Frequently asked questions

Product & technology

What is SwiftMR?

SwiftMR is a deep learning-based solution that reduces MRI scanning times. It shortens scan time by up to 50%, enhancing the patient experience and improving workflow efficiency.

A key advantage of SwiftMR is its ability to reduce scanning times without sacrificing image quality. It has been extensively validated in clinical settings and has reconstructed over 7 million scans across 1,500+ institutions worldwide.

How does SwiftMR reduce scanning time without compromising on image quality?

SwiftMR uses a two-step process. First, it modifies the parameters of the MRI equipment to reduce the scanning time. This step shortens the scan, but on its own it produces images of lower quality than the original standard.

To address this, SwiftMR utilizes advanced deep learning technology, enhancing the images to meet — and in published reader studies, on several quality measures even surpass — their original quality. This innovative approach ensures efficiency in scanning time while upholding high standards of image quality.

Which parameters are modified to reduce MRI scanning time?

We can modify nearly all parameters but primarily adjust parameters related to scan speed such as average, parallel imaging, and echo train length, among others. Occasionally, users may request not to change specific parameters, and in those instances, we respect their preferences and do not alter those settings.

If I've recently purchased new MRI equipment that already captures images rapidly, can I still benefit from SwiftMR?

Absolutely! SwiftMR remains beneficial even for modern, fast-imaging MRI equipment. SwiftMR is FDA-cleared to work in conjunction with vendor deep learning reconstruction solutions such as GE AIR Recon DL, and can further accelerate the scan time based on your current parameter settings. We optimize for additional acceleration, making it possible to achieve even faster imaging without compromising image quality.

How fast does SwiftMR process images?

Processing speed depends on network speed and the size of the study, but typically takes under 2 minutes per study. Larger studies like 3D may take longer.

This fits the typical imaging workflow: by the time a patient is moved off the MRI equipment and returns, the processed images are already on the PACS worklist — no disruption to the standard imaging procedure.

Can I try SwiftMR before purchasing it?

Yes, we provide a one-month free trial of SwiftMR. During the demo period, you have the opportunity to provide feedback and assess your satisfaction with the image quality. However, please note that as a policy, we do not offer demo licenses for more than one month. Therefore, it is advisable to schedule the demo when you can thoroughly assess its utility.

Is SwiftMR trained to reflect different MR image characteristics from different vendors?

Yes, SwiftMR was developed with different image formation and reconstruction pipelines from different vendors in mind. If an input image is from a specific vendor, the output image will also retain the unique image characteristics of that vendor.

Does SwiftMR affect ACR accreditation?

As part of ACR accreditation, centers need to submit MR images to be reviewed by radiologists, who assess image quality based on 5 categories (pulse sequence & image contrast, anatomic coverage & imaging planes, spatial & temporal resolution, artifacts, and exam identification). There are no concerns in using SwiftMR-processed images for ACR accreditation.

Performance

How much can SwiftMR reduce the imaging time?

The amount by which imaging time can be reduced with SwiftMR varies slightly depending on the imaging area and imaging technique. However, based on collaborative research conducted with numerous university hospitals, it has been proven that a reduction in imaging time of up to 50% is possible while still allowing for proper interpretation by radiologists. SwiftMR is FDA 510(k)-cleared to support images with scan time reduction by up to 50%.

Is the image quality achieved with SwiftMR suitable for clinical use?

Yes, it is. SwiftMR is designed to restore images that are interpretable even with reduced imaging time. Furthermore, it has been validated for clinical effectiveness and safety through collaborative research with numerous university hospitals, including Seoul National University Hospital and Severance Hospital.

Is it possible for SwiftMR to miss a lesion?

SwiftMR has been trained using pairs of images that display identical anatomy but vary in noise levels. This specialized training enables the SwiftMR algorithm to precisely identify and remove noise patterns while preserving anatomical structures, including lesions.

To verify this aspect, continuous collaborative research with numerous university hospitals is being conducted. So far, all published peer-reviewed SwiftMR studies to date have demonstrated at least "non-inferiority" compared to standard imaging.

