Deep learning-based reconstruction of chest ultra-high-resolution computed tomography and quantitative evaluations of smaller airways

Abstract The full-iterative model reconstruction generates ultra-high-resolution computed tomography (U-HRCT) images comprising a 1024 × 1024 matrix and 0.25 mm thickness while suppressing image noises, allowing evaluating small airways 1–2 mm in diameter. However, this technique imposes huge computational burdens and requires a long reconstruction time. This study evaluated whether a recently-established deep learning-based reconstruction, Advanced intelligent Clear-IQ Engine (AiCE), allows quantitative morphological analyses of smaller airways with equal or better quality than the full-iterative model reconstruction while shortening the reconstruction time. In phantom tubes mimicking small airways, the measurement error of 0.5-mm-thickness wall was smaller on the AiCE-based than the full-iterative model-based U-HRCT. Moreover, in five patients with chronic obstructive pulmonary disease, the AiCE-based U-HRCT decreased the reconstruction time approximately by 90% with a modest improvement in image noise, contrast, and sharpness compared to the full-iterative model-based U-HRCT. Therefore, the AiCE-based U-HRCT can be readily used clinically for morphologically evaluating peripheral small airways.