Low-dose whole-body CT using deep learning image reconstruction: image quality and lesion detection

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Whole-body computed tomography (CT) is crucial for malignancy surveillance.
• Standard-dose CT (SD-CT) involves significant radiation exposure.
• Low-dose CT (LDCT) aims to reduce radiation but may compromise image quality.

Purpose of the Study:

• To evaluate the image quality and lesion detection performance of LDCT using deep learning image reconstruction (DLIR).
• To compare DLIR-reconstructed LDCT images with standard-dose iterative reconstruction (SD-IR) and low-dose iterative reconstruction (LD-IR).

Main Methods:

• 59 patients undergoing surveillance LDCT and prior SDCT were analyzed.
• LDCT images were reconstructed using hybrid iterative reconstruction (LD-IR) and DLIR (LD-DLIR).
• Radiologists assessed image quality and lesion detection; quantitative parameters (attenuation, noise, SNR) were measured and compared.

Main Results:

• LD-DLIR demonstrated image quality and lesion detection comparable to SD-IR.
• Image quality was significantly better with SD-IR than LD-IR (p < 0.017).
• LD-DLIR showed significantly lower background noise and improved SNR compared to both SD-IR and LD-IR (p < 0.001).
• Radiation dose was significantly reduced by >75% in LD scans (mean CTDIvol 2.9 mGy, DLP 216.2 mGy•cm) versus SD scans (mean CTDIvol 13.5 mGy, DLP 1011.6 mGy•cm) (p < 0.0001).

Conclusions:

• DLIR enables substantial radiation dose reduction (>75%) in whole-body CT.
• DLIR maintains diagnostic image quality and lesion detection capabilities.
• DLIR provides superior signal-to-noise ratio compared to standard-dose iterative reconstruction.