Clinical Scorecard: Creation of Multi-Modal Datasets for Federated Learning Utilizing DICOM-Formatted Reports
At a Glance
Category
Detail
Condition
Multi-modal medical data integration for federated learning
Key Mechanisms
Use of DICOM standard structured reports for harmonized multi-modal data representation and federated cohort selection
Target Population
Medical institutions and researchers managing multi-modal imaging, waveform, and metadata datasets
Care Setting
Multi-center clinical research and federated learning environments
Key Highlights
Leveraging DICOM standards enables harmonized storage and querying of multi-modal medical data including images, waveforms, segmentations, and metadata.
An intuitive, non-programming graphical dashboard facilitates multi-center cohort creation and iterative filtering of DICOM data in PACS.
Publicly available Python scripts and SR templates support conversion and creation of DICOM structured reports for deep learning pipelines.
Guideline-Based Recommendations
Diagnosis
Utilize DICOM structured reports to standardize multi-modal medical data for consistent cohort identification across centers.
Management
Convert all local multi-modal data into DICOM format and upload into research PACS to enable federated data harmonization.
Implement federated learning by locally training models on harmonized DICOM data and aggregating model weights centrally.
Monitoring & Follow-up
Use the graphical dashboard to iteratively filter and refine cohorts, ensuring data completeness and modality presence.
Apply standardized queries across locations to maintain consistency in cohort selection.
Risks
Non-adherence to standardized DICOM templates may impair data harmonization and federated learning efficacy.
Privacy regulations restrict centralized data sharing; federated learning and local data processing mitigate this risk.
Patient & Prescribing Data
Patients with multi-modal imaging and waveform data distributed across multiple clinical centers
Federated learning on harmonized DICOM datasets enables development of robust predictive models without compromising patient privacy.
Clinical Best Practices
Adopt DICOM structured reports as the single data format for multi-modal medical data to simplify interoperability and federated cohort selection.
Deploy graphical filtering dashboards that allow users without programming expertise to create and refine multi-center cohorts.
Ensure strict adherence to DICOM templates to maintain data validity and enable effective federated learning workflows.
Leverage DICOM’s inherent matching capabilities to link images, waveforms, and metadata seamlessly within PACS.
by Malte Tölle, Lukas Burger, Halvar Kelm, Florian André, Peter Bannas, Gerhard Diller, Norbert Frey, Philipp Garthe, Stefan Groß, Anja Hennemuth, Lars Kaderali, Nina Krüger, Andreas Leha, Simon Martin, Alexander Meyer, Eike Nagel, Stefan Orwat, Clemens Scherer, Moritz Seiffert, Jan Moritz Seliger, Stefan Simm, Tim Friede, Tim Seidler, Sandy Engelhardt
