National radiotherapy imaging databank
Medical imaging is used during diagnosis, treatment planning, and follow-up after radiotherapy. A national image data bank to support the sharing of DICOM information among all Danish RT centers was developed recently. Further automation and workflow improvements will be implemented to ease the routine use of the data bank. The database will be extended to extract biomarkers from DICOM images. The calculation will be open source to aid individual centers to perform similar calculations locally if relevant.
The project focuses on optimization of the workflow related to the Danish radiotherapy data bank which is hosted and developed at Odense University Hospital (OUH). The aim is to reduce the amount of manual work needed for clinical trials to store, structure, share, and extract relevant information for clinical trials. The current implementation of the data bank is based on a manual assignment of treatment plans to individual studies which can be labor-intensive in a busy clinic; therefore focus will be on automating this process such that the assignment becomes less time-consuming. Furthermore, there will be implemented possibilities to calculate radiomics features such that individual centers can obtain features directly from the data bank for clinical studies.
A national DICOM data bank has been in clinical operation since 2010 in Denmark [Westberg et al J Phys Conf Ser, 2014;489,12100]. In a recent publication [Brink et al Acta oncologica, 2018;57,107], it was shown that the data quality from the data bank exceeds the quality of manually reported data which has been the standard in previous studies. Beside improved data quality the aim of the DICOM data bank is to provide easy access to the rich information which is available in all the medical images and treatment plans; something which is not accessible if the data are reported manually. The data bank has been used by a number of clinical randomized trials and has facilitated retrospective analysis of NTCP (normal tissue complication probability) models.
Data are provided to the data bank by a DICOM transfer of data from the treatment planning computers in the participating centers. However, the assignment of the data to individual clinical studies must currently be performed manually within a web interface of the data bank. Since this step is not a natural part of the clinical treatment planning process, and thus not an integrated step in the clinical workflow, it is a barrier for a larger degree of utilization of the national image data bank.
The wealth of information that is available in the medical images in the data bank has until now only been utilized to get a better description of the dose distribution, while the image information has not been quantified in a way useable for clinical studies. A possible way to quantify the available image information is to calculate radiomics features which currently are investigated in a number of publications as a method to improve the predictability of clinical outcome models. Specific implementations of feature extraction can impact the calculated value considerably; it would thus be beneficial if the data bank could provide a basic set of features that could be used in national trials in order to ensure comparable values across participating centers. The features in the data bank will be provided as open access, and a framework to implement the user’s own features will be developed.
The current project focuses on the development of new software tools for the national data bank; with the focus of reducing the manual workload for the individual user of the database. The development will mainly be related to the following issues:
- Creation of simple user instructions that can be downloaded from the data bank website
- Enable automatic assignment of follow-up images to specific trials in which data for the same patient have already been assigned.
- Based on predefined unique DICOM information assign a patient automatically to a given clinical study in the data bank. A possible solution is to define an empty unique structure name in the radiotherapy planning systems which can be used as a marker for inclusion in a specific clinical trial.
- Provide a basic set of image feature calculations that can be used for radiomics analysis. The features will be calculated within the structures provided in the radiotherapy plan.
- Make a framework that supports the calculation of features developed at the individual local centers using the data bank.
- Make it possible in batch mode to download full DICOM data from a specific clinical trial within the web interface of the data bank
- Optimize workflow as requested by the individual centers
The expected outcome of the current project is an improved data bank with extended functionality and reduced workload on the participating centers, such that the clinical studies will have even easier access to storage of data on a national scale as well as easy access to retrieve aggregated or raw data from the database.
The current project provides a tool for all clinical trials that are interested in sharing medical image data across centers. Implementation of the current project will facilitate a smooth use of the national data bank. There are no ethical issues related to the current project since the approvals of storing the patient confidential data are already available and no part of the current project will jeopardize the security of the current database.