SubtleHD is an image processing software that can be used for image enhancement of all body parts MRI images. It can be used for noise reduction and increasing image sharpness.

Device Description

SubtleHD is Software as a Medical Device (SaMD) consisting of a software algorithm that enhances images taken by MRI scanners. As it only processes images for the device has no user interface. It is intended to be used by radiologists and technologists in an imaging center, clinic, or hospital. The SubtleHD software can be used with MR images acquired as part of standard of care and accelerated MRI exams as the input. The outputs are the corresponding images with enhanced image quality. Original DICOM images are passed onto the SubtleHD software as an input argument and the enhanced images are saved in the designated location prescribed when running the SubtleHD software. The functionality of SubtleHD (noise reduction and sharpness enhancement) is identified from the DICOM series description and/or through configuration is specified as configuration files and OS environment variables.

SubtleHD software implements an image enhancement algorithm using a convolutional network based filtering. Original images are enhanced by running through a cascade of filter banks, where thresholding and scaling operations are applied. A single neural network is trained for adaptive noise reduction and sharpness increase. The parameters within the neural network were obtained through an image-guided optimization process. Additional nonlocal mean based denoising and unsharp masking based sharpening filters are applied to the deep learning processed image.

The software operates on DICOM files, enhances the images, and sends the enhanced images to any desired destination with an AE Title (e.g., PACS, MR device, workstation, and more). Enhanced images coexist with the original images.

AI/ML Overview

The provided text describes the acceptance criteria and study proving the device, SubtleHD (1.x), meets these criteria.

Here's the breakdown of the requested information:

1. Table of Acceptance Criteria and Reported Device Performance

The provided text includes two distinct sets of acceptance criteria and performance results: one for "Performance Validation" and one for a "Reader Study."

Performance Validation Summary:

Endpoint	Acceptance Criteria	SubtleHD Mode	Result	Conclusion
Denoising (SNR) - Primary	SNR shall improve by at least 40% in homogeneous ROI regions for at least 90% of the dataset.	Default	PASS	SubtleHD performs denoising, in terms of improved SNR, MRI images.
	SNR shall improve by at least 40% in homogeneous ROI regions for at least 95% of the dataset.	High Denoising	PASS
Sharpness (Image Intensity Change) - Primary	Slope in a line ROI is increased for at least 90% of the dataset.	Default	PASS	SubtleHD sharpens, in terms of improvement in visibility of the edge at a tissue interface by image intensity slope measure, MRI images.
	Slope in a line ROI is increased for at least 95% of the dataset.	High Sharpening	PASS
Sharpness (Image Intensity Change for Brains) - Secondary	Thickness, in terms of FWHM in a line ROI, is reduced for at least 90% of the dataset.	Default	PASS	SubtleHD sharpens, in terms of improvement in visibility of an anatomical structure by image intensity FWHM measure, MRI images.
	Thickness, in terms of FWHM in a line ROI, is reduced for at least 95% of the dataset.	High Sharpening	PASS
Sharpness and Over Smoothing (Gradient Entropy) - Primary	At least 90% of cases demonstrate a lower gradient entropy value after SubtleHD processing.	Default	PASS	SubtleHD does not result in over-smoothed images, in terms of improvement in gradient entropy.
	At least 95% of cases demonstrate a lower gradient entropy value after SubtleHD processing.	High Sharpening	PASS
	There is a statistically significant improvement in gradient entropy when comparing the original and SubtleHD enhanced images across the performance dataset per a two-sided paired t-test.	Default and High Sharpening	PASS

Reader Study Summary:

Endpoint	Endpoint Description	Acceptance Criteria	Result	Conclusion
Denoising (Primary)	Signal-to-Noise Ratio	Statistically significantly better with p-value < 0.05 or not statistically significantly different in a Wilcoxon signed rank test	PASS	SubtleHD performs denoising, in terms of improved SNR, for MRI Images.
	Overall Image Quality / Diagnostic Confidence	Statistically significantly better with p-value < 0.05 or not statistically significantly different in a Wilcoxon signed rank test	PASS
Sharpness (Primary)	Visibility of Small Structures	Statistically significantly better with p-value < 0.05 or not statistically significantly different in a Wilcoxon signed rank test	PASS	SubtleHD performs sharpening and does not over-smooth images, in terms of improved visibility of small structures, for MRI images.
Artifacts (Secondary)	Artifact Introduction	SubtleHD-enhanced images do not contain artifacts that could impact diagnosis or b) both input and SubtleHD-enhanced images are deemed to contain artifacts that could impact diagnosis.	PASS	SubtleHD does not introduce artifacts into MRI images.

2. Sample sizes used for the test set and the data provenance

Standalone Image Quality Metric Testing:

Paired Test Set (Unaligned SOC as Reference): 97 samples
Paired Test Set (Unaligned SubtleHD-enhanced SOC as the Reference): 97 samples
Aligned Test Set: 471 samples

Data Provenance for Standalone Image Quality Metric Testing:

Countries of Origin: The aligned test set included data from both the US and OUS (Outside US). The paired test set's provenance isn't explicitly stated beyond "More than 50% of data is from sites in the United States."
Retrospective or Prospective: The text states "retrospective clinical data" for the performance validation and reader study, and describes the data selection for the standalone metrics as using "a subset of our performance validation set," implying a retrospective nature for these as well.

