Radiotherapy uses ionising radiation to kill cancerous cells. The radiation is generated by a clinical linear accelerator (linac). The radiation field can be shaped using collimators mounted in the treatment head of a gantry which is free to rotate around the patient lying on a treatment couch. In the UK, approximately 130,000 cancer patients are treated with radiotherapy every year, and around 40% - second only to surgery - of patients who are cured will have had radiotherapy as part of their treatment. Bespoke radiotherapy treatment plans are generated using propriety Treatment Planning Software (TPS). The type of treatment plan varies depending on the type of tumour and the method of radiotherapy delivery. In 3D conformal radiotherapy, a small number (usually 2-4) of beams at different gantry angles are directed at the target. Alternatively, inverse planning methods can be utilised to generate even more conformal treatment plans. In Volumetric Modulated Arc Therapy (VMAT), the target is continuously irradiated as the gantry rotates around the patient whilst varying the aperture shape, the intensity of delivered radiation, and the gantry speed. These three parameters are optimised by the TPS by minimising a user-defined objective function that consists of dosimetric objectives to both the cancerous target and the surrounding healthy tissue and organs.
Aims: The apertures generated during a VMAT optimisation are often small and irregular which can potentially lead to inaccuracies during radiotherapy delivery. In this work we will develop a software tool to read and analyse information generated by the Pinnacle TPS (Philips, WI, USA) for VMAT radiotherapy plans with the ultimate goal of being able to predict which VMAT treatment plans are likely to exhibit dosimetric inaccuracies during radiotherapy delivery and to adjust these plans as necessary.
Methods, results and future work: Python is used to parse the VMAT treatment plan data that is generated by the Pinnacle TPS. Data relevant to potential dosimetric inaccuracies (aperture size and shape, the gantry speed, and the dose rate at every gantry angle) is extracted into a secondary structure of python objects. This flattens the deeply nested object tree to simplify further numerical analysis. By using the simplejson encoder (https://pypi.python.org/pypi/simplejson/), this reduced structure can be then serialised to and deserialised from a standard data format (JSON), further easing algorithm development by reducing subsequent parse times and exposing the data for future work regardless of language.
The numpy (http://www.numpy.org/) and matplotlib (http://matplotlib.org/) python packages have been utilised to develop and visualise aperture-based metrics in order to quantify the complexity of the field shape. The ‘edge effect’ metric is defined as the sum (for every gantry angle) of the ratio of the aperture’s perimeter length and area. The metric is multiplied by factor related to the amount of radiation that is delivered at that shape in order to ensure that highly-weighted small and irregular shapes influence the metric to a larger extent than the same shape would but with lower radiation intensity. We plan to correlate the ‘edge effect’ metric to the accuracy of treatment delivery and to ultimately introduce this metric into the optimisation algorithm of the TPS in order to penalise VMAT plans with complex and irregular shapes which we hypothesize should result in more accurate dose delivery.