% [cc]0.1 ± 0.4[0, 1.7]0 ± 0.2[−0.9, 0.3]NS0 ± 0.3[−1.1, 0.3]NS−0.1 ± 0.2[−0.7, 0.3]0.05
      PTV54.25V95% [%]95.4 ± 14[36.3, 99.9]−3.2 ± 14.1[−63.1, 0.8]NS−2.8 ± 14[−62.3, 1.6]NS−3 ± 14[−62.3, 0.8]NS
      ParotidRDmean [Gy]38 ± 15.2[7.9, 68.8]2.5 ± 2.6[−5.1, 5.7]0.0024 ± 2.1[−0.1, 7.4]<0.0013.1 ± 2.4[−2.3, 7.8]<0.001
      ParotidLDmean [Gy]36.4 ± 12.7[20.1, 61.9]2.4 ± 2.8[−6.3, 6.6]0.0023.7 ± 2.4[−0.3, 9.5]<0.0012.7 ± 2.4[−3.3, 7]<0.001
      SMGRDmean [Gy]56.8 ± 13.7[34.2, 71.6]0.4 ± 2.2[−5.8, 5.9]NS1.2 ± 2.4[−1.8, 7.2]NS0.9 ± 1.8[−2.2, 5.2]0.044
      SMGLDmean [Gy]50.2 ± 14.3[31.6, 71.8]−0.2 ± 3.2[−11.6, 2.7]NS0.7 ± 1.5[−2.8, 3.9]0.0360.1 ± 2.5[−8.3, 2.9]NS
      Oral CavityDmean [Gy]37.8 ± 6.8[25.7, 53.8]2.4 ± 3.2[−2, 10.2]0.0023 ± 2.9[−3.4, 8.5]0.0012.8 ± 2.6[−2, 8.4]<0.001
      MuscleSDmean [Gy]61.3 ± 6[50.2, 71.3]0.4 ± 2.1[−4.4, 2.9]NS1.4 ± 1.4[−1, 4.5]0.0020.6 ± 2.3[−4.8, 4.6]NS
      MuscleMDmean [Gy]48.4 ± 9.9[23.3, 71.1]1.3 ± 1.3[−1.4, 3.8]0.0012.1 ± 2.1[−1.4, 5.7]0.0011.9 ± 2[−1.7, 5.7]0.002
      MuscleIDmean [Gy]31.1 ± 12[1.6, 56.1]0.8 ± 1.7[−1.4, 4.8]NS1.4 ± 2.1[−2.7, 5.3]0.011.1 ± 2[−2.6, 4.4]0.033
      LarynxDmean [Gy]35.8 ± 8[11.4, 52.6]1.4 ± 2.6[−2.8, 6.7]0.031.9 ± 2.7[−2.1, 7.4]0.0062.4 ± 3.3[−2.7, 8.7]0.006
      EsophagusDmean [Gy]25 ± 9.8[0.3, 39.9]1.8 ± 5.4[−10.6, 12.8]NS4.4 ± 5.5[−1.1, 15.2]0.0034.1 ± 5.6[−2.2, 15.2]0.009
      CochleaRDmean [Gy]31.2 ± 9.8[10.4, 46.1]0 ± 5[−10.7, 11.9]NS0.2 ± 6.5[−14.7, 12]NS−1.1 ± 6.1[−11.3, 12]NS
      CochleaLDmean [Gy]34.5 ± 11.1[7.7, 69.3]0.2 ± 2.4[−2.4, 7.1]NS−1.1 ± 4.1[−14, 6.3]NS−0.9 ± 3.6[−9.6, 5.7]NS
      LensLD0.03cc [Gy]3.7 ± 2.1[0.9, 8.9]−0.9 ± 1.1[−3.5, 0.5]0.009−4.5 ± 2.8[−8.8, 0.2]<0.001−3.6 ± 1.9[−7.4, −1.2]<0.001
      LensRD0.03cc [Gy]3.7 ± 1.9[0.8, 7.6]−1.5 ± 2.5[−8.2, 0.7]NS−3.8 ± 1.6[−6.3, −1.3]<0.001−3.2 ± 1.6[−5.5, −0.1]<0.001
      Spinal CordD0.03cc [Gy]40.4 ± 4.5[29.4, 46.7]2.5 ± 3[−3.2, 7.8]0.0023.9 ± 3.9[−1.6, 12.4]<0.0011.6 ± 3.4[−3, 9.5]NS
      BrainstemD0.03cc [Gy]47.1 ± 7.9[33.9, 59.6]1.7 ± 4.2[−4.6, 11.7]NS3.7 ± 3.8[−1.8, 12.2]<0.0011.6 ± 3.3[−3.3, 12.4]NS
      BrainD0.03cc [Gy]67.3 ± 4.7[60.3, 71.3]−0.2 ± 1.2[−1.8, 1.3]NS−0.2 ± 0.9[−1.3, 1.1]NS0.3 ± 1.5[−0.8, 2.9]NS
      MandibleD0.03cc [Gy]68.3 ± 3.9[56.3, 71.3]0.5 ± 0.8[−0.9, 2.2]0.0390.3 ± 1.9[−1.9, 6.1]NS0 ± 1.6[−1.8, 4.5]NS
      Optical NerveLD0.03cc [Gy]16.5 ± 11.7[2.5, 35]−6.5 ± 4.8[−16.4, −1.3]<0.001−15.2 ± 6.3[−29.2, −6.7]<0.001−11.3 ± 6[−18.7, −3.5]<0.001
      Optical NerveRD0.03cc [Gy]15.1 ± 11[1.8, 34.1]−5 ± 5[−16.8, 0.7]0.005−15.6 ± 5.6[−27.6, −8.1]<0.001−8.4 ± 5.1[−16.8, −0.2]<0.001
      ChiasmD0.03cc [Gy]8.4 ± 5.7[2.5, 20.5]−8.6 ± 5.9[−19.1, −1.1]<0.001−20.7 ± 8.3[−30.1, −4.2]<0.001−12.7 ± 6.8[−24, −2.1]<0.001
      RetinaLD0.03cc [Gy]11.4 ± 4.5[1.5, 17.1]−1.5 ± 3.2[−6.8, 3.2]NS−6.9 ± 6.7[−15.9, 3.6]0.02−6.2 ± 4.3[−15.9, −1.1]0.002
      RetinaRD0.03cc [Gy]12.8 ± 5.3[1.3, 18.9]−1.4 ± 9.2[–22.6, 8.7]NS−2.9 ± 3.6[−8.9, 1.5]0.049−5.3 ± 4.3[−10.2, 2.8]0.01
      PatientV10Gy [cc]5326 ± 1448[1655, 7866]−692 ± 640[−2098, 681]<0.001−625 ± 413[−1542, 77]<0.001−479 ± 412[−1231, 101]<0.001