Performance Validation & Reader Study Dataset Characteristics:

Number of Image Series: 410 image series (205 input and 205 SubtleHD enhanced) for the Reader Study. The Performance Validation used the "SubtleHD performance validation test dataset," which is implied to be distinct from, but shares characteristics with, the reader study dataset.
Retrospective or Prospective: Retrospective clinical data.
Countries of Origin: The majority of performance data comes from sources in the United States (65%).

3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts

Performance Validation:

Ground Truth Establishment: Three Regions of Interest (ROIs) were drawn on each image by a Subtle Medical employee with an MD and/or PhD in a clinically-relevant field.
Quality Review: A board-certified radiologist reviewed the ROI for acceptability. The exact number of radiologists is not specified, only "a board-certified radiologist."

Reader Study:

Readers: The study involved "board-certified radiologists." The exact number of radiologists involved in the reading is not specified. Readers were blind to the image processing method.

4. Adjudication method (e.g. 2+1, 3+1, none) for the test set

Performance Validation: Not applicable in the traditional sense, as ROIs were drawn by one expert and reviewed for acceptability by another. It's not a consensus reading model for diagnostic outcomes.
Reader Study: The text does not specify an adjudication method for the reader study. It states that "Both the input images and the SubtleHD enhanced images were ranked by board-certified radiologists," implying individual readings without explicit mention of multiple readers reaching consensus or a tie-breaking mechanism.

5. If a multi-reader multi-case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance

MRMC Study: A "Reader Study" was conducted, which involved multiple readers (board-certified radiologists) and multiple cases (410 image series). This qualifies as a multi-reader multi-case study.
Comparative Effectiveness / Effect Size: The study assessed if SubtleHD-enhanced images statistically significantly improved perceived SNR, Overall Image Quality / Diagnostic Confidence, and Small Structure Visibility, and did not introduce artifacts. The results indicate a "PASS" for all these criteria based on a p-value < 0.05 from a Wilcoxon signed rank test, meaning there was a statistically significant improvement or no statistically significant difference for the positive criteria, and no artifact introduction. While it indicates an improvement in perceived image quality, the document does not report specific effect sizes (e.g., AUC uplift, specific improvement percentage in diagnostic accuracy) for how much human readers improve with AI assistance versus without. It focuses on the qualitative assessment of improvement in image characteristics and diagnostic confidence.

6. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done

Yes, a standalone performance evaluation was done through:

Standalone Image Quality Metric Testing: This section reports quantitative metrics (L1 loss, SSIM, PSNR) for the algorithm's output compared to reference images. This is purely algorithm-based performance.
Performance Validation: This also appears to be a standalone measurement of the algorithm's performance on a dataset, using quantitative metrics like SNR improvement, slope of image intensity change, FWHM reduction, and gradient entropy, without human readers directly influencing the primary outcome measurements.

Both of these sections describe the algorithm's objective performance without a human in the loop to modify or interact with the output for the purpose of the reported metrics.

7. The type of ground truth used (expert consensus, pathology, outcomes data, etc.)

Standalone Image Quality Metric Testing:
- For the Paired Test Set (Unaligned SOC as Reference), the "standard-of-care (SOC) images" were used as reference (ground truth). Also, "SubtleHD-enhanced SOC" was used as another reference.
- For the Aligned Test Set, "high-quality ground truth images were reconstructed using the vendor's commercially available deep learning pipeline and represent standard-of-care image quality."
Performance Validation: The ground truth for quantitative metrics (SNR, slope, FWHM, gradient entropy) was derived from characteristics within the images themselves, with ROIs drawn by a clinically-qualified employee and reviewed by a board-certified radiologist for acceptability. This is a form of expert-defined quantitative ground truth.
Reader Study: The ground truth for this study was the perception of board-certified radiologists using a Likert scale for SNR, Overall Image Quality / Diagnostic Confidence, Small Structure Visibility, and Imaging Artifacts. This can be considered a form of expert consensus/opinion-based ground truth or perceived improvement, although not explicitly stated as a consensus amongst multiple readers for a single case.

8. The sample size for the training set

The document explicitly states that the test sets for all evaluations (Standalone Image Quality Metric Testing, Performance Validation, and Reader Study) were "Not used for algorithm training" and that "data was selected from sources not included in the training dataset (36.50% of the dataset is from non-training sources)." However, the actual sample size of the training set is not provided in the given text.

9. How the ground truth for the training set was established

The document states that a "single neural network is trained for adaptive noise reduction and sharpness increase" and "The parameters within the neural network were obtained through an image-guided optimization process." It also implies "pre- and post-processing is applied to configure desired perceived image quality." However, the specific method for how ground truth was established for the training set is not detailed in the provided text. It is generally understood that for deep learning models like this, the "ground truth" for training purposes would be pairs of original (possibly noisy/blurry) images and corresponding "ideal" or "high-quality" images that the model learns to transform inputs into. The text indicates that some of the performance data came from training sources.

