Deep Learning Reconstruction Enhances Liver Lesion Detection in Low-Dose CT

Overview

This study evaluated deep learning reconstruction (DLR) versus iterative reconstruction (IR) for detecting low-contrast liver lesions across standard, medium, and low-dose abdominal CT scans. DLR demonstrated robust noise reduction and improved lesion detectability at reduced radiation doses without compromising image texture, suggesting potential for dose reduction in clinical practice.

Background

Detecting liver lesions on abdominal CT is challenging, especially at low radiation doses due to increased image noise. While iterative reconstruction (IR) reduces noise, it can alter image texture and impair low-contrast lesion detection. Deep learning reconstruction (DLR) algorithms have emerged to enhance image quality by reducing noise while preserving texture, but clinical evidence on their impact on lesion detection and dose reduction remains mixed. This study investigates the diagnostic performance of DLR compared to IR across different dose levels in patients with known liver metastases.

Data Highlights

Key Findings

• DLR algorithms like TrueFidelity reduce image noise comparably or better than IR across dose levels without compromising image texture.
• DLR improves low-contrast lesion detectability in phantom studies and shows promise in clinical settings.
• Subjective image quality is enhanced with DLR, but evidence on dose reduction potential for lesion detection is mixed.
• This study prospectively included patients with known liver metastases and used blinded radiologist readings to assess lesion detection across standard, medium, and low-dose scans.
• DLR at high strength maintained or improved lesion detection rates at reduced radiation doses compared to IR.

Clinical Implications

DLR can be integrated into abdominal CT protocols to reduce radiation dose while maintaining diagnostic confidence in detecting liver lesions. This supports adherence to the ALARA principle by balancing image quality and radiation exposure. Radiologists should consider DLR as a tool to optimize low-contrast lesion detection, especially in oncological follow-up imaging.

Conclusion

Deep learning reconstruction offers a promising approach to reduce radiation dose in abdominal CT without sacrificing liver lesion detectability. Further clinical validation will solidify its role in routine imaging protocols.

References

1. Jensen et al. 2023 -- Utilizing Deep Learning Techniques for Identifying Liver Lesions in Standard and Low-Dose Abdominal CT Scans