BrainInsight is intended for automatic labeling, spatial measurement, and volumetric quantification of brain structures from a set of low-field MR images and returns annotated and segmented images, color overlays and reports.

Device Description

BrainInsight is a fully automated MR imaging post-processing medical software that provides image alignment, whole brain segmentation, ventricle segmentation, and midline shift measurements of brain structures from a set MR images. The BrainInsight processing architecture includes a proprietary automated internal pipeline based on machine learning tools. The output annotated and segmented images are provided in standard image format using segmented color overlays and reports that can be displayed on third-party workstations and FDA-cleared Picture Archive and Communications Systems (PACS). The high throughput capability makes the software suitable for use in routine patient care as a support tool for clinicians in assessment of low-field (0.064 T) structural MRIs. BrainInsight provides overlays and reports based on 0.064 T 3D MRI series of T1 Gray/White, T2-Fast, and FLAIR images.

AI/ML Overview

Here's a breakdown of the acceptance criteria and the study details for the BrainInsight™ device, based on the provided text:

1. Table of Acceptance Criteria and Reported Device Performance

The acceptance criteria were defined based on non-inferiority testing, aiming for the model performance to be no worse than the average annotator's discrepancy.

Midline Shift Discrepancy (Lower is Better)

Application	Modality	Acceptance Criteria (Model <= Mean Annotator)	Reported Device Performance (Model Discrepancy)	Reported Mean Annotator Discrepancy
Midline Shift	T1	Model <= 1.42	0.99	1.42
Midline Shift	T2	Model <= 1.00	0.76	1.00
Midline Shift	T2-Fast	Model <= 1.38	1.00	1.38
Midline Shift	FLAIR	Model <= 1.21	0.90	1.21

Lateral Ventricle Segmentation Discrepancy (Lower is Better)

Application	Modality	Acceptance Criteria (Model <= Mean Annotator)	Reported Device Performance (Model Discrepancy)	Reported Mean Annotator Discrepancy
Lateral Ventricle Left	T1	Model <= 0.18	0.17	0.18
Lateral Ventricle Left	T2	Model <= 0.24	0.20	0.24
Lateral Ventricle Left	T2-Fast	Model <= 0.18	0.16	0.18
Lateral Ventricle Left	FLAIR	Model <= 0.12	0.12	0.12
Lateral Ventricle Right	T1	Model <= 0.19	0.19	0.19
Lateral Ventricle Right	T2	Model <= 0.24	0.22	0.24
Lateral Ventricle Right	T2-Fast	Model <= 0.16	0.15	0.16
Lateral Ventricle Right	FLAIR	Model <= 0.13	0.13	0.13

Mean Absolute Error for Midline Shift (Lower is Better)

Application	Modality	Acceptance Criteria (Implicitly, to be within acceptable clinical error)	Reported Device Performance (Error)
Midline Shift	T1	Not explicitly stated, but clinical acceptability implied by meeting non-inferiority	1.01 mm
Midline Shift	T2	Not explicitly stated, but clinical acceptability implied by meeting non-inferiority	0.80 mm
Midline Shift	T2-Fast	Not explicitly stated, but clinical acceptability implied by meeting non-inferiority	0.89 mm
Midline Shift	FLAIR	Not explicitly stated, but clinical acceptability implied by meeting non-inferiority	0.75 mm

Dice Overlap and Volume Differences for Segmentation (Higher Dice, Lower Volume Difference are Better)