Summary

{0}------------------------------------------------

Image /page/0/Picture/0 description: The image shows the logo of the U.S. Food and Drug Administration (FDA). On the left is the Department of Health & Human Services logo. To the right of that is the FDA logo, which consists of the letters "FDA" in a blue square, followed by the words "U.S. FOOD & DRUG" in blue, and then the word "ADMINISTRATION" in a smaller font in blue.

February 12, 2025

Subtle Medical, Inc. % Jared Seehafer Regulatory Consultant Enzyme Corporation 611 Gateway Blvd Ste 120 South San Francisco. California 94080

Re: K243250

Trade/Device Name: SubtleHD (1.x) Regulation Number: 21 CFR 892.2050 Regulation Name: Medical Image Management And Processing System Regulatory Class: Class II Product Code: OIH Dated: January 13, 2025 Received: January 13, 2025

Dear Jared Seehafer:

We have reviewed your section 510(k) premarket notification of intent to market the device referenced above and have determined the device is substantially equivalent (for the indications for use stated in the enclosure) to legally marketed predicate devices marketed in interstate commerce prior to May 28, 1976, the enactment date of the Medical Device Amendments, or to devices that have been reclassified in accordance with the provisions of the Federal Food, Drug, and Cosmetic Act (the Act) that do not require approval of a premarket approval application (PMA). You may, therefore, market the device, subject to the general controls provisions of the Act. Although this letter refers to your product as a device, please be aware that some cleared products may instead be combination products. The 510(k) Premarket Notification Database available at https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm identifies combination product submissions. The general controls provisions of the Act include requirements for annual registration, listing of devices, good manufacturing practice, labeling, and prohibitions against misbranding and adulteration. Please note: CDRH does not evaluate information related to contract liability warranties. We remind you, however, that device labeling must be truthful and not misleading.

If your device is classified (see above) into either class II (Special Controls) or class III (PMA), it may be subject to additional controls. Existing major regulations affecting your device can be found in the Code of Federal Regulations, Title 21, Parts 800 to 898. In addition, FDA may publish further announcements concerning your device in the Federal Register.

{1}------------------------------------------------

FDA's substantial equivalence determination also included the review and clearance of your Predetermined Change Control Plan (PCCP). Under section 515C(b)(1) of the Act, a new premarket notification is not required for a change to a device cleared under section 510(k) of the Act, if such change is consistent with an established PCCP granted pursuant to section 515C(b)(2) of the Act. Under 21 CFR 807.81(a)(3), a new premarket notification is required if there is a major change or modification in the intended use of a device. or if there is a change or modification in a device that could significantly affect the safety or effectiveness of the device, e.g., a significant change or modification in design, material, chemical composition, energy source, or manufacturing process. Accordingly, if deviations from the established PCCP result in a major change or modification in the intended use of the device, or result in a change or modification in the device that could significantly affect the safety or effectiveness of the a new premarket notification would be required consistent with section 515C(b)(1) of the Act and 21 CFR 807.81(a)(3). Failure to submit such a premarket submission would constitute adulteration and misbranding under sections 501(f)(1)(B) and 502(o) of the Act, respectively.

Additional information about changes that may require a new premarket notification are provided in the FDA guidance documents entitled "Deciding When to Submit a 510(k) for a Change to an Existing Device" (https://www.fda.gov/media/99812/download) and "Deciding When to Submit a 510(k) for a Software Change to an Existing Device" (https://www.fda.gov/media/99785/download).

Your device is also subject to, among other requirements, the Quality System (QS) regulation (21 CFR Part 820), which includes, but is not limited to, 21 CFR 820.30, Design controls; 21 CFR 820.90, Nonconforming product; and 21 CFR 820.100. Corrective and preventive action. Please note that regardless of whether a change requires premarket review, the QS regulation requires device manufacturers to review and approve changes to device design and production (21 CFR 820.30 and 21 CFR 820.70) and document changes and approvals in the device master record (21 CFR 820.181).

Please be advised that FDA's issuance of a substantial equivalence determination does not mean that FDA has made a determination that your device complies with other requirements of the Act or any Federal statutes and regulations administered by other Federal agencies. You must comply with all the Act's requirements, including, but not limited to: registration and listing (21 CFR Part 807); labeling (21 CFR Part 801); medical device reporting of medical device-related adverse events) (21 CFR Part 803) for devices or postmarketing safety reporting (21 CFR Part 4, Subpart B) for combination products (see https://www.fda.gov/combination-products/guidance-regulatory-information/postmarketing-safety-reportingcombination-products); good manufacturing practice requirements as set forth in the quality systems (QS) regulation (21 CFR Part 820) for devices or current good manufacturing practices (21 CFR Part 4, Subpart A) for combination products; and, if applicable, the electronic product radiation control provisions (Sections 531-542 of the Act); 21 CFR Parts 1000-1050.

