Quantitative Evaluation of Low-Dose CT Image Quality Using Deep Learning Reconstruction: A Comparative Study of...

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Healthcare

Background:

• Reducing radiation exposure in CT scans is crucial for patient safety, especially in routine and repeat imaging.
• Deep learning image reconstruction (DLIR) offers a promising solution for maintaining diagnostic image quality at reduced radiation doses.

Purpose of the Study:

• To compare the performance of two leading DLIR algorithms, Philips Precise Image (PI) and GE TrueFidelity (TF), under low-dose CT conditions.
• To evaluate the effectiveness of different DLIR presets in preserving image quality and diagnostic information.

Main Methods:

• The AAPM CIRS-610 phantom was scanned using low-dose 80 kVp protocols on Philips CT 5300 and GE Revolution CT scanners.
• Images were reconstructed with five DLIR presets: PI (Smoother, Standard, Sharper) and TF (Middle, High).
• Quantitative metrics including SNR, CNR, nRMSE, PSNR, SSIM, FSIM, UQI, GMSD, and gradient magnitude were used for evaluation.

Main Results:

• GE TrueFidelity (TF)-High demonstrated the highest Signal-to-Noise Ratio (SNR), showing a 54-57% improvement over Philips Precise Image (PI)-Smoother.
• TF-High also achieved superior Peak Signal-to-Noise Ratio (PSNR) and the lowest Geometric Mean Squared Difference (GMSD), indicating better structure preservation.
• Philips Precise Image (PI)-Sharper provided the strongest gradient magnitude, effectively enhancing fine edge details.

Conclusions:

• GE TrueFidelity (TF)-High offers an optimal balance between noise reduction and structural fidelity for low-dose CT imaging.
• Philips Precise Image (PI)-Sharper is effective for highlighting fine details when edge enhancement is prioritized.
• The choice of DLIR algorithm and preset should be tailored to specific clinical requirements in low-dose CT protocols.