IODeep: An IOD for the introduction of deep learning in the DICOM standard
These videos have been matched automatically. Contact us if they are not relevant.
These videos have been matched automatically. Contact us if they are not relevant.
View abstract on PubMed
Summary
This summary is machine-generated.IODeep, a new DICOM Information Object Definition, integrates trained Artificial Intelligence (AI) models into clinical workflows. This enables tailored AI for radiology, improving physician decision-making and diagnostic accuracy.
Area Of Science
- Biomedical Imaging
- Artificial Intelligence in Medicine
- Medical Informatics
Background
- Deep Neural Networks (DNNs) show promise in biomedical image segmentation but are underutilized in clinical practice.
- Integration into diagnostic workflows and explainability are key barriers to AI adoption in healthcare.
- Standardization is crucial for the effective use of AI tools in clinical settings.
Purpose Of The Study
- To introduce IODeep, a novel DICOM Information Object Definition (IOD) for storing trained DNN models.
- To facilitate the seamless integration of AI models into existing Picture Archiving and Communication Systems (PACS).
- To support the tailoring of AI models to specific clinical data and improve diagnostic decision-making.
Main Methods
- Developed IODeep to store DNN architecture and weights, linked to image acquisition details (modality, anatomy, disease).
- Implemented a DNN selection algorithm for PACS retrieval based on metadata labels.
- Created a PACS viewer for demonstrating DICOM integration without PACS modification.
Main Results
- IODeep enables full integration of trained AI models within a DICOM infrastructure.
- Demonstrated effective DICOM integration with a dedicated PACS viewer.
- Developed a service-based architecture supporting the AI model integration workflow.
Conclusions
- IODeep facilitates the integration of trained AI models into DICOM environments.
- Supports model fine-tuning with local hospital data or federated learning across institutions.
- Enables AI models to be adapted to real-world radiology data, enhancing physician decision support.
These videos have been matched automatically. Contact us if they are not relevant.
These videos have been matched automatically. Contact us if they are not relevant.