Wednesday, November 29 | 9:30 a.m.-9:40 a.m. | W3-SSNR10-1 | Room E353C
In this scientific session, researchers will present study findings that suggest that a deep learning-based fast MRI reconstruction model improves the efficiency and quality of brain MRI exams in both spin-echo and gradient-echo sequences.
The findings could improve patient care and curb healthcare expenditures, wrote a team led by presenter Kyu Sung Choi, MD, PhD, of Seoul National University Hospital in South Korea.
"Enhanced image quality and reduced acquisition time provided by deep learning-based fast reconstruction may facilitate better diagnostics at lower healthcare costs," the group noted.
The investigators sought to evaluate the effectiveness of brain MRI used with a deep learning-based fast reconstructed technique, comparing qualitative and quantitative image quality to conventional MRI for both spin-echo and gradient-echo sequences.
Four neuroradiologists blinded to the type of MRI exam assessed both synthetic and conventional MRI studies with T1-weighted, T2-weighted, T2-FLAIR, and 3D T1-weighted images from 100 patients, using a five-point scale. The researchers calculated interreader agreement among the four clinicians and evaluated signal-to-noise and contrast-to-noise ratio for 3D T1-weighted and T2 FLAIR sequences.
The deep-learning fast reconstruction MRI technique showed a median reduced acquisition time of 41%, as well as improved image quality (3.8 on the rating scale compared with 3.4 for conventional MRI exams, p < 0.001) and structure visualization (3.6 compared with 3.4). Lesion conspicuity and image artifacts were comparable between the two types of exams.
The group also found the following:
- Interreader agreement was moderate to substantial (kappa = 0.74 for structure and lesion delineation; 0.52 for artifacts; and 0.55 for overall quality)
- The fast reconstruction technique increased signal-to-noise and contrast-to-noise ratios compared with conventional MRI, at 82 versus 31.4 and 12.4 versus 4.4, respectively
- The researchers found no significant volumetric analysis differences in 98.2% of regional volumes
"[We found that] deep learning-based fast reconstructed MRI … reduced acquisition time with improvements in overall image quality," the group surmised.
Check out this session for full details!
