RPI ID: 2018-007-401

Innovation Summary:
This invention introduces a 3D convolutional autoencoder that enhances low-dose CT image quality by learning from high-quality 2D training data. The system leverages transfer learning to bridge the gap between 2D and 3D imaging, enabling efficient training and robust performance. It corrects artifacts and improves image fidelity, making it suitable for clinical deployment. The model is optimized for speed and accuracy in real-world diagnostic environments.

Challenges / Opportunities:
Training 3D models for CT correction is typically data-intensive and computationally demanding. This invention overcomes those barriers by using 2D-trained networks as a foundation, significantly reducing training requirements. It provides a scalable solution for improving diagnostic imaging while maintaining low radiation exposure. The technology supports the growing demand for AI-driven tools in radiology and medical imaging.

Key Benefits / Advantages:
✔ Transfer learning reduces training time and data requirements
✔ 3D correction enhances volumetric image quality
✔ Artifact reduction improves diagnostic confidence
✔ Clinically ready design for integration into imaging systems

Applications:
• Radiology and diagnostic imaging
• AI-assisted CT interpretation
• Low-dose screening programs
• Medical imaging research

Keywords:
#3Dautoencoder #CTcorrection #transferlearning #medicalAI #imageenhancement

Intellectual Property:
US Issued Patent 11,580,410 B2