How does deep learning technology enhance image quality?

Deep learning involves two key processes: training and inference. During the training phase, the model is 'trained' using pairs of images: one set captured in a short duration with low quality, and another set taken over a longer duration with high quality. The deep learning model learns to convert a low-quality image into a high-quality one. SwiftMR is a model that has been trained with approximately 3 million images from multiple vendors' scanners.

Once the training phase is complete, the model performs 'inference' to transform short-duration, low-quality images into their high-quality, long-duration counterparts. This deep learning process effectively enhances images that may appear coarse or blurred due to shorter scan times. It reduces noise in grainy images and converts low-resolution, blurred images into higher resolutions — not only restoring the original image quality but, in published studies, sometimes surpassing it.

Coverage

Which MRI manufacturers are supported by SwiftMR?

SwiftMR is compatible with all MRI vendors and scanner models, regardless of their age or magnet type, at field strengths from 0.25T to 3.0T — including open and upright MRI systems.

If you're looking to implement SwiftMR in MRI vendors other than GE, Siemens, Philips, Canon (Toshiba), Fujifilm (Hitachi), and Fonar, please reach out to our team!

Does SwiftMR support protocols targeted for specific clinical cases (e.g., vessel wall imaging, brain metastasis imaging, TLE imaging)?

SwiftMR coverage is determined by the sequences that make up each study. For instance, the vessel wall imaging order primarily consists of sequences such as 3D PDWI (SPACE), 3D T1WI (SPACE), CE 3D T1WI (SPACE), and so on, and all of these sequences are supported.

Does SwiftMR support derivative images (e.g., MIP, ADC maps)?

SwiftMR supports certain derivative images like MIP and ADC. In this case, the source image is processed, and the resulting image is sent to the PACS system.

Additionally, it is possible to send the reconstructed image back to the MRI scanner for further processing to generate derivative images.

Does SwiftMR work on images from pediatric patients?

SwiftMR supports images from pediatric patients. The validation dataset submitted for FDA 510(k) clearance includes pediatric data (ages 0–21).

Workflow

Does the SwiftMR-processed image replace the original MR image, or can you get both?

SwiftMR does not replace the original MR image. You can retain both the original MR image and the SwiftMR-processed image if you have sufficient storage capacity. However, it's important to note that the original image will be the accelerated one, since the protocols have been changed for the acceleration process.

Before sending the image to SwiftMR, are you able to tell whether the image has an artifact or needs a rescan?

We're introducing noise by reducing scan time, but not changing any artifact or anatomical information. This is a common and good question we get — yes, you will still be able to see it and discern it on the source images to know if you need to rescan. Please note that we're not adding signal from native images. We are simply removing noise, so naturally the signal will reveal itself.

Installation

What is the installation process like?

The installation is carried out over a total of two days: (1) gateway PC installation and protocol setting will occur on the first day; (2) user training and monitoring will occur on the second day.

Protocol setting must be done on the actual MR console, and the working hours may vary depending on whether the equipment is in operation or not. User training begins along with actual use on patients, and we monitor to ensure there is no problem with the operation.

Additionally, in case of issues, you can contact the responsible customer success manager for immediate response. If the issue persists for an extended period, we recommend scanning using your previous method (original protocols + direct transmission to PACS) until the issue is resolved.

Is there a risk of equipment malfunction due to SwiftMR?

No, there is no such risk. SwiftMR operates within the safe parameters already supported by the equipment, and its time-saving approach does not strain the equipment. SwiftMR optimizes data acquisition, maintaining the quality of images while reducing the operational time. This efficiency can potentially extend the lifespan of the equipment for the same number of images captured.

Additionally, SwiftMR has been developed as a stand-alone system and is incapable of installing any software on or affecting the equipment in any way. The integration that enables image transfer between the equipment and SwiftMR adheres strictly to the existing DICOM image transmission protocol. From the MRI equipment's point of view, SwiftMR looks like a PACS server. There have been no reported cases of equipment malfunction attributable to SwiftMR in the 1,500+ institutions that have implemented it.

