Vista OS, Vista AI Scan, RTHawk
K251029Vista AI, Inc. · cleared 2025-08-21 · product code LNH · Radiology
Premarket evidence — what FDA accepted
source quote (p.4)
“Vista OS is an accessory to 1.5T and 3.0T whole-body magnetic resonance diagnostic devices (MRDD). It is intended to operate alongside, and in parallel with, the existing MR console to acquire traditional, real-time and accelerated images.”
source quote (p.7)
“The AI algorithms within the Vista OS system are designed to assist MRI technologists which are always in control of the scan process. The software automates aspects of MRI setup and parameter selection to help reduce exam time, simplify the workflow, and increase reliability. The AI models are employed to perform specific tasks such as: Prescription Localization: The AI provides an initial estimate of image prescriptions, which the technologist can adjust. Cardiac Segmentation: The AI provides measurements of the subject's anatomy to aid in optimizing imaging parameters, which the technologist can modify. Artifact Detection: The AI alerts the technologist to poor image quality resulting from breathing artifacts or irregular rhythm. Cardiac Motion Registration: The AI improves image quality by reducing motion across images when scanned in the cardiac short-axis view. The system uses neural networks for image analysis, with no generative AI employed. These models have a multi-layered architecture that reduces data to the most relevant set for inline image analysis.”
source quote (p.7)
“No clinical decisions are made by the software. The outputs of the AI models are imaging acquisition settings and preliminary analyses useful for assisting the acquisition of images, which are then visually presented to the technologist. The technologist retains the ability to reject or modify the AI's outputs.”
Validation studies (6)
Retrospective clinical
n=120 images
endpoints: 80% agreement between neural-network assessment at its default sensitivity level and the cardiologist reader
standards: IEC 60601-2-33:2022-08 (Ed. 4.0), IEC 60601-1:2020 (Ed. 3.2), MS1-2008, MS3-2008, MS4-2010, MS8-2016, NEMA PS3.1 - 3.20 (2023e), ISO 14971:2019
Retrospective clinical
n=100 cases
endpoints: 80% agreement between neural-network assessment at its default sensitivity level and the cardiologist reader
standards: IEC 60601-2-33:2022-08 (Ed. 4.0), IEC 60601-1:2020 (Ed. 3.2), MS1-2008, MS3-2008, MS4-2010, MS8-2016, NEMA PS3.1 - 3.20 (2023e), ISO 14971:2019
Retrospective clinical
n=209 images
endpoints: denoising should not detract from diagnostic accuracy in all cases; diagnostic quality of the denoised data be judged superior to its paired non-denoised series in more than 80% of test cases
standards: IEC 60601-2-33:2022-08 (Ed. 4.0), IEC 60601-1:2020 (Ed. 3.2), MS1-2008, MS3-2008, MS4-2010, MS8-2016, NEMA PS3.1 - 3.20 (2023e), ISO 14971:2019
Retrospective clinical
n=323 images
endpoints: mean error in plane angulation of less than 3 degrees with standard deviation less than 5 degrees; mean plane position error less than 5 mm with standard deviation less than 15 mm
standards: IEC 60601-2-33:2022-08 (Ed. 4.0), IEC 60601-1:2020 (Ed. 3.2), MS1-2008, MS3-2008, MS4-2010, MS8-2016, NEMA PS3.1 - 3.20 (2023e), ISO 14971:2019
Retrospective clinical
n=329 images
endpoints: mean 3D Intersection-over-Union (IoU) metrics of at least 0.65 for each volumetric scan prescription
standards: IEC 60601-2-33:2022-08 (Ed. 4.0), IEC 60601-1:2020 (Ed. 3.2), MS1-2008, MS3-2008, MS4-2010, MS8-2016, NEMA PS3.1 - 3.20 (2023e), ISO 14971:2019
Retrospective clinical
n=42 other
endpoints: average velocity error should be less than 10% individually for all vessels and views
standards: IEC 60601-2-33:2022-08 (Ed. 4.0), IEC 60601-1:2020 (Ed. 3.2), MS1-2008, MS3-2008, MS4-2010, MS8-2016, NEMA PS3.1 - 3.20 (2023e), ISO 14971:2019
Reported performance (1 observation)
source quote (p.10)
“The primary acceptance criterion for this test was 80% agreement between neural-network assessment at its default sensitivity level and the cardiologist reader.”
Each value carries its own analysis unit and task — never compare or pool across devices. Source: 510(k) summary PDF.
Predicate network
Postmarket — what happened after clearance
- recall_reason_pattern
Software/algorithm-related recall in product code LNH (Philips North America, initiated 2026-04-14): "The potential for stiffness value errors when a specific range of image reconstruction parameters is used in combination with Resoundant's algorithm, leading to the reconstruction " Recalling firm is another firm in the same product code.
first seen 2026-07-08 · recall res_event_number:98779
- recall_reason_pattern
Software/algorithm-related recall in product code LNH (Philips North America, initiated 2025-12-03): "The potential for stiffness value errors when viewing exported MR Elastography (MRE) stiffness maps to viewer Picture Archiving and Communication System (PACS)." Recalling firm is another firm in the same product code.
first seen 2026-07-08 · recall res_event_number:98111
Recall and MAUDE counts are product-code-level (reports aren't reliably attributable to one device). Signals are descriptive observables with sources — never a judgment that the device is unsafe or drifting. Snapshot 2026-07-08.
Reimbursement — how devices like this got paid
Not yet tracked — no payment pathway indexed for this clearance (the reimbursement corpus is a growing seed set).