DeepFoqus (DeepFoqus-Accelerate)

DeepFoqus-Accelerate is an AI-powered software system that reconstructs brain MRI images from accelerated (up to 4x undersampled) scans to produce clinical-quality images comparable to standard unaccelerated scans. It accepts DICOM and HDF5 files from 1.5T or 3T Siemens and GE MRI scanners for T1, T2, and FLAIR sequences in sagittal, axial, or coronal orientations, intended for adult use only by professional MRI/radiology staff.

The primary difference from the predicate device (SwiftMR) is the reconstruction algorithm: DeepFoqus-Accelerate uses ensembles of deep learning models (CNN and U-net architectures) to reconstruct accelerated image data, whereas SwiftMR uses convolutional neural network-based filtering for image enhancement. Both systems follow the same workflow (upload images, apply deep learning processing, export DICOM output) and neither directly interfaces with MRI equipment.

ISO 14971:2019 (risk management), IEC 62366-1:2015+AMD1:2020 (usability), IEC 62304:2006/A1:2016 (software development), NEMA PS 3.1-3.20 2022d (DICOM standard), ISO 15223-1:2021 (labeling), AAMI TIR 57:2016, AAMI TIR 97:2019, and AAMI SW96:2023.

Substantial equivalence is established through functional equivalence despite algorithmic differences. Both devices produce the same essential output (reconstructed/enhanced brain MRI images in DICOM format) for the same intended use. The algorithmic difference was validated through identical performance metrics (SSIM, PSNR, HaarPSI), phantom bench testing, artifact assessment, and clinical radiologist review—all showing comparable performance to the predicate. Direct comparison of major functionality demonstrated largely identical results, supporting that the different reconstruction methodology achieves equivalent clinical outcomes.