Legal, compliance & security

Does SwiftMR transmit patient data externally in any way?

Patient information is not sent outside of the customer's network. For the cloud deployment, Protected Health Information (PHI) is separated from the image data and deidentified. Only the images are transmitted to the cloud server. Any information sent to our cloud server is deleted within 24 hours after processing.

Is SwiftMR compliant with patient data protection laws?

SwiftMR adheres to domestic data protection laws, ensuring that patient information remains confidential and is not exposed externally. Both images and patient-level data are promptly anonymized or deleted after the required service period.

Furthermore, at AIRS Medical headquarters, we handle personal information, image data, and cloud database management in accordance with ISO/IEC 27001:2022 information security standards and patient data protection standards, alongside GDPR, HIPAA, and SOC 2 Type II attestations.

How do you anonymize patient-level data?

When the DICOM images are sent to the gateway PC, the gateway PC removes fields containing patient information (such as name and gender) following the Safe Harbor Method as defined in HIPAA before transmitting the image data to the cloud server.

An encrypted serial number is assigned before transmission to the server. Once the images are reconstructed on the server and sent back to the gateway PC, the patient information is recombined based on the serial number before storing the final MR image to the PACS.

Are there any security concerns when using the cloud server?

There are no concerns about patient data being compromised because SwiftMR servers do not store patient personal information. Additionally, the transmission of anonymized image data to and from the cloud is supported by encryption protocols.

The cloud server is hosted on AWS, which adheres to not only ISO 27001 but also ISO 27017 and ISO 27018 standards. These regulations are internationally recognized as the most stringent for information security and personal data protection related to cloud services.

What responsibility does the company take if there is an issue with the image interpretation when using SwiftMR?

Image interpretation is a unique responsibility and authority of the interpreters, and the company does not take responsibility for this aspect. This is similar to how MRI equipment manufacturers do not assume responsibility for interpretation.

However, if there are situations where the core functionality of SwiftMR — image reconstruction — fails to work due to errors, resulting in the omission of image restoration or exceeding the examination time, we will promptly address and resolve the issue, preferably on the same day. If compensation is required due to such situations, we will engage in appropriate discussions to provide support.

Up to this point, we have not encountered any such issues, but to ensure flexible responses, we have also secured product liability insurance globally.

Competitors

What are the advantages of SwiftMR compared to other solutions?

When compared to competitors, SwiftMR has several strengths that can be summarized as follows:

(1) Vendor-neutral: SwiftMR can be applied to various MRI models without being limited by MRI manufacturers.

(2) Wide coverage range: SwiftMR is FDA-cleared for MR images of all body parts, with broad sequence coverage.

(3) Enhanced image quality: SwiftMR increases image quality by providing denoising and resolution enhancement at the same time.

Specific competitive advantages depending on the type of competitor are as follows:

Versus vendor-specific solutions developed by MRI equipment manufacturers — SwiftMR can be easily applied in hospitals that have more than two brands of MRI equipment, while vendor-specific solutions require separate investments for each brand. SwiftMR can also be applied to all MRI models without the need for hardware and software upgrades.

Versus vendor-neutral solutions applicable to multiple manufacturers — SwiftMR provides noise reduction and sharpness enhancement together across supported sequences, rather than denoising alone, and is FDA-cleared for MR images of all body parts. Coverage is determined by the sequences that make up each study — reach out to our team to review your protocol list.

What sets SwiftMR apart from other solutions in terms of its unique features or advantages?

Existing protocols are optimized with comprehensive consideration of parameters affecting scan time — not just the number of averages.

With the deep learning reconstruction engine for denoising and resolution enhancement always working in unison, higher acceleration can be achieved compared to having to choose between the trade-off of noise and resolution.

About MRI

Why does an MRI scan take a long time?

Acquiring a high-quality MR image takes time. It is technically possible to scan faster, but that usually comes at the cost of image quality — and standards for acceptable image quality vary from one radiology center to another.