      V30Gy [cc]2508 ± 741[591, 4104]142 ± 93[−67, 349]<0.001206 ± 130[−24, 475]<0.001223 ± 135[–32, 486]<0.001
      V50Gy [cc]1235 ± 396[274, 1990]18 ± 48[−137, 98]0.0222 ± 38[−46, 99]0.02523 ± 39[−48, 125]0.015
      Figure thumbnail gr2
      Fig. 2Differences between AutoVMAT and VMAT+5 (yellow), VMATCS60 (purple) and VMATCS90 (blue) in healthy tissue plan parameters; positive in case of advantage for VMAT+5, VMATCS60 and VMATCS90. For plan parameters with an * the difference is statistically significant. All plans were normalized for equal PTV70 coverage. Yellow, purple, and blue boxes show IQR, separated by blue lines as median. The whiskers extend to a maximum of 1.5 × IQR beyond the box, to each side
      [
      • Krzywinski M.
      • Altman N.
      Visualizing samples with box plots.
      ]
      . Values outside this range are plotted individually as outliers (+, in red). Subscripts L and R: Left and Right. MuscleS, MuscleM, MuscleI: superior, middle, and inferior swallowing muscles, respectively. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
      The angular distribution of the selected IMRT beams in the VMAT+5 plans of the 20 study patients is presented in Fig. 3. Based on this distribution, the class solution VMATCS60 was defined with partial arcs at couch angles −60° and 60° (purple boxes in Fig. 3). Partial arcs were used to avoid collisions. To investigate plan quality with the smallest possible number of non-coplanar arcs (one) for keeping treatment times as low as possible, the VMATCS90 class solution was defined with one partial arc at couch angle 90° (blue box in Fig. 3).
      Figure thumbnail gr3
      Fig. 3Angular distribution and frequency of the 100 non-coplanar IMRT beams in the VMAT+5 plans of the 20 study patients. SearchSpace (grey markers): feasible non-coplanar beam set-ups not resulting in a collision. Frequency marks (black circles): number of times an available beam angle was selected. The purple boxes represent the partial-arcs for VMATCS60. The blue box represents the partial-arc for VMATCS90. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

      3.3 Comparison of VMATCS60, VMATCS90 and VMAT+5 with AutoVMAT

      All plans included in the analyses were rescaled to 98% PTV70 coverage, after confirming they had adequate PTV coverage and fulfilled the clinical constraints. Data is presented in Table 2 and Fig. 2. In the non-coplanar VMATCS60, VMATCS90 and VMAT+5, doses in several high priority structures reduced compared to coplanar AutoVMAT, at the cost of some clinically less relevant losses, mainly related to structures with a D0.03cc constraint that were anyway respected and mostly with low parameter values. As shown in Table 3, compared to VMATCS90, VMATCS60 had lower parotids Dmean (-0.8 ± 1.4 Gy and −1 ± 2.1 Gy) and spinal cord and brainstem D0.03cc, while D0.03cc for the right optical nerve and the chiasm were on average higher (all D0.03cc values with constraints). Larger differences between VMATCS60 and VMATCS90 were patient-specific.
