Vienna, Austria

ESTRO 2023

Session Item

May 14
10:30 - 11:30
Translational outputs from clinical trials
Chloe Brooks, United Kingdom;
Jens Overgaard, Denmark
Proffered Papers
10:40 - 10:50
Assessing plan quality in the ‘PLATO anal cancer trial 5’ pilot phase with automated planning
Megan Barrell, United Kingdom


Assessing plan quality in the ‘PLATO anal cancer trial 5’ pilot phase with automated planning

Megan Barrell1,1, Natalie Abbott1, Richard Adams2, Maria Hawkins3, David Sebag-Montefiore4, Anthony Millin1, Philip Wheeler1

1Velindre Cancer Centre, Physics, Cardiff, United Kingdom; 2Velindre Cancer Centre, Oncology, Cardiff, United Kingdom; 3University College London, Medical Physics and Bioengineering, London, United Kingdom; 4Leeds Cancer Centre, Oncology, Leeds, United Kingdom

Show Affiliations
Purpose or Objective

Treatment efficacy relies on plan quality. Within trials, plan quality may vary due to training and equipment differences, which may influence treatment outcome or trial results. This study uses automated planning to assess plan quality and variation within the PersonaLlising rAdioTherapy dOse (PLATO) Anal Cancer Trial 5 (ACT5).

Material and Methods

A protocol based automatic iterative optimisation (PBAIO) planning solution [1], implemented in RayStation, was calibrated for anal cancer using 5 pre-trial benchmark patient plans and 10 non-trial patients.

Plans were generated for the pilot phase of PLATO ACT5; a dataset of 51 patients from 11 centres. Patients with prosthetic hips, replans, or unavailable suitable planning data were excluded (n=9). All trial plans were approved by the PLATO national trials QA team.

The trial and automated plans were quantitatively compared using the ACT5 planning protocol parameters, small bowel V15Gy in cm^3, and planning target volume (PTV) conformity index (CI) and homogeneity index (HI). Statistical analysis was completed using a Wilcoxon signed rank test.


At a population level, automation generally yielded higher quality plans with less variation when compared to trial plans. Automation reduced mandatory and optimal objective failures from 4 to 3 and 137 to 80 respectively.

34/46 metrics showed statistically significant (p<0.05) differences between automated and trial plans (Table1). Automation significantly reduced OAR dose. Genitalia D50% and D35% reduced by >5.5Gy, femoral heads (FHs) by >2.5Gy and bladder D50% by 1.8Gy. Small bowel D200cc and D150cc reduced by 5.0Gy, and V15Gy by 41cm3. These reductions did not adversely impact PTV D98%, D2%, HI or CI, which were within 0.6Gy, 0.6Gy, 0.018, and 0.017 respectively.

At a per patient level, substantial variation in the difference between trial and automated plan metrics indicated noteworthy plan quality variability (Figure1). For the genitalia and FHs, interquartile range (IQR) of the difference (trial-auto) was largest for D35%; 5.8Gy and 5.2Gy respectively. For the bladder, D50% IQR was 4.5Gy. The small bowel D200cc and V15Gy IQRs were 7.7Gy and 46cm3 respectively. Meaningful variations in PTV D98%, D2%, CI and HI were also observed (Figure1) with IQRs of up to 2.4Gy, 2.4Gy, 0.018, and 0.060 respectively.


Automated planning highlighted significant variations in plan quality within the pilot phase of PLATO ACT5. Evaluating plan quality in this manner may encourage improvements in training, QA and future trial approaches. This may reduce variation and improve overall plan quality.


[1] P. Wheeler, “Utilisation of Pareto navigation techniques to calibrate a fully automated radiotherapy treatment planning solution”, Phys Img Radiat Oncol, vol. 16, no. 10, pp. 41-48, 2019