PULMONARY NODULE DETECTION IN CHEST CT USING A DEEP LEARNING-BASED RECONSTRUCTION ALGORITHM

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Computed tomography (CT) is crucial for pulmonary nodule detection.
• Image reconstruction algorithms significantly impact nodule detection performance.
• Deep learning image reconstruction (DLIR) offers potential for improved image quality and dose reduction.

Purpose of the Study:

• To evaluate if deep learning image reconstruction (DLIR) techniques are non-inferior to Adaptive Statistical Iterative Reconstruction V (ASIR-V) for pulmonary nodule detection in chest CT.
• To assess DLIR performance across varying radiation doses and reconstruction strengths.

Main Methods:

• A lung phantom with artificial nodules was scanned at multiple dose levels (0.38–7.6 mGy CTDIvol).
• Images were reconstructed using ASIR-V and three DLIR strengths (DL-L, DL-M, DL-H).
• Four radiologists evaluated nodule detection and scoring on 256 image series.

Main Results:

• No statistically significant difference in nodule detection performance was observed between ASIR-V and DLIR algorithms (p=0.987).
• Average area under the curve (AUC) across readers was similar for all reconstruction methods (0.555–0.558).
• Performance remained consistent across different radiation doses and DLIR strengths.

Conclusions:

• Deep learning image reconstruction (DLIR) is non-inferior to ASIR-V for pulmonary nodule detection in chest CT.
• DLIR is a viable alternative to conventional reconstruction methods, especially in low-dose CT protocols.
• The findings support the clinical utility of DLIR for lung nodule detection without compromising diagnostic accuracy.