      Table 3Comparison of dosimetric plan parameters for VMATCS60 and VMATCS90. Mean values, standard deviations (SD) and ranges refer to the 20 patients in the study. Statistically significant differences (p-value < 0.05) are in bold. NS: non-significant. NS: non-significant. Parallel OARs: parotid glands, SMGs, oral cavity, swallowing muscles, larynx, esophagus, cochleas and lenses. Serial OARs: spinal cord, brainstem, brain, optical nerves, chiasm, retina and mandible.
      StructureParameterVMATCS60VMATCS90VMATCS60 - VMATCS90
      Mean ± SD[Min, Max]Mean ± SD[Min, Max]Mean ± SD[Min, Max]p-value
      PTV70V95% [%]98.1 ± 0[98, 98.1]98.1 ± 0[98, 98.1]0 ± 0[−0.1, 0.1]NS
      V107% [cc]0.2 ± 0.5[0, 1.7]0.2 ± 0.5[0, 1.8]0 ± 0.4[−1, 1.4]NS
      PTV54.25V95% [%]98.3 ± 1.1[95.5, 99.7]98.5 ± 0.9[96.8, 99.8]−0.2 ± 0.4[−1.4, 0.2]0.021
      ParotidRDmean [Gy]34 ± 15[7, 69]34.8 ± 15.6[6.6, 68.9]−0.8 ± 1.4[−4.2, 1.3]0.009
      ParotidLDmean [Gy]32.7 ± 12.2[20.1, 59.3]33.7 ± 12.8[19.2, 64]−1 ± 2.1[−8, 1.2]0.028
      SMGRDmean [Gy]55.6 ± 14[33.9, 72]55.9 ± 14.5[32.4, 72.1]−0.3 ± 1.8[−4.7, 2.3]NS
      SMGLDmean [Gy]49.5 ± 14.3[32.4, 71.4]50.1 ± 14.4[32.3, 70.1]−0.6 ± 2.4[−9.5, 1.8]NS
      Oral CavityDmean [Gy]34.8 ± 6.8[21.6, 51.7]35 ± 6.9[22.1, 51.7]−0.2 ± 2.1[−5.4, 4.9]NS
      MuscleSDmean [Gy]59.9 ± 6.9[48.3, 70.6]60.7 ± 7.1[48.3, 71.8]−0.8 ± 2.2[−6.7, 3.5]NS
      MuscleMDmean [Gy]46.3 ± 10.4[23.8, 70.6]46.5 ± 10.2[25, 71.6]−0.2 ± 1.2[−2.7, 2.9]NS
      MuscleIDmean [Gy]29.6 ± 10.8[4.3, 54]30 ± 11.2[4.2, 55.4]−0.3 ± 0.9[−2, 0.9]NS
      LarynxDmean [Gy]33.8 ± 7.2[12.7, 45.2]33.3 ± 7.6[14.1, 45.7]0.5 ± 2[−4, 6]NS
      EsophagusDmean [Gy]20.5 ± 8.7[0.8, 35]20.9 ± 8.6[1.2, 35.4]−0.3 ± 2.2[−7.9, 2.7]NS
      CochleaRDmean [Gy]31 ± 9.4[9.6, 45.9]32.3 ± 8.3[13.7, 45.6]−1.4 ± 4.3[−13.8, 3.4]NS
      CochleaLDmean [Gy]35.6 ± 10.2[21.6, 71]35.4 ± 10.3[17.3, 71]0.1 ± 2.6[−7.2, 4.4]NS
      LensLD0.03cc [Gy]8.2 ± 3[2.4, 11.9]7.3 ± 2.7[3.1, 11]0.9 ± 2.6[−3.5, 7]NS
      LensRD0.03cc [Gy]7.5 ± 2.5[4.2, 11.3]7 ± 2.6[0.9, 10.8]0.6 ± 1.5[−1.5, 3.3]NS
      Spinal CordD0.03cc [Gy]36.5 ± 6.1[24.5, 44.7]38.8 ± 5.7[29.1, 46.6]−2.3 ± 2.8[−8.4, 0.4]0.001
      BrainstemD0.03cc [Gy]43.5 ± 8.7[31.9, 59.2]45.5 ± 8.7[32.3, 59.2]−2 ± 2.3[−7.3, 0.5]<0.001
      BrainD0.03cc [Gy]67.6 ± 4.5[60.3, 71.3]67 ± 5.9[57.4, 71.3]0.6 ± 1.5[−0.9, 2.9]NS
      MandibleD0.03cc [Gy]68 ± 5.4[53.1, 72.4]68.3 ± 5.2[53.7, 71.5]−0.3 ± 1.1[−2, 2.3]NS
      Optical NerveLD0.03cc [Gy]31.7 ± 11.5[14.1, 46.2]27.8 ± 11.4[7.5, 45.6]3.9 ± 6.7[−9.9, 12.1]NS
      Optical NerveRD0.03cc [Gy]30.8 ± 10.7[12.6, 48.3]23.6 ± 12.9[2, 48.3]7.2 ± 6[0, 20.5]0.001
      ChiasmD0.03cc [Gy]29.1 ± 10.2[10, 42.4]21.2 ± 8.9[8.2, 35.9]8 ± 7.7[0, 23.2]0.002
      RetinaLD0.03cc [Gy]18.2 ± 7.4[7.1, 33]17.6 ± 6.6[7.1, 33]0.7 ± 5.5[−7.2, 10.6]NS
      RetinaRD0.03cc [Gy]15.6 ± 5.5[6.4, 27.8]18.1 ± 8[1.3, 27.8]−2.4 ± 5.5[−10.4, 5.1]NS
      PatientV10Gy [cc]5951 ± 1558[1930, 8741]5805 ± 1511[1687, 8225]146 ± 304[−515, 976]0.025
      V30Gy [cc]2301 ± 686[476, 3636]2284 ± 683[506, 3753]17 ± 69[−118, 170]NS
      V50Gy [cc]1213 ± 394[258, 1929]1212 ± 386[254, 1916]1 ± 24[−49, 45]NS
      Figure B2 (in Electronic Supplement B) shows for each of the 20 study patients separately, differences in plan parameter values between AutoVMAT and VMATCS60. Figure B2a and B2b show consistent dose reductions for the higher priority OARs for VMATCS60. For the lower priority OARs (non-constrained), the level of sparing is patient and OAR-specific. For esophagus, patients 9, 10 and 12 have dose reductions of at least 10 Gy, while the average is 4.4 ± 5 Gy. However, for patient 2 and 9, cochlea doses were higher with VMATCS60. Nevertheless, in none of the plans OAR doses surpassed the clinically applied goal. AutoVMAT resulted in lower low dose bath and VMATCS60 in lower high dose bath, for all patients (Figure B2f).

