Deep Learning-Based Reconstruction Algorithm With Lung Enhancement Filter for Chest CT: Effect on Image Quality and...

Area of Science:

• Radiology
• Medical Imaging
• Image Reconstruction

Background:

• Ground-glass nodules (GGNs) are crucial indicators in chest CT scans.
• Optimizing image quality and nodule sharpness is essential for accurate diagnosis, especially at low radiation doses.

Purpose of the Study:

• To evaluate the impact of a novel lung enhancement filter combined with deep learning image reconstruction (DLIR) on image quality and GGN sharpness.
• To compare this combined approach against traditional hybrid iterative reconstruction and DLIR alone.

Main Methods:

• Five artificial GGNs of varying densities were placed in an anthropomorphic phantom.
• CT scans were performed at four radiation dose levels using a 256-slice CT scanner.
• Images were reconstructed using adaptive statistical iterative reconstruction-V (AR50), DLIR (TrueFidelity - TF), and DLIR with the lung enhancement filter (TF + Lu).
• Image noise, signal-to-noise ratio, contrast-to-noise ratio, and nodule sharpness (full-width at half-maximum) were analyzed. Subjective image quality was also assessed.

Main Results:

• The TF + Lu and TF algorithms reduced image noise compared to AR50 (P = 0.001).
• TF + Lu significantly enhanced GGN sharpness across all radiation doses compared to TF alone (P = 0.001).
• Nodule sharpness with TF + Lu was comparable to AR50, while TF alone yielded the lowest sharpness.

Conclusions:

• Incorporating a lung enhancement filter with DLIR (TF + Lu) markedly improves GGN sharpness over DLIR alone (TF).
• TF + Lu is a promising technique for improving image quality and GGN evaluation in ultralow-dose chest CT imaging.