All medical devices, including Class I and unclassified devices and combination product device constituent parts are required to be in compliance with the final Unique Device Identification System rule ("UDI Rule"). The UDI Rule requires, among other things, that a device bear a unique device identifier (UDI) on its label and package (21 CFR 801.20(a)) unless an exception or alternative applies (21 CFR 801.20(b)) and that the dates on the device label be formatted in accordance with 21 CFR 801.18. The UDI Rule (21 CFR 830.300(a) and 830.320(b)) also requires that certain information be submitted to the Global Unique Device Identification Database (GUDID) (21 CFR Part 830 Subpart E). For additional information on these

{2}------------------------------------------------

requirements, please see the UDI System webpage at https://www.fda.gov/medical-device-advicecomprehensive-regulatory-assistance/unique-device-identification-system.

Also, please note the regulation entitled, "Misbranding by reference to premarket notification" (21 CFR 807.97). For questions regarding the reporting of adverse events under the MDR regulation (21 CFR Part 803), please go to https://www.fda.gov/medical-device-safety/medical-device-reportingmdr-how-report-medical-device-problems.

For comprehensive regulatory information about medical devices and radiation-emitting products, including information about labeling regulations, please see Device Advice (https://www.fda.gov/medicaldevices/device-advice-comprehensive-regulatory-assistance) and CDRH Learn (https://www.fda.gov/training-and-continuing-education/cdrh-learn). Additionally, you may contact the Division of Industry and Consumer Education (DICE) to ask a question about a specific regulatory topic. See the DICE website (https://www.fda.gov/medical-device-advice-comprehensive-regulatoryassistance/contact-us-division-industry-and-consumer-education-dice) for more information or contact DICE by email (DICE@fda.hhs.gov) or phone (1-800-638-2041 or 301-796-7100).

Sincerely,

Daniel M. Krainak, Ph.D. Assistant Director Magnetic Resonance and Nuclear Medicine Team DHT8C: Division of Radiological Imaging and Radiation Therapy Devices OHT8: Office of Radiological Health Office of Product Evaluation and Quality Center for Devices and Radiological Health

Enclosure

{3}------------------------------------------------

Indications for Use

Form Approved: OMB No. 0910-0120 Expiration Date: 07/31/2026 See PRA Statement below.

Submission Number (if known)

K243250

Device Name

SubtleHD (1.x)

Indications for Use (Describe)

SubtleHD is an image processing software that can be used for image enhancement of all body parts MRI images. It can be used for noise reduction and increasing image sharpness.

Type of Use (Select one or both, as applicable)

Prescription Use (Part 21 CFR 801 Subpart D)

Over-The-Counter Use (21 CFR 801 Subpart C)

CONTINUE ON A SEPARATE PAGE IF NEEDED.

This section applies only to requirements of the Paperwork Reduction Act of 1995.

DO NOT SEND YOUR COMPLETED FORM TO THE PRA STAFF EMAIL ADDRESS BELOW.

The burden time for this collection of information is estimated to average 79 hours per response, including the time to review instructions, search existing data sources, gather and maintain the data needed and complete and review the collection of information. Send comments regarding this burden estimate or any other aspect of this information collection, including suggestions for reducing this burden, to:

Department of Health and Human Services Food and Drug Administration Office of Chief Information Officer Paperwork Reduction Act (PRA) Staff PRAStaff(@fda.hhs.gov

"An agency may not conduct or sponsor, and a person is not required to respond to, a collection of information unless it displays a currently valid OMB number."

{4}------------------------------------------------

SubtleHD 510(k) Summary

Date Summary Prepared:	2025-02-10
Contact Details:
Applicant Name:	Subtle Medical, Inc.
Applicant Address:	883 Santa Cruz Ave, Suite 205Menlo Park, CA 94025 United States
Applicant Contact:	Ms. Ronny Elor
Applicant Contact Telephone:	(925) 324-8467
Applicant Contact Email:	ronny@subtlemedical.com
Correspondent Name:	Enzyme Corporation
Correspondent Address:	611 Gateway Blvd, Ste 120South San Francisco, CA 94080 United States
Correspondent Contact:	Mr. Jared Seehafer
Correspondent Contact Telephone:	(415) 638-9554
Correspondent Contact Email:	jared@enzyme.com
Device Name:
Device Trade Name:	SubtleHD (1.x)
Common Name:	Medical image management and processing system
Classification Name:	System, Image Processing, Radiological
Regulation Number:	892.2050
Product Code:	QIH
Device Class:	Class II
Legally Marketed Predicate Devices:	Primary Predicate #: K230854Predicate Trade Name: SwiftMRPredicate Manufacturer: AIRS Medical Inc.
	Secondary Predicate #: K223623Predicate Trade Name: SubtleMRPredicate Manufacturer: Subtle Medical, Inc.

Table 1. Contact Details and Device Name

{5}------------------------------------------------

Device Description Summary

Intended Use / Indications for Use

SubtleHD is an image processing software that can be used for image enhancement of all body parts MRI images. It can be used for noise reduction and increasing image sharpness.

