Evaluation of the time‑of‑flight-enhanced deep learning image reconstruction method in 18F‑FDG PET/CT for breast cancer imaging.

BACKGROUND

Breast cancer is the most frequently diagnosed cancer among women. Accurate diagnosis and effective management rely heavily on high-quality positron emission tomography (PET) imaging. A novel time-of-flight (TOF)-enhanced deep learning reconstruction (DLR) technique has recently been introduced for the Omni Legend (GE Healthcare) PET/CT system. However, its clinical utility in breast cancer imaging has not yet been fully established. This study aims to assess the impact of the DLR method on 18F-FDG PET/CT imaging in patients with breast cancer.

METHODS

This retrospective study included 30 female breast cancer patients who underwent 18F-FDG PET/CT using the Omni Legend system. PET images were reconstructed using the Bayesian penalized likelihood (BPL) method and a DLR method with three TOF enhancement levels: low (L-DLR), medium (M-DLR), and high (H-DLR). Image quality was evaluated using liver noise level (Noise) and lesion signal-to-background ratios (SBR). Percentage changes in these metrics between BPL and each DLR setting were calculated. The four reconstruction methods were compared using the Friedman test with Bonferroni correction. P-values < 0.05 were used to denote statistical significance.

RESULTS

Noise values for BPL, L-DLR, M-DLR, and H-DLR were 0.08, 0.06, 0.06, and 0.08, respectively (P < 0.001), whereas SBR values were 3.75, 3.85, 4.09, and 4.39, respectively (P < 0.001). Compared with BPL, L-DLR and M-DLR significantly reduced Noise by 33.20% (P < 0.001) and 22.21% (P < 0.001), respectively, whereas M-DLR and H-DLR significantly improved SBR by 8.96% (P < 0.001) and 16.79% (P < 0.001), respectively.

CONCLUSIONS

The TOF-enhanced DLR method improves PET image quality metrics compared with the BPL method and has the potential to enhance image quality in 18F-FDG PET/CT for patients with breast cancer.