Raw data from 400 patients with conventional dose head and neck CTA, coronary CTA, and abdominal CTA were included. Filtered back-projection, iteration, and deep learning reconstruction were performed. To evaluate the impact of deep learning reconstruction on image quality and diagnostic performance in patients with conventional dose CTA.

Raw data from 800 patients with low tube voltage and contrast medium head and neck CTA, coronary CTA, and abdominal CTA were included. Filtered back-projection, iteration, and deep learning reconstruction were performed. To evaluate the impact of deep learning reconstruction on image quality and diagnostic performance in patients with double-low-dose CTA.

Deep learning image reconstruction (DLIR) is a newly developed artificial intelligence noise reduction algorithm in recent years. It trains massive high-quality FBP data sets to learn to distinguish noise and signal, so as to selectively reduce noise and reconstruct high-quality images with low-quality image data.

The specificity and sensitivity were calculated separately for the standard dose group and the double low-dose group using the optimal cutoff value from the receiver operating characteristic curve, for the purpose of comparing diagnostic accuracy between the two groups.

The signal-to-noise ratio was calculated separately for the standard dose group and the double low-dose group using image CT values and noise, to assess the image quality between the two groups.