Intended Use / Indications for Use Comparison

SubtleHD and its predicates are all intended to denoise and sharpen MRI images of various input protocols, imaging scanners, and anatomical regions.

Technological Comparison

SubtleHD and its predicates are all used for image enhancement. They operate on DICOM files, enhance the images, and send the enhanced images to any desired destination. The receipt of original DICOM image files and delivery of enhanced images as DICOM files depends on other software systems. Both subject and predicate devices use convolutional network based filtering. Original images are enhanced by running through a cascade of filter banks, where thresholding and scaling operations are applied. The software performs noise reduction and sharpness increase. The parameters within the software are obtained through an image-quided optimization process. Additional pre- and post-processing is applied to configure desired perceived image quality.

The following table provides a detailed description of the technological characteristics of subject and predicate devices.

{6}------------------------------------------------

Comparison	SubtleHD(Subject Device)	SwiftMR(PrimaryPredicateDevice)(K230854)	SubtleMR(SecondaryPredicateDevice)(K223623)	NotedDifferences
Workflow	The software operateson DICOM files,enhances the images,and sends theenhanced images toany desired destinationwith an AE Title (e.g.,PACS, MR device,workstation, andmore). Enhancedimages coexist withthe original images.	The software operateson DICOM files,enhances the images,and stores theenhanced images onPACS or on a MRdevice. Enhancedimages coexist withthe original images.	The software operateson DICOM files on thefile system, enhancesthe images, and storesthe enhanced imageson the file system. Thereceipt of originalDICOM image filesand delivery ofenhanced images asDICOM files dependson other softwaresystems.	SubtleHD can sendenhanced images toany destination as longas it is associated withan IP and AE Title.SubtleMR by default isintended to sendenhanced images toPACS, but can beconfigured to otherdestinations. SwiftMRcan send enhancedimages to PACS or anMR Device.SubstantiallyEquivalent.
Product Code	QIH	LLZ	LLZ	SubstantiallyEquivalent.
PhysicalCharacteristics	Software packageOperates on a virtualmachine	Same	Same	Same
Intended User	Radiologists	Same	Same	Same
Intended Location	Medical facility(hospitals, clinics,imaging center, etc.)	Same	Same	Same
Modalities	MRI	Same	Same	Same
Operating System /Computer	Linux Compatible; PCor Mac	PC Compatible	Same	SubtleHD andSubtleMR are PC orMac, while SwiftMR isonly PC. SubstantiallyEquivalent.
Rx or OTC	Rx	Same	Same	Same
User Interface	None	Same	Same	Same
DICOMStandard Compliance	The softwareprocessesDICOM-compliantimage data.	Same	Same	Same
Comparison	SubtleHD(Subject Device)	SwiftMR(PrimaryPredicateDevice)(K230854)	SubtleMR(SecondaryPredicateDevice)(K223623)	NotedDifferences
Image EnhancementAlgorithm Description	SubtleHD softwareimplements an imageenhancementalgorithm using aconvolutional neuralnetwork based filtering.Original images areenhanced by runningthrough a cascade offilter banks, wherethresholding andscaling operations areapplied. A singleneural network istrained for adaptivenoise reduction andsharpness increase.The parameters withinthe neural networkwere obtained throughan image-guidedoptimization process.Additional nonlocalmean based denoisingand unsharp maskingbased sharpeningfilters are applied tothe deep learningprocessed image.	SwiftMR implementsan imageenhancementalgorithm usingconvolutional neuralnetwork-basedfiltering. Originalimages are enhancedby running through acascade of filter banks,where thresholdingand scaling operationsare applied. Neuralnetwork-based filtersthat perform noisereduction and/orsharpening areobtained. Theparameters of thefilters were obtainedthrough animage-guidedoptimization process.Sharpening filter isadditionally applied tothe deep learningprocessed image.	SubtleMR softwareimplements an imageenhancementalgorithm usingconvolutional neuralnetwork based filtering.Original images areenhanced by runningthrough a cascade offilter banks, wherethresholding andscaling operations areapplied. Separateneural network basedfilters are obtained fornoise reduction andsharpness increase.The parameters of thefilters were obtainedthrough animage-guidedoptimization process.	SubstantiallyEquivalent.
Model Architecture	Single SRE/DNEmodel with filters/pre/post-processing.	DNE is a model; SREis achieved via multiplefilters	SRE is a model, DNEis a model, with filters/pre/post-processing	Deep-learningalgorithm andprocessing steps fornoise reduction andsharpnessenhancement.SubstantiallyEquivalent.
Function	Combined SRE andDNE for all anatomies.	Combined SRE andDNE, can turn eitheroff as an option.	Separate, DNE head,spine, neck, abdomen,pelvis, prostate,breast, andmusculoskeletal, SREhead only.	SubstantiallyEquivalent.
Enhancement levels	Optional highdenoising level andoptional highsharpening level.	Denoising level fromlevel 0 to level 8,Sharpness level fromlevel 0 to level 5.	No levels (single level).	SubtleHD and SwiftMRare substantiallyequivalent. SubtleMRhas a single SRE andDNE option.
Performance Validation	Endpoints andacceptance criteriausing retrospectiveclinical images for bothnoise reduction andsharpness increasefunctions.	Same	Same	Same

