Dynamic Representation of Anatomical Structures in Radiation Therapy Treatment Planning and Evaluation
Segmentation is a ubiquitous operation in radiation therapy (RT) and in medical image computing (MIC) in general. Various data representations can describe segmentation results, such as binary volumes or surface models. Conversions between them are often required, which include complex data processing. There are challenges involved in managing multiple representations, which pose a hindrance to research application development and usability. The challenges are related to conversion method selection, data provenance, consistency, coherence of in-memory objects, and fidelity. The fundamental contribution is a software methodology for dynamic management of multiple segmentation representations, which facilitates RT and MIC research software development and improves end-user experience. At the core, a complex data container preserves identity and provenance of the contained representations and ensures data coherence. Conversions from one representation to another are executed automatically. A graph containing the conversion algorithms determines each execution, ensuring consistency between representations. The accuracy of the core conversion algorithms was evaluated. The software infrastructure is made available as an open-source library, based on which a manual and semi-automated segmentation application was created that has become one of the most versatile segmentation tools available, corroborated by the number of projects using it worldwide. Numerous clinical research applications have also been developed based on the proposed framework, facilitating different aspects of radiation therapy research: gel and film dosimetry analysis, MRI-ultrasound contour propagation, and external beam planning system. An issue related to a typical conversion step is explored, focusing on the calculation of a universal RT plan evaluation metric. Accuracy depends on spatial resolution and thus the conversion parameters, which need to be different based on structure size and complexity. A fuzzy-based algorithm is presented to calculate the parameters for each structure. In summary, a software methodology is proposed for dynamic management of representations in image segmentation, offering a solution to the challenges that arise when working with multiple representations. The implemented infrastructure facilitates rapid and robust application development, and allows creating more user-friendly software. The community impact of the applications developed using the framework and the ongoing research and education projects show its potential in the clinical and research community.
URI for this recordhttp://hdl.handle.net/1974/26422
Request an alternative formatIf you require this document in an alternate, accessible format, please contact the Queen's Adaptive Technology Centre
The following license files are associated with this item: