Author: Andrew R. Godley, Steve B. Jiang, Mu-Han Lin, Austen Matthew Maniscalco, Dan Nguyen, Yang Kyun Park 👨🔬
Affiliation: Department of Radiation Oncology, UT Southwestern Medical Center, Medical Artificial Intelligence and Automation (MAIA) Lab, Department of Radiation Oncology, UT Southwestern Medical Center, UT Southwestern Medical Center, Medical Artificial Intelligence and Automation (MAIA) Lab & Department of Radiation Oncology, UT Southwestern Medical Center 🌍
Purpose:
Preparing DICOM datasets for research and education is challenging due to the complexity of the format and the necessity for patient-specific handling. Existing workflows demand substantial technical expertise and time to ensure data fidelity. To address these limitations, we developed a user-friendly, codeless graphical-user-interface(GUI), MedicalDataHandler, to simplify the reading, visualization, and processing of DICOM data for research and artificial intelligence(AI)–driven pipelines. This tool streamlines data preparation, reduces barriers for non-technical users, and enables reproducible handling without coding expertise.
Methods:
MedicalDataHandler was implemented in Python using the DearPyGUI(v2.0.0) toolkit. Users select folders containing DICOM files, which the tool reads and organizes by patient identifiers. CT images, structure sets, radiotherapy(RT) plans, and RT doses are grouped by shared metadata, producing a comprehensive table of accessible patient data. The interface supports interactive, real-time visualization in axial, coronal, and sagittal views with intuitive controls for scrolling, zooming, panning, and adjusting window width/level. Segmentation labels, colors, and data orientation are easily modified. Hovering over a voxel reveals its image/dose values and segmented structures. Multicore processing and multithreading facilitate rapid data inspection and conversion to the deep-learning–friendly NIfTI format. Advanced features include metadata inspection, adjustable voxel grid spacing, CT Hounsfield-Unit-to-Relative-Electron-Density mapping, and custom plan-sum dose file creation.
Results:
We validated MedicalDataHandler by cleaning data from 61 radiotherapy patients for input into training a deep-learning-based dose prediction model. It streamlined workflows by eliminating the need for complex, patient-specific code and enabling rapid preparation of research-ready datasets, improving resource efficiency and ensuring data consistency for subsequent AI applications.
Conclusion:
MedicalDataHandler reduces the technical barrier to DICOM data management and accelerates preprocessing, serving as a valuable resource for researchers and trainees. With intuitive visualization, flexible editing, and rapid data conversion, it empowers a broader audience to manage DICOM data efficiently and consistently for research and education.