Table 2. Comparison of Technological Characteristics

{7}------------------------------------------------

Subtle Medical Inc. - 510(k) – SubtleHD – 510(k) Summary

{8}------------------------------------------------

Predetermined Change Control Plan (PCCP)

SubtleHD has a Predetermined Change Control Plan (PCCP), which details planned device modifications, the associated methodology to develop, validate, and implement those modifications, and an assessment of the impact of those modifications. In general, Subtle Medical utilizes PCCPs for planned modifications that aim to improve customer satisfaction with respect to perceived image quality, generalizability, and flexibility of the product.

The SubtleHD PCCP contains one planned modification: to add more denoising and sharpening levels as a configuration option. This modification is intended to provide additional flexibility for the user. Prior to release, the software verification and performance validation described in this summary shall be re-executed with the same endpoints and validation dataset as those used to support this submission, with the acceptance criteria for the new configurable levels of denoising and sharpness being adjusted to have increasing stringency. Re-executing the software verification and validation and performance validation will ensure substantial equivalence is maintained following the modification. The algorithm will be locked with fixed model parameters prior to release.

This modification shall be performed as a part of Subtle Medical's Design Change Control process in a Quality Management System that is ISO 13485:2016 and MDSAP certified. It shall be accompanied by a software version-specific Customer Release Notes describing the change along with software version-specific User Manual. The updated labeling shall be delivered to users by Subtle Medical customer success personnel.

Non-Clinical and/or Clinical Tests Summary & Conclusions

Subtle Medical conducted the following performance testing:

Software Verification and Validation testing (unit, integration, and system testing) to demonstrate ● that software requirements are implemented. These tests passed.
. Standalone image quality metric testing to report the Least Absolute Deviations (L1 loss), Structural Similarity (SSIM), and peak signal-to-noise ratio (PSNR) values. This testing demonstrated significant reduction in L1 loss and significant increase in SSIM and PSNR.
. Performance Validation testing utilizing retrospective clinical data to demonstrate the software enhanced image quality in MR images via a reduction of noise or sharpness enhancement. These tests passed.
. A Reader Study utilizing retrospective clinical data to demonstrate the software enhanced image quality in MR images via a reduction of noise or sharpness enhancement. These tests passed.

Standalone Image Quality Metric Testing:

A subset of our performance validation set, which is detailed below, where standard-of-care (SOC) images were acquired, was used to create a paired test set for evaluating L1 loss, SSIM, and PSNR. To enhance the spatial alignment between the two separately acquired images, affine registration was applied to the accelerated images (i.e., inputs) to align them with the SOC images. However, perfect alignment of image pairs was not achievable due to inter-scan motions (e.g., non-rigid respiratory/pulsation). Additionally, slight image contrast differences in focal regions between the two scans (e.g., non-repeatable blood/CSF suppression, protocol adjustment-induced minor contrast changes) were possible. Furthermore, the acquired SOC images may not have all been of high quality, with residual noise and image blurring potentially persisting on the SOC images. All of these factors could induce additional errors (beyond the performance of the SubtleHD algorithm) when measuring pixel-wise

{9}------------------------------------------------

metrics using the SOC images as the reference. To eliminate residual noise and reduce blurring on SOC images, the metrics were also calculated using the SubtleHD-enhanced SOC as the reference, and then metric analysis was performed between the input and SubtleHD enhanced input.

The following selection criteria was used for the paired test data:

Not used for algorithm training. ●
Split between all anatomies (abdomen, ankle, brain, breast, C-spine, foot, hand, hip, knee, . L-spine, neck, pelvis, prostate, shoulder, t-spine, and wrist).
. Have pediatric data:
- o 20 children (ages 2 to 11)
- O 22 adolescents (ages 12 to 21)
Have 0%, 20%, 30%, 40%, 50%, 60%, 70%, and 80% time reduced images. ●
Have axial, coronal, and sagittal orientations.
Have 2D and 3D acquisitions.
Have the following protocols: DWI, FLAIR, GRE, MRA, PD, RADIAL, STIR, SWI, T1, T2, and MIP
Contrasted and non-contrasted images
Have various Acceleration Method Distribution.
Have 3.0T, 1.5T, and low field strengths.
Have a variety of imaging scanner vendors and models.
Have some data from sites that did not also supply training data.
More than 50% of data is from sites in the United States.
. Have various clinical conditions and normal.
Reconstruction Matrix (Image Sizes) up to 512 x 448

This selection criteria represents a well characterized clinically-relevant reference dataset.

