RPI ID: 2019-005-201 / 2019-005-601

Innovation Summary:
This invention introduces a deep learning framework called GAN-Circle for super-resolution reconstruction in computed tomography (CT) imaging. The system is based on a generative adversarial network (GAN) architecture constrained by three synergistic learning strategies: identical mapping, residual learning, and cycle consistency. This ensemble enables the model to generate high-resolution CT images from low-resolution inputs while preserving anatomical accuracy and minimizing artifacts. GAN-Circle is designed for clinical integration, offering a scalable solution for enhancing image quality in low-dose or fast-scan CT protocols.

Challenges / Opportunities:
Conventional CT super-resolution techniques often struggle to balance image sharpness with artifact suppression, especially when training data is limited or noisy. GAN-Circle addresses these limitations by combining multiple learning constraints that guide the network toward anatomically faithful reconstructions. This opens opportunities for safer, faster, and more accurate CT imaging in clinical diagnostics, particularly in oncology, cardiology, and emergency care.

Key Benefits / Advantages:
✔ High-resolution CT reconstruction from low-resolution inputs
✔ GAN constrained by identical, residual, and cycle learning
✔ Reduces noise and artifacts while preserving fine detail
✔ Adaptable to various CT imaging protocols and scanner types
✔ Supports low-dose imaging and rapid acquisition workflows

Applications:
• Oncology and cardiovascular diagnostics
• Low-dose and pediatric CT imaging
• Emergency and trauma care
• AI-enhanced radiology systems
• Medical imaging research and development

Keywords:
#CTsuperresolution #GANCircle #deeplearning #medicalimaging #imageenhancement #lowdoseCT

Intellectual Property:
US Issued Patent(s) 11,232,541 B2; 11,854,160 B2