Intentional creation of suboptimal, realistic dose distributions
Open Access
- 27 October 2025
- journal article
- research article
- Published by Wiley in Journal of Applied Clinical Medical Physics
- Vol. 26 (11) , e70305
Abstract
Background: Radiation oncology residents report a lack of understanding and confidence in assessing radiotherapy plan quality. A contributing factor is the environment in which plan review is taught during residency, that is, routine clinical practice, which does not provide ample time for self‐guided practice in a low‐stakes setting. Expertise in plan review requires diverse case presentation and many examples, which are often not achievable in smaller programs and for less common cancer types. As plan quality affects patient outcomes, it is important to address these pitfalls in the education of residents on plan review. Purpose: To address the identified pitfalls of clinic‐based training, we have developed techniques to create realistic dose distributions that appear suboptimal in a controllable way. These plans can provide many more case examples in the training curriculum and present a low‐stakes technique for safe and effective education of radiation oncology residents. Methods: High‐quality dose distributions were first generated with a pre‐trained deep learning model (trained using only high‐quality plans). The dose distributions were then altered directly to create three classes of suboptimal dose distributions: (1) decreased organ‐at‐risk sparing, (2) decreased target conformality, and (3) hotspots in the target. Experienced clinicians then reviewed a subset of these suboptimal dose distributions to assess realism. Results: We successfully decreased the quality of radiotherapy dose distributions. The decreased organ‐at‐risk sparing, decreased target conformality, and increased target hotspots were statistically significant ( p < 0.05) when assessed by dose‐volume histogram metrics for all parameters evaluated, and the magnitude of dose change was controllable. The resulting dose distributions were overall scored by experienced clinicians as realistic. Conclusion: In this study, we developed techniques to generate realistic but suboptimal dose distributions. The techniques operate directly on existing dose distributions without the need for a treatment planning system and produce dose distributions that appear realistic to experienced clinicians.Keywords
This publication has 15 references indexed in Scilit:
- OpenKBP: The open‐access knowledge‐based planning grand challenge and datasetMedical Physics, 2021
- Technical Note: Dose prediction for head and neck radiotherapy using a three‐dimensional dense dilated U‐net architectureMedical Physics, 2021
- Development and dosimetric assessment of an automatic dental artifact classification tool to guide artifact management techniques in a fully automated treatment planning workflowComputerized Medical Imaging and Graphics, 2021
- A Blinded, Prospective Study of Error Detection During Physician Chart Rounds in Radiation OncologyPractical Radiation Oncology, 2020
- An Innovative Learning Tool for Radiation Therapy Treatment Plan Evaluation: Implementation and EvaluationInternational Journal of Radiation Oncology - Biology - Physics (IJROBP), 2020
- Three‐dimensional dose prediction for lung IMRT patients with deep neural networks: robust learning from heterogeneous beam configurationsMedical Physics, 2019
- A feasibility study for predicting optimal radiation therapy dose distributions of prostate cancer patients from patient anatomy using deep learningScientific Reports, 2019
- DoseNet: a volumetric dose prediction algorithm using 3D fully-convolutional neural networksPhysics in Medicine & Biology, 2018
- Quantifying Unnecessary Normal Tissue Complication Risks due to Suboptimal Planning: A Secondary Study of RTOG 0126International Journal of Radiation Oncology - Biology - Physics (IJROBP), 2015
- Experience-Based Quality Control of Clinical Intensity-Modulated Radiotherapy PlanningInternational Journal of Radiation Oncology - Biology - Physics (IJROBP), 2011