Ezra Flash

Ezra Flash is image processing software that enhances MRI images by reducing noise using a convolutional neural network-based algorithm. It processes non-contrast MR images from 1.5T and 3T Siemens and GE scanners for head, abdomen, and pelvis regions in patients over 18 years old, outputting DICOM-compliant enhanced images alongside the originals.

Ezra Flash uses a convolutional neural network-based filtering approach with separate dedicated models for head and body regions, operating on Linux with DICOM-compliant input/output. The primary predicate (K230264) employs identical CNN-based filtering but was limited to 3T scanners and head-only indications; this submission expands to 1.5T and adds abdomen and pelvis regions. Both use cascade filter banks with thresholding and scaling operations optimized through image-guided processes.

ISO 14971:2019 (risk management), IEC 62304 Edition 1.1:2015 (software lifecycle), NEMA PS 3.1-3.20 (2021e) DICOM standard, and AAMI TIR57:2016 (medical device security). No FDA-mandated performance standards apply.

Ezra Flash maintains identical intended use (noise reduction in non-contrast MRI) and employs the same core CNN-based image enhancement technology as the primary predicate K230264. The expansion from 3T-only to both 1.5T and 3T, and from head-only to head, abdomen, and pelvis, represents a broadening of the predicate's scope rather than a fundamental technological change. Performance testing demonstrates SNR improvements ≥5%, CNR improvements >0%, and acceptable perceived image quality via Likert scoring, all consistent with the predicate's established safety and effectiveness profile. The minor algorithmic refinements (head vs. body-specific models) do not raise new safety or efficacy concerns given the identical underlying architecture and demonstrated performance metrics.