RPI ID: 2020-111-201

Innovation Summary:
A deep learning network is designed to reduce metal artifacts in CT imaging by constraining feature disentanglement within a low-dimensional manifold. The model separates anatomical structures from artifact-induced distortions, improving image clarity. It is trained on synthetic and real-world datasets and supports integration with clinical imaging workflows. The approach enhances diagnostic accuracy in patients with implants.

Challenges / Opportunities:
Metal artifacts obscure critical anatomical details in CT scans, complicating diagnosis. This invention introduces a novel disentanglement strategy that isolates and suppresses artifact signals. It opens opportunities for improved imaging in orthopedics, dentistry, and oncology. The method is compatible with existing CT systems and PACS infrastructure.

Key Benefits / Advantages:
✔ Artifact suppression in CT imaging
✔ Low-dimensional feature disentanglement
✔ Enhanced diagnostic clarity
✔ Compatible with clinical workflows
✔ Applicable to multiple implant types

Applications:
• Medical imaging
• Orthopedic diagnostics
• Dental imaging

Keywords:
#CTimaging #metalartifacts #deeplearning #medicaldiagnostics #imageclarity #featuredisentanglement

Intellectual Property:
US Application 17/859186 US20230026961A1 filed 07-Jul-2022