Deep learning based ultra-low dose fan-beam computed tomography image enhancement algorithm: Feasibility study in...

Area of Science:

• Medical Imaging
• Radiotherapy
• Artificial Intelligence

Background:

• Clinical radiotherapy requires high-quality imaging for accurate tumor targeting and treatment planning.
• Ultra-low dose computed tomography (LDCT) reduces radiation exposure but often compromises image quality.
• Deep learning offers potential solutions for enhancing LDCT images to meet clinical standards.

Purpose of the Study:

• To evaluate the feasibility of a deep learning-based algorithm for enhancing ultra-low dose kV-fan-beam CT (kV-FBCT) images.
• To assess the clinical applicability of the enhanced LDCT images for abdominal and pelvic tumor radiotherapy.

Main Methods:

• A CycleGAN-based deep learning model was developed for image enhancement.
• 76 patients with abdominal and pelvic tumors were prospectively included.
• Images were analyzed using subjective and objective metrics, comparing normal dose CT (NDCT), LDCT, and deep learning-enhanced CT (DLR).

Main Results:

• DLR significantly reduced image noise and improved contrast-to-noise ratio (CNR) compared to LDCT, approaching NDCT levels.
• Low-density resolution and spatial frequencies (MTF10, MTF50) were significantly improved in DLR compared to LDCT and slightly exceeded NDCT.
• Subjective image quality scores for DLR were comparable to NDCT, with statistically significant improvements over LDCT.

Conclusions:

• Deep learning-enhanced LDCT (DLR) achieves image quality close to NDCT while maintaining reduced radiation dose.
• The DLR technique meets the quality requirements for conventional image-guided adaptive radiotherapy (ART).
• This method provides a viable technical foundation for LDCT-guided ART in clinical practice.