AI framework turns single-energy CT scans into high-contrast images
New deep learning model generates 50 keV images from standard CT without dual-energy hardware.
A team of researchers has developed a deep learning framework that can generate high-quality virtual monochromatic images (VMI) at 50 keV from standard single-energy CT (SECT) scans. Currently, achieving such contrast-enhanced images requires dual-energy CT (DECT) hardware, which is costly and limits clinical adoption. The proposed model uses contrast phase information—Angio, Arterial, Portal, and Delayed—as a conditional prior to guide the energy transformation from 70 keV (common in DECT training pairs) to 50 keV. By training on DECT-derived image pairs across all four phases, the unified model learns to produce phase-specific 50 keV-like images that preserve contrast dynamics.
The framework's architecture incorporates a novel prior conditioning mechanism that integrates the contrast phase label directly into the neural network, enabling a single model to handle all phases with high generalization. Experiments show the synthesised images closely match true 50 keV DECT images in terms of contrast enhancement and noise characteristics. This work could significantly reduce the hardware barrier for VMI in clinical settings, allowing hospitals with standard CT scanners to benefit from improved contrast resolution without purchasing expensive DECT systems. The model also maintains temporal consistency across phases, making it suitable for dynamic studies like perfusion imaging.
- Synthesizes 50 keV virtual monochromatic images from single-energy CT (SECT) data
- Trained on DECT-derived 70 keV and 50 keV pairs across four contrast phases: Angio, Arterial, Portal, Delayed
- A single unified model generalizes across phases, preserving contrast dynamics with a novel prior conditioning architecture
Why It Matters
Could expand access to high-contrast CT imaging by eliminating need for expensive dual-energy hardware.