To investigate the diagnostic performance for low-contrast liver lesion detection using deep learning reconstruction (DLR) and iterative reconstruction (IR) across different dose levels, with a focus on comparing the two methods.
Key Findings:
DLR demonstrated robust noise reduction comparable to IR without compromising image texture. Clinical studies showed improved subjective image quality with DLR, but findings are conflicting regarding the substantial dose reduction potential for low-contrast lesion detection, which may impact clinical decision-making.
Interpretation:
DLR may enhance image quality and maintain diagnostic performance for liver lesion detection at reduced radiation doses, but further evaluation is necessary to confirm its efficacy in clinical practice.
Limitations:
Study limited to participants with known liver metastasis, which may not represent the general population. Potential bias in subjective assessments by radiologists despite blinding. Results may not be generalizable to other types of liver lesions or cancers.
Conclusion:
DLR shows promise for improving liver lesion detection at lower radiation doses, warranting further investigation to validate its clinical utility and potential impact on patient outcomes.
by Tormund H. Njølstad, Kristin Jensen, Hilde K. Andersen, Audun E. Berstad, Gaute Hagen, Cathrine K. Johansen, Kjetil Øye, Jan Glittum, Anniken Dybwad, Emma Thingstad, Marianne G. Guren, Johann Baptist Dormagen, Anselm Schulz
Radiologists assigned to receive step-by-step explanations from a large language model achieved higher diagnostic accuracy in a randomized vignette study, while differential-diagnosis outputs may have increased inappropriate reliance on incorrect model suggestions.