      4. Discussion

      High quality non-coplanar VMAT class solutions can in principle circumvent the current lack of algorithms for individualized non-coplanar arc selection in most commercial treatment planning systems, or prevent complex, time- and workload intensive selection of patient-specific non-coplanar arcs with manual planning. Alternatively, patients can be treated with regular coplanar VMAT, which may result in suboptimal dose for some tumor sites or for individual patients. The aim of this study was to use automated planning to develop and evaluate non-coplanar VMAT class solution for nasopharyngeal cancer, with only few non-coplanar static arcs to keep treatment times clinically feasible. One CS, consisting of a full coplanar arc supplemented with 2 fixed non-coplanar partial arcs at couch angles −60° and 60° (VMATCS60) was derived from a population distribution of patient-specific IMRT beam angles obtained with our in-house developed algorithm for automated multi-criterial plan generation with integrated beam angle optimization. For comparison, the simplest non-coplanar CS with only one non-coplanar arc at couch 90° was investigated as well.
      To the best of our knowledge, the use of automated planning for development and evaluation of VMAT CSs has never been investigated for NPC. Many studies have demonstrated superior plan quality for automated planning compared to manual planning, for many tumor sites [
      • Heijmen B.
      • Voet P.
      • Fransen D.
      • Penninkhof J.
      • Milder M.
      • Akhiat H.
      • et al.
      Fully automated, multi-criterial planning for Volumetric Modulated Arc Therapy – An international multi-center validation for prostate cancer.
      ,
      • Voet P.W.J.
      • Dirkx M.L.P.
      • Breedveld S.
      • Fransen D.
      • Levendag P.C.
      • Heijmen B.J.M.
      Toward fully automated multicriterial plan generation: A prospective clinical study.
      ,
      • Voet P.W.J.
      • Dirkx M.L.P.
      • Breedveld S.
      • Al-Mamgani A.
      • Incrocci L.
      • Heijmen B.J.M.
      Fully automated volumetric modulated arc therapy plan generation for prostate cancer patients.
      ,
      • Sharfo A.W.M.
      • Voet P.W.J.
      • Breedveld S.
      • Mens J.W.M.
      • Hoogeman M.S.
      • Heijmen B.J.M.
      Comparison of VMAT and IMRT strategies for cervical cancer patients using automated planning.
      ,
      • Sharfo A.W.M.
      • Breedveld S.
      • Voet P.W.J.
      • Heijkoop S.T.
      • Mens J.M.
      • Hoogeman M.S.
      Validation of fully automated VMAT plan generation for library-based plan-of-the-day cervical cancer radiotherapy.
      ,
      • Della Gala G.
      • Dirkx M.L.P.
      • Hoekstra N.
      • Fransen D.
      • Lanconelli N.
      • van de Pol M.
      • et al.
      Fully automated VMAT treatment planning for advanced-stage NSCLC patientsVollautomatische VMAT-Behandlungsplanung für Patienten mit fortgeschrittenem NSCLC.
      ,
      • Buergy D.
      • Sharfo A.W.M.
      • Heijmen B.J.M.
      • Voet P.W.J.
      • Breedveld S.
      • Wenz F.
      • et al.
      Fully automated treatment planning of spinal metastases - A comparison to manual planning of Volumetric Modulated Arc Therapy for conventionally fractionated irradiation.
      ,
      • Sharfo A.W.M.
      • Dirkx M.L.P.
      • Bijman R.G.
      • Schillemans W.
      • Breedveld S.
      • Aluwini S.
      • et al.
      Late toxicity in the randomized multicenter HYPRO trial for prostate cancer analyzed with automated treatment planning.
