DVH Analysis: A Custom, Scalable Software Platform Designed for Flexible Dosimetric Quantitative Evaluation

Abstract

Quantitative dosimetric analysis of an arbitrary number of related DVHs is not a feature of most commercial treatment planning systems. We have developed a robust, flexible software platform capable of providing a simple-to-use interface for complex DVH analysis that can accept an arbitrary amount of DVHs. Using Python 3.6 and the Pandas open-source library, we built a software package that will read in an arbitrary number of DVHs exported from a treatment planning system as a .csv file. The program will parse the text of the DVH file to construct a well-formatted Pandas Dataframe for each DVH. Important meta-data such as ROI volumes and patient information are maintained for each DVH. Standard dosimetric queries can be passed in a flexible way, allowing for fine-tuned control over which ROIs are analyzed. Queries can be specified to return both relative and absolute dose/volume parameters. Statistical analysis functions from the SciPy library are incorporated into the software for advanced analytic capabilities. Robust plotting functionality from the Matplotlib library was incorporated to easily visualize these large datasets. Using this platform we navigated the dosimetric analysis of 216 generated plans for a recent dosimetric analysis comparing protons to photons for use in treating renal tumors. This represented 2376 individual DVH curves. At a 1 cGy bin size, this amounted to 13,305,600 dose bins. We were able to easily query volume and dose parameters for any ROI, comparing means and statistical measures of variance and significance. All data analysis results were reviewed and confirmed. This allowed us to focus our effort on generating radiotherapy plans, rather than laborious statistical analysis. We were able to successfully build a useful tool for DVH data analysis. This tool has enabled large scale DVH-analytic projects without the need for additional statistician support. It has acted as a force multiplier, allowing residents physicians and other non-statisticians to rigorously interrogate large scale DVH data intuitively in an efficient manner. Leveraging Python and open source data-processing libraries can be used to create powerful, radiation-oncology specific tool sets to enhance research scope and statistical analysis.