Reaching the image quality a facility expects typically requires a longer scan, so a single exam can run anywhere from 15 to 60 minutes. That hurts the patient experience and the operational efficiency of the radiology center.

Are there ways to reduce the duration of an MRI scan?

Yes, reducing the scan duration is achievable by acquiring less information during the imaging process. This can be done by adjusting equipment console parameters such as lowering the 'Average' setting, increasing GRAPPA (or SENSE or CS), or decreasing the matrix size.

All of these methods, however, trade away image quality — the resulting images look coarser or more blurred than standard imaging. Shorter scan times have to be weighed against that compromise.

Product & technology

What is SwiftSight?

SwiftSight is AI-powered brain MRI quantification software. It automatically processes brain MRI scans to produce quantitative volumetric measurements and MS lesion analysis, then generates reports that are returned to PACS. It is designed to support clinicians and patients with objective imaging data.

SwiftSight-Brain is FDA 510(k)-cleared for automatic labelling, visualization, and volumetric quantification of segmentable brain structures and lesions from MR images, with volumetric data compared to reference percentile data.

What does the volumetry analysis actually measure?

SwiftSight segments the brain into anatomical subregions and calculates the volume of each. Those volumes are compared against an age- and sex-matched normative database and expressed as a percentile, giving clinicians an objective reference point for how a region compares to a healthy reference population.

What is "Brain Age" and how is it calculated?

Brain Age is a summary metric designed to make the volumetric analysis easier to interpret. It compares volumes of key regions (whole brain, cerebrum, cerebellum, the four lobes, and hippocampus) against age- and sex-matched references from the normative database, with each region weighted by its effect size — hippocampal volume, for example, carries more weight than occipital volume because it is more strongly associated with neurodegeneration. The full methodology is detailed in a forthcoming white paper.

What advanced regional metrics are available?

Beyond standard lobar volumes, SwiftSight reports more granular structures such as the choroid plexus and detailed regional metrics, supporting research and advanced diagnostic evaluation by the clinician.

What features does SwiftSight provide?

SwiftSight delivers three core outputs:

What reports are currently available?

SwiftSight currently generates the following reports:

What are some future offerings?

Planned expansions include:

Coverage

What does the MS lesion analysis provide?

On FLAIR sequences, SwiftSight quantifies total lesion volume, tracks lesion change over time, identifies newly appearing lesions, and supports longitudinal progression tracking — giving clinicians quantitative measures to inform their own assessment of disease activity and treatment response.

What are the scan parameter requirements?

SwiftSight-Brain requires two sequences:

Can brain volumetry be used to fully diagnose diseases such as dementia or epilepsy?

SwiftSight analyzes and quantifies brain atrophy patterns that may be associated with various neurological disorders. These quantitative measures are intended to support — not replace — the clinician's own diagnosis.

How reliable is the volumetry across different scan parameters and vendors?

When combined with SwiftMR, our solution has shown improved volumetric accuracy, and we are continuously enhancing the algorithm to further reduce variations across different vendors and scan parameters.

Performance

How large is the normative database?

The normative database comprises 32,615 subjects, with broad representation from U.S. and worldwide populations. A white paper detailing the methodology is forthcoming.

Workflow

Does SwiftSight change the technologist's workflow?

No additional manual steps are required. Sites configure a SwiftSight series name once; the system then recognizes the scan automatically, generates the report, and routes it back to PACS.

How fast are results returned?

Typical processing is 3–5 minutes, with parallel processing of multiple studies, so reports return to PACS without holding up the queue.

How does longitudinal comparison work?

SwiftSight matches follow-up scans by PatientID and automatically generates volume-change and progression reports when prior studies are available — no manual setup per patient.

Competitors

How does SwiftSight compare to other solutions?

Compared to other solutions, SwiftSight offers automated series-based report routing (rather than sending images to different destinations), faster processing, parallel study handling, broader protocol tolerance, and patient-friendly reporting.

Is DTI analysis available?

DTI plus volumetry is a planned capability.