Author: Annie Cooney, Kenneth L. Homann, Somayeh Taghizadehghahremanloo, Adam D. Yock, Hong Zhang 👨🔬
Affiliation: Assistant Professor, Vanderbilt University Medical Center 🌍
Purpose:
Digital radiotherapy data has been standardized using the DICOM format. However, different radiotherapy software environments interpret identical data differently due to inherent software development decisions. This work investigates how shape characteristics like size and curvature affect the variable representation of standardized datasets across multiple software environments. In addition, it assesses their influence on the accuracy of three morphological operations –expansion, contraction, and Boolean subtraction– that are critical during the treatment planning process
Methods:
A Python script was used to generate ground truth DICOM radiotherapy structure files containing regular polyhedra (tetrahedra, hexahedra, and dodecahedra) with side lengths ranging from 5-50 mm in 5-mm increments. These shapes were represented as polygonal slices stacked with a 1-mm image slice thickness. They were chosen for their well-defined geometry, precisely characterized properties, and consistent interior angles (acute, right, or obtuse). The ground truth files were imported into multiple software environments (Eclipse and MIM). The structure volumes calculated by these environments were compared to the analytical volumes of the original polyhedra. Each polyhedron was then expanded and contracted by 1-5 mm in 1-mm increments, and the resulting margin volumes were compared to analytical volumes derived using a rolling ball algorithm with infinite precision. Finally, Boolean subtractions were performed where polyhedra were subtracted from one another, and the resulting volumes were compared analytically. Differences in the performance of these operations between software environments were evaluated and analyzed.
Results:
Compared to the analytical volume, each software environment represented polyhedra more accurately when they had longer side lengths, larger interior angles, and larger magnitude operations. The two software environments demonstrated systematic differences in their accuracy.
Conclusion:
The accuracy and consistency that different software environments represent radiotherapy structures, and the result of subsequent morphological operations depends on the specific software environment and the size and curvature of the structure.