In addition to a paired test set, an aligned test set was also evaluated. The purpose of the aligned test set was to better facilitate pixel-wise metrics analysis, which necessitates spatially aligned ground truth reference data. Any misalignment could contribute to metrics that do not reflect the model's performance. Furthermore, the quality of input images should align with the accelerated clinical protocols to ensure that the test results closely approximate practical scenarios. For this test set, therefore, the low-quality input images and high-quality ground truth images were paired from the same undersampled k-space data. Undersampling was performed using commercially available acceleration methods on each vendor's MRI scanner. This ensures that low-quality input images are representative of the types of images that would be processed by these devices in real-world clinical scenarios, where acceleration techniques are frequently used to improve workflow efficiency. The high-quality ground truth images were reconstructed using the vendor's commercially available deep learning pipeline and represent standard-of-care image quality. Importantly, these pairs have perfect spatial alignment, which better facilitates pixel-wise metric analysis, such as L1 loss, SSM, and PSNR. This alignment is critical for accurately assessing the performance of the image enhancement algorithm.

The following selection criteria was used for the aligned test data:

Not used for algorithm training.
Inputs (low-quality images) and targets (high-quality images) are spatially aligned to support the pixelwise based metrics (e.g. L1 loss, SSIM, PSNR) analysis
. Split between all anatomies (abdomen, ankle, arm, brain, extremity, knee, I-spine, pelvis, shoulder, spine)
For known ages, have a variety of age ranges (28 years old - 80 years old)
Have axial, coronal, and sagittal orientations.

{10}------------------------------------------------

Have 2D acquisitions. ●
Have the following protocols: DWI, PD, STIR, T1, T1 FLAIR, T2, T2 FLAIR, T2 STIR. ●
The quality of inputs are aligned with the quality of accelerated clinical protocols
Have 3.0T, 1.5T, and low field strengths.
Have various acceleration factors: 1.0 - 6.0
Have GE and Siemens scanners with various models. ●
. Have US and OUS data.

Each of the paired test sets (SOC as reference, SubtleHD-enhanced SOC as reference) had 97 samples for statistical comparison. The aligned test set had 471 samples. For each test pair, each metric was evaluated across all 2D slices and then averaged across slices to obtain the metric value for that test pair. The mean and standard deviation were calculated for L1 loss, SSIM, and PSNR.

The following table summarizes the results of this testing.

Test Set	Study Test	Result (mean ± std dev)	T-Test P-Value
Paired Test Set(Unaligned SOC asthe Reference)	L1 Loss	9.993% ± 92.487%	0.189
	SSIM	0.0115 ± 0.0403	0.001
	PSNR	-0.307 dB ± 2.863 dB	0.150
Paired Test Set(UnalignedSubtleHD-enhancedSOC as theReference)	L1 Loss	-6.908% ± 22.422%	0.022
	SSIM	0.0210 ± 0.0448	<0.0001
	PSNR	0.583 dB ± 2.754 dB	<0.0001
Aligned Test Set	L1 Loss	-38.837% ± 15.469%	<0.0001
	SSIM	0.0367 ± 0.0219	<0.0001
	PSNR	3.844 dB ± 2.172 dB	<0.0001

Table 3. Standalone Image Quality Metric Summary

Performance Validation & Reader Study Dataset Characteristics:

To represent the patient population and use of MRI in the field, the SubtleHD performance validation test dataset consists of:

Body (breast, abdomen, prostate, and pelvis), Cardiac, Neuro (head, neck, and cervical, lumbar, ● and thoracic spine), and Musculoskeletal (shoulder, wrist, hand, hip, knee, foot, and ankle) anatomical regions
DIXON, DWI, FLAIR, GRE, MRA, PD, RADIAL, STIR, SWI, T1, T2, T2*, and MIP input protocols
Contrasted and non-contrasted images
Axial, coronal, and sagittal orientations
2D and 3D acquisition types ●
. 0.25T, 0.30T, 0.50T, 0.55T, 0.60T, 1.16T, 1.50T, and 3.0T field strengths

{11}------------------------------------------------

ASG Superconductors, ESAOTE, FONAR, Toshiba/Canon, Philips Medical Systems, Hitachi/Fujifilm, GE MEDICAL SYSTEMS, and SIEMENS imaging scanner vendors
Reconstruction Matrix (Image Sizes) up to 1344x1344
. Standard-of-care data and time reduced data (up to 80% time reduction)
Even distribution of subject sex (53% female, 47% male)
Pediatric and Adult subject ages (ranging from 3 to 94 years old) ●

To show that the performance of the device is not hindered by site variability, in the validation dataset, data was selected from sources not included in the training dataset (36.50% of the dataset is from non-training sources). The majority of performance data comes from sources in the United States (65%).

Performance Validation:

Three Regions of Interest (ROIs) were drawn on each image in the SubtleHD performance validation test dataset by a Subtle Medical employee with an MD and/or PhD in a clinically-relevant field. Two rectangular ROI were drawn in homogeneous regions to compare the Signal-to-Noise ratio (SNR) before and after SubtleHD processing. One ROI line was drawn across a tissue interface or anatomy to compare image intensity change along the line before and after SubtleHD processing in terms of slope for all images and full width half max (FWHM) for brain images. A board-certified radiologist reviewed the ROI for acceptability for use in the performance validation study design.

