ARRS 2022 Abstracts

RETURN TO ABSTRACT LISTING


E1531. Deep Learning Technology Subtle Imaging Matches Conventional Knee MR Image with Reduced Acquisition Time
Authors
  1. Rola Husain; Mount Sinai Hospital
  2. Stephen Belmustakov; Mount Sinai Hospital
  3. Etan Dayan; Mount Sinai Hospital
  4. Ajit Shankaranarayanan; Subtle Medical Inc.
  5. Yuan Xie; Subtle Medical Inc.
  6. Mingqian Huang; Mount Sinai Hospital
  7. Idoia Corcuera-Solano; Mount Sinai Hospital
Objective:
Knee MRI is one of the most performed MR examinations in the outpatient setting and has a crucial role in detecting pathology. To increase access of advanced imaging for all patients and to provide valuable diagnostic information to the clinician, reducing imaging acquisition time while preserving good image quality is crucial. The deep learning (DL) subtle MRI (s-MRI) technique may be an alternative to conventional MRI, with reduced scanning time while preserving image quality and diagnostic value. The purpose of this study was to compare DL s-MRI software to conventional MRI (c-MRI) acquisition for the knee.

Materials and Methods:
In this prospective study, patients who underwent 1.5 T MRI of the knee from November–December 2020 in our institution were included. Knee MR images were obtained using our conventional 5-sequence protocol and equivalent subtle axial PD FS, sagittal T2 FS, coronal T2 FS, and coronal PD . Images were evaluated in an independent blinded fashion by two musculoskeletal radiologists for image quality, diagnostic quality, and artifacts, with scores ranging from 1–3. Data were compared using paired Wilcoxon tests.

Results:
A total of 50 consecutive patients were evaluated (26 men and 24 women) with a mean age of 43.4 ± 19.5 years (range, 13–78 years). The average acquisition time saving with s-MRI was 6 minutes and 42 seconds per patient (50.2% reduction). Overall diagnostic performance between c-MRI and s-MRI was comparable with all the sequences (p > 0.05). Image quality was comparable in all sequences except axial PD FS (p < 0.002) and coronal T2 for reader 1 (p = 0.004), where c-MRI performed better. No differences in artifacts were found between c-MRI and s-MRI, except for axial PD FS images (p < 0.004), where c-MRI performed better. Nevertheless, all subtle images were rated with a score above 2 in all sequences, meaning image quality of these structures were “as expected” by a musculoskeletal radiologist.

Conclusion:
DL s-MRI provides similar image quality and diagnostic value when compared to c-MRI for the study of the knee with more than 50% reduction in acquisition time, which can potentially increase efficiency while improving patient’s comfort.