Automated Treatment Planning in Radiation Oncology

Automated treatment planning harnesses computational algorithms to generate radiotherapy plans with minimal human intervention. By integrating patient anatomy, tumour volumes and organ-at-risk delineations, these systems optimise dose distributions to maximise target coverage while sparing healthy tissue. Early knowledge-based planning methods relied on statistical models trained on libraries of prior clinical cases to predict achievable dose-volume objectives. More recent approaches employ machine learning and deep convolutional networks to infer voxel-level dose maps directly from contours or imaging data. Automation reduces inter-planner variability, shortens planning times from hours to minutes and supports adaptive strategies in which plans are updated in near real time. This shift towards data-driven and artificial-intelligence-enabled workflows promises consistent, high-quality treatment across diverse tumour sites and clinical settings, accelerating access to precision radiotherapy worldwide.

Research from Nature Portfolio

Recent studies have demonstrated that convolutional neural networks can accurately predict three-dimensional dose distributions from patient anatomy and target/OAR contours. A modified U-net architecture applied to prostate IMRT achieved an average Dice similarity of 0.91 between predicted and clinical isodose volumes and kept absolute dose deviations below 5 % across both targets and critical structures. This foundational work shows that deep learning can guide optimisation engines, potentially bypassing iterative manual adjustment and enabling automated generation of clinically acceptable plans with limited training data.

Automated Treatment Planning in Radiation Oncology publication trend

The graph below shows the total number of articles in automated treatment planning in radiation oncology across all publications each year (not limited to Nature Index journals).

Technical terms

Knowledge-based planning (KBP): A data-driven method that derives optimisation objectives from historical treatment plans to guide new plan generation.

Intensity-modulated radiation therapy (IMRT): A technique that modulates beam intensity across multiple fields to conform dose to complex tumour geometries.

Volumetric modulated arc therapy (VMAT): A rotational form of IMRT in which dose rate, gantry speed and beam shape vary continuously around the patient.

Planning target volume (PTV): A geometric expansion of the clinical target to account for motion and setup uncertainties.

Organ at risk (OAR): A normal tissue structure whose radiation tolerance limits the dose deliverable to the PTV.

Deep convolutional network: A multilayered machine-learning model that learns spatial hierarchies of features to predict outputs such as dose distributions.

References

1. A feasibility study for predicting optimal radiation therapy dose distributions of prostate cancer patients from patient anatomy using deep learning. Scientific Reports (2019).
2. Knowledge‐based planning for intensity‐modulated radiation therapy: A review of data‐driven approaches. Medical Physics (2019).
3. Evaluation of an automated knowledge based treatment planning system for head and neck. Radiation Oncology (2015).
4. DeepDose: Towards a fast dose calculation engine for radiation therapy using deep learning. Physics in Medicine and Biology (2020).