Room: Karl Dean Ballroom A1
Purpose: DVH Analytics was designed in an open and vendor-agnostic manner to collect DICOM-RT data into a scalable database and provide analytical tools such as a time-series plot, correlation matrix, and multi-variable linear regression.
Methods: DVH Analytics uses Python code to parse data from RT Plan, RT Dose, and RT Structure DICOM files using PyDICOM; DVH data is calculated using dicompyler. Data from more than 80 DICOM tags and 1-cGY binned DVHs are stored in a SQL database; PTV overlap and distances are calculated. DVHs are categorized based on user preferences, allowing for naming variations. Although no imaging data is parsed, a catalogue is kept of all input DICOM files. This code was designed to accommodate linac, brachytherapy, and proton based treatments. DVH endpoints, EUD, TCP, and NTCP can be calculated and analyzed across all retrieved DVH data. Numerical data can be plotted over time along with a rolling average, percentile limits, and histograms. Numerical datasets can be plotted against each other for the purposes of building a multi-variable regression.
Results: Using a single physician's head-and-neck practice, a database of over 3,000 DVHs across 89 patients was imported. Differences in means, correlations, and temporal trends were observed. For demonstration purposes, the multi-variable regression tool was used to generate a model to predict maximum dose to the brainstem based on the minimum distance to the PTV and the treatment beam average SSD.
Conclusion: A free and open-source application has been developed that can store, organize, parse, and analyze non-image based DICOM-RT data. This software provides means to query large patient datasets, display ROIs, view data over time, perform statistical tests, and build predictive models.