A new, fast deep learning-based MR imaging technique shows excellent performance for diagnosing knee injuries, researchers have reported.
The under-five-minute, sixfold-accelerated 3-tesla knee MRI exam uses deep learning (DL) for super-resolution image reconstruction, and could improve patient care, wrote a team led by Jan Vosshenrich, MD, of NYU Grossman School of Medicine in New York City. The group’s findings were published April 23 in the American Journal of Roentgenology.
“Modern acceleration techniques can shorten acquisition time without compromising the balance of speed, signal-to-noise ratio, and spatial resolution,” the authors noted.
MRI is the standard of care for the noninvasive diagnosis of a variety of acute and chronic knee abnormalities, and shorter exam times are better for patients, the investigators explained. That’s where deep learning could help. Vosshenrich and colleagues sought to validate a deep-learning MRI technique that uses parallel imaging (PI) and simultaneous multislice (SMS) acceleration to quickly image the knee.
The team conducted a study that included 124 patients who underwent sixfold PI-SMS-accelerated deep-learning super-resolution 3-tesla knee MRI and arthroscopic surgery between October 2022 and July 2023. (The protocol consisted of turbo spin-echo pulse sequences capable of combined PI-SMS acceleration and included axial fat-suppressed T2-weighted, coronal proton density-weighted, coronal fat-suppressed proton density-weighted, sagittal fat-suppressed T2-weighted, and sagittal proton density-weighted images; total acquisition time was 4 minutes 49 seconds.)
MRI scans in an 18-year-old man with right knee pain following a noncontact twisting injury playing basketball. Combined threefold parallel imaging (PIx3) and twofold simultaneous multislice (SMSx2) accelerated deep learning super-resolution turbo spin-echo sequences (A-E) were acquired at 3 tesla. Coronal proton density (PD) fat-suppressed (FS) (A), coronal PD-weighted (B), axial T2-weighted fat-suppressed (C), sagittal T2-weighted fat-suppressed (D), and sagittal PD-weighted (E) MRI scans demonstrate findings suggestive of a pivot-shift mechanism injury with a transchondral fracture of the central lateral femoral condyle (arrow in A) and transchondral bone contusion of the posterolateral tibial plateau, a femoral attachment full-thickness tear of the anterior cruciate ligament (arrows in B and C), joint effusion (asterisk in C), and a nondisplaced vertical-longitudinal tear along the posterior segment of the lateral meniscus (arrows in D and E). Readers correctly diagnosed the findings. An arthroscopic knee surgery photograph (F) demonstrates the lateral meniscus tear (arrow in F). FS = fat-suppressed. PD = proton density. Images and caption courtesy of the AJR.
Seven musculoskeletal radiologists assessed the MRI studies for image quality, presence of artifacts, and structural visibility using the five-point Likert scale. They also identified any cruciate ligament tears, collateral ligament tears, meniscal tears, cartilage defects, or fractures.
The researchers reported the following regarding the technique:
- Overall image quality was good to very good, with a median Likert score of 4.
- Interreader agreement was very good, with kappa correlation of 0.86.
- Image noise was minimal, with a median Likert score of 4.
- Visibility of anatomic structures was very good, with a median Likert score of 5.
- The technique’s performance for diagnosing arthroscopy-validated structural abnormalities was good to excellent, with an area under the receiver operating curve (AUC) equal to or higher than 0.81 and kappa interreader agreement equal to or greater than 0.72.
They also reported the technique’s diagnostic performance across a range of measures and types of injury:
“DL image reconstruction methods enable highly accelerated clinical PI-SMS knee MRI with better image quality than conventional methods and substantially shorter scan times, adding value through maintained diagnostic accuracy,” Vosshenrich and colleagues wrote.
Yet more study regarding this type of MR imaging is needed, according to the authors.
“Future research directions include validating even higher acceleration factors and developing DL super-resolution image reconstruction methods for 3D knee MRI,” they concluded.
The complete study can be found here.
Leave a Reply