Sonic DL

Sonic DL is a deep learning-based MR image reconstruction software feature for GE HealthCare MRI systems (1.5T, 3.0T, and 7.0T). It reconstructs images from highly under-sampled data to enable accelerated acquisitions while maintaining diagnostic quality, applicable to cardiac cine imaging and 3D Cartesian imaging across all anatomies and patient ages.

The proposed Sonic DL extends the predicate device (K223523) by adding a new deep learning convolutional neural network module (Sonic DL 3D) for 3D Cartesian acquisitions using fast spin echo and gradient echo sequences at up to 12x acceleration. The predicate only supported 2D cardiac cine imaging (FIESTA) at up to 12x acceleration on 1.5T and 3.0T systems.

Not stated in this summary. The document references nonclinical testing using digital reference objects, ACR phantoms, and quantitative metrics (PSNR, RMSE, SSIM), plus clinical Likert-score based radiologist evaluations and peer-reviewed publications, but does not cite specific consensus standards like ISO or IEC.

Sonic DL 3D demonstrates substantial equivalence because nonclinical testing shows comparable or superior image quality metrics (PSNR ≥30dB, SSIM ≥0.8) and low-contrast detectability versus conventional parallel imaging at equivalent acceleration; clinical studies confirm diagnostic quality across multiple anatomies, field strengths, and pathologies comparable to ARC+HyperSense; quantitative post-processing analysis shows volumetric measurements within clinically acceptable error margins; and the technology uses the same fundamental deep learning reconstruction approach as the predicate, with no new safety hazards or performance concerns identified relative to standard MR imaging.