Impact of a new deep-learning image reconstruction algorithm on potential dose reduction and quality of chest CT...

To evaluate the impact of a new deep-learning image reconstruction (DLR) algorithm on image quality and potential dose reduction compared with a hybrid iterative reconstruction (IR) algorithm under chest CT conditions.

Acquisitions were performed on an image quality phantom at 5 CTDI_vol (0.4/2.5/5.0/7.5/9.5 mGy). Raw data were reconstructed using soft tissue and lung kernels at Level 5 among the 9 levels available on the IR algorithm (IR-5) and the 5 levels on the DLR algorithm (from D1 to D5). The noise power spectrum (NPS) and task-based transfer function (TTF) were computed. The detectability index (d') was computed to model subsolid pulmonary nodule and high-contrast pulmonary nodule for lung images and low-contrast soft tissue mediastinal nodule for soft tissue images.

For both kernels, noise magnitude and average NPS spatial frequencies decreased from D1 to D5 and were lower than those obtained with IR-5. For all inserts studied and both kernels, TTF values at 50% decreased from D1 to D5 and were lower than those obtained with IR-5. For all three simulated lesions and both kernels, d' values increased from D1 to D5 and were higher than those obtained with K5. Compared to IR-5, a dose reduction potential was found for each DLR level.

Compared to IR-5, noise magnitude and detectability values were better, but noise texture and spatial resolution values were degraded, and this degradation increased as the DLR level increased. The results obtained with this new DLR algorithm open up new possibilities for improving image quality and offer significant potential for radiation dose reduction.