Clinical value of ultra-low-dose chest CT with deep learning image reconstruction in the detection and characterization of pulmonary nodules

To evaluate the clinical value of ultra-low-dose CT (ULDCT) with deep learning image reconstruction (DLIR) in the diagnosis of pulmonary nodules.

This was a prospective study that included 115 patients with suspected pulmonary nodules who provided informed consent to participate. All patients were examined by both standard-dose CT (SDCT) and ULDCT. The SDCT images were reconstructed with adaptive statistical iterative reconstruction‑V at 40% (ASIR-V40%; group A), while the ULDCT images were reconstructed with ASIR-V40% (group B) and with high-strength DLIR (DLIR‑H; group C). The radiation dose and the number of detected lung nodules of both scanning modes were recorded. The CT values and noise values (SD) of lung tissue, aorta, and muscle were measured, and the signal-to-noise ratio (SNR) was calculated. Overall image quality and malignant signs of lung nodules were evaluated in a double-blind manner and analyzed using the pathological diagnosis as the gold standard. Measurements were statistically compared between ULDCT and SDCT.

The radiation dose of ULDCT was 0.25 ± 0.08 mSv, that is, a 92.6% reduction compared with the 3.38 ± 0.97 mSv of SDCT (P < 0.001). There was no significant difference in CT values of the lung, aorta, and muscle among the three groups (all P > 0.05). There was no difference in SD, SNR, subjective diagnostic confidence, and malignant signs of pulmonary nodules between groups A and C (P > 0.05), but these parameters were all better than in group B (P < 0.05). The agreement between the two observers on malignant signs was high (kappa = 0.81). The number of pulmonary nodules detected in groups A, B, and C was 247, 238, and 247, respectively.

Image quality obtained using ULDCT with DLIR‑H is comparable to that of SDCT with ASIR-V40%. The ULDCT method enables excellent detection and diagnosis of nodules with significant dose reduction.