      ,
      • Sharfo A.W.M.
      • Stieler F.
      • Kupfer O.
      • Heijmen B.J.M.
      • Dirkx M.L.P.
      • Breedveld S.
      • et al.
      Automated VMAT planning for postoperative adjuvant treatment of advanced gastric cancer.
      ,
      • Buschmann M.
      • Sharfo A.W.M.
      • Penninkhof J.
      • Seppenwoolde Y.
      • Goldner G.
      • Georg D.
      • et al.
      Automated volumetric modulated arc therapy planning for whole pelvic prostate radiotherapyAutomatisierte volumenmodulierte Arc-Therapieplanung für Ganzbecken-Prostatabestrahlung.
      ,
      • Bijman R.
      • Rossi L.
      • Sharfo A.W.
      • Heemsbergen W.
      • Incrocci L.
      • Breedveld S.
      • et al.
      Automated Radiotherapy planning for patient-specific exploration of the trade-off between tumor dose coverage and predicted radiation-induced toxicity—A proof of principle study for prostate cancer..
      ,
      • Rossi L.
      • Cambraia Lopes P.
      • Marques Leitão J.
      • Janus C.
      • van de Pol M.
      • Breedveld S.
      • et al.
      On the importance of individualized, non-coplanar beam configurations in mediastinal lymphoma radiotherapy, optimized with automated planning.
      ,
      • Fjellanger K.
      • Hysing L.B.
      • Heijmen B.J.M.
      • Pettersen H.E.S.
      • Sandvik I.M.
      • Sulen T.H.
      • et al.
      Enhancing radiotherapy for locally advanced non-small cell lung cancer patients with ice, a novel system for automated multi-criterial treatment planning including beam angle optimization.
      ]. Especially for development of CSs, high plan quality is required, as the aim is to in principle treat all future patients with the developed CS. A suboptimal CS would result in an on average suboptimal dose in these patients. The CSs investigated in this study and all plans generated for evaluation were created for regular linac delivery.
      Prior to the investigations on CSs, our optimizer for automated plan generation was configured for NPC treatment in line with our clinical planning aims by creating a ‘wish-list’. Automatically generated coplanar AutoVMAT plans were then compared to the clinically delivered coplanar VMAT plans (ClinVMAT). This first analysis pointed at an important opportunity for improving plan quality by replacing manual planning with automated planning, especially for structures related to xerostomia and dysphagia. Sparing of these structures was feasible while maintaining adequate target coverage and strictly obeying hard constraints on other OARs. This observed gain with autoplanning without any manual fine-tuning is remarkable, given the large anatomy variations in this patient group (Figure A1).
      For establishment VMATCS60, our optimizer was used to first automatically generate coplanar VMAT plans with 5 additional optimized patient-specific non-coplanar IMRT beams. The idea behind this approach was to identify the most important non-coplanar directions for NPC and to explore how they could be combined in few arcs. Although there were substantial variations in the patient group, there were two clusters of directions that pointed at an opportunity for VMATCS60 to improve plan quality. In a recent study, the applied consecutive beam angle selection in Erasmus-iCycle was compared to a new approach that uses so-called total-beam-space plans, confirming the high quality of automated beam angle selection in Erasmus-iCycle [
      • Schipaanboord B.W.K.
      • Heijmen B.J.M.
      • Breedveld S.
      TBS-BAO: fully automated beam angle optimization for IMRT guided by a total-beam-space reference plan.
      ].
      Overall, all investigated non-coplanar approaches i.e., VMAT+5, VMATCS60 and VMATCS90, outperformed coplanar AutoVMAT with respect to sparing of structures related to xerostomia and dysphagia (with the highest clinical priority), at the price of some losses that were considered clinically less relevant, e.g., because the related structures only had a D0.03cc constraint that had to be respected (which was always achieved) with minor clinical interest in further dose reduction. Table 2 suggests (no p-values provided for direct pairwise comparisons) that dose reductions with VMATCS60 and VMATCS90 in xerostomia and dysphagia related OARs relative to AutoVMAT were larger than reductions with VMAT+5 relative to AutoVMAT, while for some clinically considered less important OARs, VMAT+5 seemed favorable. With this observation we conclude that there are no indications that VMAT+5 is dosimetrically superior to VMATCS60 or VMATCS90. As VMAT+5 also has 5 couch rotations per fraction instead of 2 (VMATCS60) or 1 (VMATCS90), the latter two approaches seem overall favourable, especially as they also avoid time-consuming patient-specific BAO. As shown in Table 3, VMATCS60 seems to have on average slightly better dosimetry than VMATCS90, but this advantage must be weighed against the use of 2 non-coplanar arcs instead of one. Figure B2 shows that the gain of non-coplanar treatment is highly patient-specific. This observation points at an option to generate for each patient in the treatment planning phase three plans: coplanar VMAT, non-coplanar VMATCS60 and VMATCS90 and then only select a non-coplanar approach in case of clinically relevant dosimetric advantages. With automated planning, generation of multiple plans per patient can be performed without excessive increase in planning time and workload [
      • Sharfo A.W.M.
