Deep-learning denoising lifts low-field 0.55T knee MRI accuracy to near 3T

Deep-learning denoising raised low-field 0.55T knee MRI sensitivity from 0.83 to 0.97 and diagnostic accuracy from 0.83 to 0.98, approaching conventional 3T (sensitivity 0.95, accuracy 0.97) in a retrospective study of 33 knee MRIs.

• Retrospective paired design: 33 knee MRIs from 26 patients imaged at both 0.55 T and 3 T; four radiologists assessed image quality, landmark visibility, and diagnostic confidence, with consensus reading as reference.
• DL denoising raised sensitivity from 0.83 to 0.97 and accuracy from 0.83 to 0.98; 3 T sensitivity was 0.95 and accuracy 0.97.
• Image quality and diagnostic confidence improved with DL (image quality 2.92 vs 2.55, p=0.005; confidence 8.21 vs 7.66, p<0.001), though 3 T scored highest.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.