Author: Shaojie Chang, Thomas A. Foley, Hao Gong, Emily Koons, Shuai Leng, Cynthia H. McCollough, Eric E. Williamson 👨🔬
Affiliation: Mayo Clinic 🌍
Purpose: To enhance coronary CT angiography (cCTA) capabilities on conventional energy integrating detector CT (EID-CT) using artificial intelligence (AI). The AI framework incorporates high-resolution and multi-energy imaging features learned from photon-counting detector CT (PCD-CT) to improve EID-CT image quality, increasing spatial resolution and reducing blooming artifacts.
Methods: An AI-powered cCTA framework, named AI-cCTA, was developed which included two cascaded subnetworks focusing on high-resolution and multi-energy imaging. The network was trained on ultra-high resolution, multi-energy PCD-CT data (NAEOTOM Alpha, Siemens) and applied to conventional EID-CT. The high-resolution subnetwork utilizes PCD-CT images reconstructed with an identical kernel as EID-CT (e.g., Qr40) as inputs, and uses sharper kernel (Qr72) PCD-CT images as labels. The training dataset consists of paired low-resolution (Qr40) and high-resolution (Qr72) PCD-CT images from eight cCTA patients. The multi-energy subnetwork aligns PCD-CT virtual monoenergetic images (VMIs) with EID-CT’s effective energy levels. Specifically, 70 keV PCD VMIs, equivalent to EID-CT scans at 120 kV, are used as inputs, while paired 100 keV PCD VMIs are chosen as labels due to their effectiveness in reducing blooming artifacts. AI-cCTA’s performance was evaluated on a conventional EID-CT system (SOMATOM Force, Siemens) through qualitative visual inspection and quantitative line profile analysis and lumen diameter measurement.
Results: Visual inspection showed AI-cCTA images had fewer blooming artifacts and better lumen visualization than EID-CT. One case noted increased lumen diameter from 1.85 mm to 2.52 mm in a patient with dense calcifications. Line profile analysis confirmed AI-cCTA improved spatial resolution over traditional EID-CT. Compared to PCD-CT images of the same patient scanned on same day, AI-cCTA effectively minimizes blooming artifacts, providing calcium and lumen details closely approximating those of PCD-CT.
Conclusion: AI-cCTA equips EID-CT systems with high-resolution and virtual monoenergetic imaging capabilities and substantially reduces blooming artifacts, which could potentially improve the coronary artery disease diagnosis using cCTA.