      • Breedveld S.
      • Voet P.W.J.
      • Heijkoop S.T.
      • Mens J.M.
      • Hoogeman M.S.
      Validation of fully automated VMAT plan generation for library-based plan-of-the-day cervical cancer radiotherapy.
      ].
      The observed superiority of non-coplanar set-ups for NPC treatment confirms the findings by Wild et al. [
      • Wild E.
      • Bangert M.
      • Nill S.
      • Oelfke U.
      Noncoplanar VMAT for nasopharyngeal tumors: Plan quality versus treatment time.
      ]. These authors applied different non-coplanar techniques to three nasopharyngeal tumour cases, including comparisons between coplanar VMAT, couch optimized non-coplanar VMAT and VMAT with dynamic couch rotation. While their results indicate that couch tilts do not yield as high plan quality as simultaneous couch and gantry rotations, the technology for rotating beam and couch trajectories is not currently commercially available.
      Two investigators (JL, RB) independently filled out the RATING score list [
      • Hansen C.R.
      • Crijns W.
      • Hussein M.
      • Rossi L.
      • Gallego P.
      • Verbakel W.
      • et al.
      Radiotherapy Treatment plannINg study Guidelines (RATING): A framework for setting up and reporting on scientific treatment planning studies.
      ] for quality assessment of treatment planning studies, arriving at scores of 87% and 86% (maximum 100%).

      5. Conclusions

      In this study we used an in-house algorithm for automated treatment planning with integrated beam angle optimization to develop and evaluate non-coplanar VMAT class solutions for nasopharyngeal carcinoma, consisting of coplanar VMAT supplemented with only few fixed non-coplanar VMAT arcs to keep treatment times clinically feasible. Proposed non-coplanar class solutions with one or two non-coplanar arcs better spared xerostomia and dysphagia related OARs than coplanar VMAT. Also, the use of automated planning instead of manual planning resulted in dosimetric gains. Advantages of using a non-coplanar class solution and of using automated planning were both highly patient-specific. [
      • Alber M.
      • Reemtsen R.
      Intensity modulated radiotherapy treatment planning by use of a barrier-penalty multiplier method.
      ].

      Declaration of Competing Interest

      The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

      Appendix A. Supplementary data

      The following are the Supplementary data to this article:

      References

        • Chen Y.P.
        • Chan A.T.C.
        • Le Q.T.
        • Blanchard P.
        • Sun Y.
        • Ma J.
        Nasopharyngeal carcinoma.
        Lancet. 2019; 394: 64-80https://doi.org/10.1016/S0140-6736(19)30956-0
        • Rana S.
        Intensity modulated radiation therapy versus volumetric intensity modulated arc therapy.
        J Med Radiat Sci. 2013; 60: 81-83https://doi.org/10.1002/jmrs.19
        • Teoh M.
        • Clark C.H.
        • Wood K.
        • Whitaker S.
        • Nisbet A.
        Volumetric modulated arc therapy: A review of current literature and clinical use in practice.
        Br J Radiol. 2011; 84: 967-996https://doi.org/10.1259/bjr/22373346
        • Wild E.
        • Bangert M.
        • Nill S.
        • Oelfke U.
        Noncoplanar VMAT for nasopharyngeal tumors: Plan quality versus treatment time.
        Med Phys. 2015; 42: 2157-2168https://doi.org/10.1118/1.4914863
        • Sharbo G.
        • Hashemi B.
        • Bakhshandeh M.
        • Rakhsha A.
        Radiobiological assessment of nasopharyngeal cancer IMRT using various collimator angles and non-coplanar fields.
        J Radiother Pract. 2021; 20: 168-175
        • Bangert M.
        • Ziegenhein P.
        • Oelfke U.
        Characterizing the combinatorial beam angle selection problem.
        Phys Med Biol. 2012; 57: 6707-6723https://doi.org/10.1088/0031-9155/57/20/6707
        • Bangert M.
        • Ziegenhein P.
        • Oelfke U.
        Comparison of beam angle selection strategies for intracranial IMRT.
        Med Phys. 2013; 40: 011716
        • Rossi L.
        • Breedveld S.
        • Heijmen B.J.M.
        • Voet P.W.J.
        • Lanconelli N.
        • Aluwini S.
        On the beam direction search space in computerized non-coplanar beam angle optimization for IMRT - Prostate SBRT.
        Phys Med Biol. 2012; 57: 5441-5458https://doi.org/10.1088/0031-9155/57/17/5441
        • Rocha H.
        • Dias J.M.
        • Ferreira B.C.
        • Lopes M.C.
        Beam angle optimization for intensity-modulated radiation therapy using a guided pattern search method.
        Phys Med Biol. 2013; 58: 2939-2953https://doi.org/10.1088/0031-9155/58/9/2939
        • Akbas U.
        • Koksal C.
        • Kesen N.D.
        • Ozkaya K.
        • Bilge H.
        • Altun M.
        Nasopharyngeal carcinoma radiotherapy with hybrid technique.
