Clinical Scorecard: Utilizing Deep Learning Techniques for Identifying Liver Lesions in Standard and Low-Dose Abdominal CT Scans

At a Glance

Key Highlights

• Low-dose CT imaging increases image noise, complicating detection of low-contrast liver lesions.
• DLR algorithms like TrueFidelity reduce noise comparably to iterative reconstruction (IR) without compromising image texture.
• Clinical evaluation of DLR shows improved subjective image quality but mixed evidence on dose reduction potential for lesion detection.

Guideline-Based Recommendations

Diagnosis

• Use standard-dose CT scans as reference standard for liver lesion detection and characterization.
• Apply DLR and IR reconstruction techniques to improve image quality, especially in low-dose settings.
• Radiologists should review images blinded to dose level and reconstruction method to minimize bias.

Management

• Balance radiation dose and image quality following the ALARA principle to minimize long-term radiation risks.
• Consider high-strength DLR algorithms for enhanced noise reduction in low-dose abdominal CT.
• Maintain weight-based contrast administration protocols to optimize lesion visualization.

Monitoring & Follow-up

• Quantitatively assess image noise using regions of interest in liver parenchyma and paraspinal muscle.
• Monitor radiologist confidence and lesion detection rates across different dose levels and reconstruction methods.
• Follow up with cross-sectional imaging to confirm lesion characterization.

Risks

• Higher radiation doses improve image quality but increase long-term radiation-related risks.
• Iterative reconstruction may alter image noise texture, potentially impairing low-contrast lesion detection.
• DLR algorithms require rigorous clinical validation to ensure diagnostic performance is not compromised.

Patient & Prescribing Data

Patients undergoing abdominal CT for liver lesion evaluation, particularly with known liver metastases from gastrointestinal or pancreatic adenocarcinoma

DLR allows potential dose reduction while maintaining diagnostic image quality, but clinical evidence on improved lesion detection and radiologist confidence remains mixed.

Clinical Best Practices

• Implement DLR algorithms trained on high- and low-dose datasets to enhance image quality in low-dose CT scans.
• Use a standardized protocol for contrast administration tailored to patient body habitus.
• Conduct blinded, randomized image review by experienced radiologists to assess lesion detection and characterization.
• Apply quantitative noise measurements to objectively evaluate image quality across reconstruction methods and dose levels.
• Adhere to ALARA principles balancing diagnostic needs with radiation exposure risks.

References

• European Congress of Radiology 2023 poster on preliminary results
• GE Healthcare TrueFidelity deep learning reconstruction
• ALARA principle in radiology