The following are the endpoints, acceptance criteria, results, and conclusions from the SubtleHD Performance Validation:

Endpoint	Acceptance Criteria	SubtleHDMode	Result	Conclusion
Denoising (SNR)Primary Endpoint	SNR shall improve by at least 40% inhomogenous ROI regions for at least 90%of the dataset.	Default	PASS	SubtleHD performsdenoising, in terms ofimproved SNR, MRI images.
	SNR shall improve by at least 40% inhomogenous ROI regions for at least 95%of the dataset.	HighDenoising	PASS
Sharpness(Image IntensityChange) PrimaryEndpoint	Slope in a line ROI is increased for at least90% of the dataset.	Default	PASS	SubtleHD sharpens, in termsof improvement in visibility ofthe edge at a tissue interfaceby image intensity slopemeasure, MRI images.
	Slope in a line ROI is increased for at least95% of the dataset.	HighSharpening	PASS
Sharpness(Image IntensityChange forBrains)SecondaryEndpoint	Thickness, in terms of FWHM in a line ROI,is reduced for at least 90% of the dataset.	Default	PASS	SubtleHD sharpens, in termsof improvement in visibility ofan anatomical structure byimage intensity FWHMmeasure, MRI images.
	Thickness, in terms of FWHM in a line ROI,is reduced for at least 95% of the dataset.	HighSharpening	PASS
Sharpness andOver Smoothing(GradientEntropy) PrimaryEndpoint	At least 90% of cases demonstrate a lowergradient entropy value after SubtleHDprocessing.	Default	PASS	SubtleHD does not result inover-smoothed images, interms of improvement ingradient entropy.
	At least 95% of cases demonstrate a lowergradient entropy value after SubtleHDprocessing.	HighSharpening	PASS

Table 4. Performance Validation Summary

{12}------------------------------------------------

Endpoint	Acceptance Criteria	SubtleHDMode	Result	Conclusion
	There is a statistically significantimprovement in gradient entropy whencomparing the original and SubtleHDenhanced images across the performancedataset per a two-sided paired t-test.	Default andHighSharpening	PASS

Reader Study:

This study utilized human data gathered under the auspices of IRB-approved clinical protocols. 410 image series (205 input and 205 SubtleHD enhanced) were anonymized and randomized prior to the reader study. Readers were blind to the image processing method.

Both the input images and the SubtleHD enhanced images were ranked by board-certified radiologists for perceived SNR, Overall Image Quality / Diagnostic Confidence, Small Structure Visibility, and Imaging Artifacts, utilizing a 4 point Likert scale.

The following are the endpoints, acceptance criteria, results, and conclusions from the SubtleHD Reader Study:

Endpoint	EndpointDescription	Acceptance Criteria	Result	Conclusion
Denoising(PrimaryEndpoint)	Signal-to-Noise Ratio	Statistically significantlybetter with p-value < 0.05or not statisticallysignificantly different in aWilcoxon signed rank test	PASS	SubtleHD performsdenoising, in terms ofimproved SNR, for MRIImages.
	Overall Image Quality/ DiagnosticConfidence	Statistically significantlybetter with p-value < 0.05or not statisticallysignificantly different in aWilcoxon signed rank test	PASS
Sharpness(PrimaryEndpoint)	Visibility of SmallStructures	Statistically significantlybetter with p-value < 0.05or not statisticallysignificantly different in aWilcoxon signed rank test	PASS	SubtleHD performssharpening and does notover-smoooth images, interms of improved visibilityof small structures, for MRIimages.

Table 5. Reader Study Summary

{13}------------------------------------------------

Endpoint	EndpointDescription	Acceptance Criteria	Result	Conclusion
Artifacts(SecondaryEndpoint)	Artifact Introduction	SubtleHD-enhancedimages do not containartifacts that could impactdiagnosis or b) both inputand SubtleHD-enhancedimages are deemed tocontain artifacts that couldimpact diagnosis.	PASS	SubtleHD does not introduceartifacts into MRI images.

Conclusion:

These results demonstrate that SubtleHD is substantially equivalent to the predicate devices.

Regulation Number and Section

§ 892.2050 Medical image management and processing system.

(a)
Identification. A medical image management and processing system is a device that provides one or more capabilities relating to the review and digital processing of medical images for the purposes of interpretation by a trained practitioner of disease detection, diagnosis, or patient management. The software components may provide advanced or complex image processing functions for image manipulation, enhancement, or quantification that are intended for use in the interpretation and analysis of medical images. Advanced image manipulation functions may include image segmentation, multimodality image registration, or 3D visualization. Complex quantitative functions may include semi-automated measurements or time-series measurements.(b)
Classification. Class II (special controls; voluntary standards—Digital Imaging and Communications in Medicine (DICOM) Std., Joint Photographic Experts Group (JPEG) Std., Society of Motion Picture and Television Engineers (SMPTE) Test Pattern).