        Med Dosim. 2019; 44: 251-257https://doi.org/10.1016/j.meddos.2018.09.003
        • Sharfo A.W.M.
        • Dirkx M.L.P.
        • Breedveld S.
        • Romero A.M.
        • Heijmen B.J.M.
        VMAT plus a few computer-optimized non-coplanar IMRT beams (VMAT+) tested for liver SBRT.
        Radiother Oncol. 2017; 123: 49-56https://doi.org/10.1016/j.radonc.2017.02.018
        • Sharfo A.W.M.
        • Rossi L.
        • Dirkx M.L.P.
        • Breedveld S.
        • Aluwini S.
        • Heijmen B.J.M.
        Complementing prostate SBRT VMAT with a two-beam non-coplanar IMRT class solution to enhance rectum and bladder sparing with minimum increase in treatment time.
        Front Oncol. 2021; 11https://doi.org/10.3389/fonc.2021.620978
        • Hansen C.R.
        • Crijns W.
        • Hussein M.
        • Rossi L.
        • Gallego P.
        • Verbakel W.
        • et al.
        Radiotherapy Treatment plannINg study Guidelines (RATING): A framework for setting up and reporting on scientific treatment planning studies.
        Radiother Oncol. 2020; 153: 67-78
        • Heijmen B.
        • Voet P.
        • Fransen D.
        • Penninkhof J.
        • Milder M.
        • Akhiat H.
        • et al.
        Fully automated, multi-criterial planning for Volumetric Modulated Arc Therapy – An international multi-center validation for prostate cancer.
        Radiother Oncol. 2018; 128: 343-348
        • Breedveld S.
        • Storchi P.R.M.
        • Voet P.W.J.
        • Heijmen B.J.M.
        ICycle: Integrated, multicriterial beam angle, and profile optimization for generation of coplanar and noncoplanar IMRT plans.
        Med Phys. 2012; 39: 951-963https://doi.org/10.1118/1.3676689
        • Voet P.W.J.
        • Dirkx M.L.P.
        • Breedveld S.
        • Fransen D.
        • Levendag P.C.
        • Heijmen B.J.M.
        Toward fully automated multicriterial plan generation: A prospective clinical study.
        Int J Radiat Oncol Biol Phys. 2013; 85: 866-872https://doi.org/10.1016/j.ijrobp.2012.04.015
        • Voet P.W.J.
        • Dirkx M.L.P.
        • Breedveld S.
        • Al-Mamgani A.
        • Incrocci L.
        • Heijmen B.J.M.
        Fully automated volumetric modulated arc therapy plan generation for prostate cancer patients.
        Int J Radiat Oncol Biol Phys. 2014; 88: 1175-1179https://doi.org/10.1016/j.ijrobp.2013.12.046
        • Sharfo A.W.M.
        • Voet P.W.J.
        • Breedveld S.
        • Mens J.W.M.
        • Hoogeman M.S.
        • Heijmen B.J.M.
        Comparison of VMAT and IMRT strategies for cervical cancer patients using automated planning.
        Radiother Oncol. 2015; 114: 395-401https://doi.org/10.1016/j.radonc.2015.02.006
        • Sharfo A.W.M.
        • Breedveld S.
        • Voet P.W.J.
        • Heijkoop S.T.
        • Mens J.M.
        • Hoogeman M.S.
        Validation of fully automated VMAT plan generation for library-based plan-of-the-day cervical cancer radiotherapy.
        PLoS ONE. 2016; 11 (169202)https://doi.org/10.1371/journal.pone.0169202
        • Della Gala G.
        • Dirkx M.L.P.
        • Hoekstra N.
        • Fransen D.
        • Lanconelli N.
        • van de Pol M.
        • et al.
        Fully automated VMAT treatment planning for advanced-stage NSCLC patientsVollautomatische VMAT-Behandlungsplanung für Patienten mit fortgeschrittenem NSCLC.
        Strahlenther Onkol. 2017; 193: 402-409
        • Buergy D.
        • Sharfo A.W.M.
        • Heijmen B.J.M.
        • Voet P.W.J.
        • Breedveld S.
        • Wenz F.
        • et al.
        Fully automated treatment planning of spinal metastases - A comparison to manual planning of Volumetric Modulated Arc Therapy for conventionally fractionated irradiation.
        Radiat Oncol. 2017; 12https://doi.org/10.1186/s13014-017-0767-2
        • Sharfo A.W.M.
        • Dirkx M.L.P.
        • Bijman R.G.
        • Schillemans W.
        • Breedveld S.
        • Aluwini S.
        • et al.
        Late toxicity in the randomized multicenter HYPRO trial for prostate cancer analyzed with automated treatment planning.
        Radiother Oncol. 2018; 128: 349-356
        • Sharfo A.W.M.
        • Stieler F.
        • Kupfer O.
        • Heijmen B.J.M.
        • Dirkx M.L.P.
        • Breedveld S.
        • et al.
        Automated VMAT planning for postoperative adjuvant treatment of advanced gastric cancer.
        Radiat Oncol. 2018; 13https://doi.org/10.1186/s13014-018-1032-z
        • Buschmann M.
        • Sharfo A.W.M.
