A Native Strategy for Integrating Deep-Learning Models for Segmentation into a Radiological Viewer

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Deep learning (DL) models show promise for automating medical imaging diagnostics.
• Clinical integration of DL solutions remains a significant challenge due to workflow complexity and technical barriers.

Purpose of the Study:

• To propose and validate a native integration strategy for DL segmentation models within a DICOM viewer.
• To enhance clinical adoption and usability of DL tools in medical imaging workflows.

Main Methods:

• Developed a native integration strategy embedding DL model execution directly into a CE-marked open-source DICOM viewer.
• Created a dedicated DL module within the viewer, eliminating the need for external software or complex APIs.
• Validated the approach using vertebral bodies and liver segmentation use cases with models from different DL libraries.

Main Results:

• The native integration strategy allows direct use of DL segmentation models within the DICOM viewer.
• The method is compatible with heterogeneous DL architectures and requires minimal user interaction.
• Clinical usability is preserved without disrupting existing diagnostic workflows.

Conclusions:

• The proposed methodology offers a simple, flexible, and clinically ready solution for viewer-native DL deployment.
• This approach facilitates the adoption of DL tools in regulated healthcare environments.
• Enables efficient sharing and reuse of DL models across institutions, advancing medical imaging diagnostics.