        • Penninkhof J.
        • Seppenwoolde Y.
        • Goldner G.
        • Georg D.
        • et al.
        Automated volumetric modulated arc therapy planning for whole pelvic prostate radiotherapyAutomatisierte volumenmodulierte Arc-Therapieplanung für Ganzbecken-Prostatabestrahlung.
        Strahlenther Onkol. 2018; 194: 333-342
        • Bijman R.
        • Rossi L.
        • Sharfo A.W.
        • Heemsbergen W.
        • Incrocci L.
        • Breedveld S.
        • et al.
        Automated Radiotherapy planning for patient-specific exploration of the trade-off between tumor dose coverage and predicted radiation-induced toxicity—A proof of principle study for prostate cancer..
        Front Oncol. 2020; 10https://doi.org/10.3389/fonc.2020.00943
        • Rossi L.
        • Cambraia Lopes P.
        • Marques Leitão J.
        • Janus C.
        • van de Pol M.
        • Breedveld S.
        • et al.
        On the importance of individualized, non-coplanar beam configurations in mediastinal lymphoma radiotherapy, optimized with automated planning.
        Front Oncol. 2021; 11https://doi.org/10.3389/fonc.2021.619929
        • Fjellanger K.
        • Hysing L.B.
        • Heijmen B.J.M.
        • Pettersen H.E.S.
        • Sandvik I.M.
        • Sulen T.H.
        • et al.
        Enhancing radiotherapy for locally advanced non-small cell lung cancer patients with ice, a novel system for automated multi-criterial treatment planning including beam angle optimization.
        Cancers (Basel). 2021; 13https://doi.org/10.3390/cancers13225683
        • Voet P.W.J.
        • Breedveld S.
        • Dirkx M.L.P.
        • Levendag P.C.
        • Heijmen B.J.M.
        Integrated multicriterial optimization of beam angles and intensity profiles for coplanar and noncoplanar head and neck IMRT and implications for VMAT.
        Med Phys. 2012; 39: 4858-4865https://doi.org/10.1118/1.4736803
        • Rossi L.
        • Sharfo A.W.
        • Aluwini S.
        • Dirkx M.
        • Breedveld S.
        • Heijmen B.
        First fully automated planning solution for robotic radiosurgery–comparison with automatically planned volumetric arc therapy for prostate cancer.
        Acta Oncol. 2018; 57: 1490-1498https://doi.org/10.1080/0284186X.2018.1479068
        • Schipaanboord B.W.K.
        • Breedveld S.
        • Rossi L.
        • Keijzer M.
        • Heijmen B.
        Automated prioritised 3D dose-based MLC segment generation for step-and-shoot IMRT.
        Phys Med Biol. 2019; 64https://doi.org/10.1088/1361-6560/ab1df9
        • Bijman R.
        • Rossi L.
        • Janssen T.
        • de Ruiter P.
        • van Triest B.
        • Breedveld S.
        • et al.
        MR-linac radiotherapy – the beam angle selection problem.
        Front Oncol. 2021; 11https://doi.org/10.3389/fonc.2021.717681
        • Schipaanboord B.W.K.
        • Giżyńska M.K.
        • Rossi L.
        • de Vries K.C.
        • Heijmen B.J.M.
        • Breedveld S.
        Fully automated treatment planning for MLC-based robotic radiotherapy.
        Med Phys. 2021; 48: 4139-4147https://doi.org/10.1002/mp.14993
        • Schipaanboord B.W.K.
        • Heijmen B.J.M.
        • Breedveld S.
        TBS-BAO: fully automated beam angle optimization for IMRT guided by a total-beam-space reference plan.
        Phys Med Biol. 2022; 67035004https://doi.org/10.1088/1361-6560/AC4B37
        • Hussein M.
        • Heijmen B.J.M.
        • Verellen D.
        • Nisbet A.
        Automation in intensity modulated radiotherapy treatment planning-a review of recent innovations.
        Br J Radiol. 2018; 91https://doi.org/10.1259/bjr.20180270
        • Bijman R.
        • Sharfo A.W.
        • Rossi L.
        • Breedveld S.
        • Heijmen B.
        Pre-clinical validation of a novel system for fully-automated treatment planning.
        Radiother Oncol. 2021; 158: 253-261https://doi.org/10.1016/j.radonc.2021.03.003
        • Ma T.M.
        • Emami B.
        • Grimm J.
        • Xue J.
        • Asbell S.O.
        • Kubicek G.J.
        • et al.
        Volume effects in radiosurgical spinal cord dose tolerance: how small is too small?.
        J Radiat Oncol. 2019; 8: 53-61https://doi.org/10.1007/s13566-018-0371-6
        • Alber M.
        • Reemtsen R.
        Intensity modulated radiotherapy treatment planning by use of a barrier-penalty multiplier method.
        Optim Methods Software. 2007; 22: 391-411https://doi.org/10.1080/10556780600604940
        • Krzywinski M.
        • Altman N.
        Visualizing samples with box plots.
        Nat Methods. 2014; 11: 119-120https://doi.org/10.1038/nmeth.2813