Application	Modality	Performance Metric	Acceptance Criteria (Implicitly, to be clinically acceptable and comparable to annotators)	Device Performance	Annotator Performance
Left Ventricle	T1	Dice Overlap (%)	Not explicitly stated	85	90
Right Ventricle	T1	Dice Overlap (%)	Not explicitly stated	83	90
Whole Brain	T1	Dice Overlap (%)	Not explicitly stated	95	97
Left Ventricle	T1	Volume Differences (%)	Not explicitly stated	25	9
Right Ventricle	T1	Volume Differences (%)	Not explicitly stated	26	11
Whole Brain	T1	Volume Differences (%)	Not explicitly stated	3	2
Left Ventricle	T2	Dice Overlap (%)	Not explicitly stated	84	88
Right Ventricle	T2	Dice Overlap (%)	Not explicitly stated	82	87
Whole Brain	T2	Dice Overlap (%)	Not explicitly stated	96	97
Left Ventricle	T2	Volume Differences (%)	Not explicitly stated	27	21
Right Ventricle	T2	Volume Differences (%)	Not explicitly stated	26	20
Whole Brain	T2	Volume Differences (%)	Not explicitly stated	5	5
Left Ventricle	T2-Fast	Dice Overlap (%)	Not explicitly stated	86	91
Right Ventricle	T2-Fast	Dice Overlap (%)	Not explicitly stated	86	92
Left Ventricle	T2-Fast	Volume Differences (%)	Not explicitly stated	26	17
Right Ventricle	T2-Fast	Volume Differences (%)	Not explicitly stated	23	13
Left Ventricle	FLAIR	Dice Overlap (%)	Not explicitly stated	89	93
Right Ventricle	FLAIR	Dice Overlap (%)	Not explicitly stated	88	94
Left Ventricle	FLAIR	Volume Differences (%)	Not explicitly stated	9	7
Right Ventricle	FLAIR	Volume Differences (%)	Not explicitly stated	11	8

Summary of Device Performance against Acceptance Criteria:
The document states: "The test results show high accuracy of BrainInsight performance as compared to the reference and annotators and the subject device met all acceptance criteria." This implies that for all metrics where non-inferiority criteria were set (Midline Shift Discrepancy and Lateral Ventricle Discrepancy), the model performed as well as or better than the mean annotator. For other metrics, the performance was presented as being accurate and acceptable.

2. Sample Size Used for the Test Set and Data Provenance

Sample Size for Test Set: The exact numerical sample size for the test set is not explicitly stated. However, the document mentions that each model and application were validated using an appropriate sample size to yield statistically significant results.
Data Provenance:
- Country of Origin: Not specified.
- Retrospective or Prospective: Not specified.
- Acquisition Device: All test images were acquired using Hyperfine Swoop Portable MR imaging system with software versions 8.3 and 8.4.
Test Set Distribution:
- Age: >2 to 12 years (20.6%), >12 to <18 years (8.8%), >18 to 90 years (70.6%)
- Gender: 33% Female / 41% Male / 25% Anonymized
- Pathology: Stroke (Infarct), Hydrocephalus, Hemorrhage (SAH, SDH, IVH, IPH), Mass/Edema, Tumor.

3. Number of Experts Used to Establish the Ground Truth for the Test Set and Their Qualifications

Number of Experts: The document states that the datasets for training and validation were annotated by "multiple experts." It then mentions that "The entire group of training image sets was divided into segments and each segment was given to a single expert." This phrasing is somewhat ambiguous for the test set specifically. It is implied that multiple experts were involved in the ground truth establishment for the overall process, but it doesn't clearly state how many experts independently evaluated each case in the test set, nor if the "single expert per segment" approach also applied to the test set ground truth.
Qualifications of Experts: Not specified beyond being referred to as "experts" and "annotators."

4. Adjudication Method for the Test Set

The adjudication method varies by application:

Midline Shift: Ground truth was determined based on the average shift distance of all annotators. This implies a form of consensus or averaging method rather than a strict adjudication by a senior expert.
Segmentation (Lateral Ventricles, Whole Brain): Ground truth for segmentation was calculated using Simultaneous Truth and Performance Level Estimation (STAPLE). STAPLE is an algorithm that estimates a "true" segmentation from multiple segmentations, weighting them based on their estimated performance. This is an algorithmic adjudication method.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study

Was a MRMC study done? No, a traditional MRMC comparative effectiveness study that measures how human readers improve with AI vs. without AI assistance was not explicitly described for this submission. The study focuses on standalone performance of the AI model against expert annotations and the "mean annotator" performance.
Effect Size of Human Improvement (if applicable): Not applicable, as an MRMC comparative effectiveness study was not detailed.

6. Standalone (Algorithm Only Without Human-in-the-Loop Performance) Study

Was a standalone study done? Yes, the described performance evaluation appears to be a standalone (algorithm only) study. The device's performance is compared directly against the ground truth established by annotators, and against the mean discrepancy of the annotators themselves. There is no mention of human readers using the AI output to improve their performance compared to a baseline.

7. Type of Ground Truth Used

The type of ground truth used varies by the measurement:

Midline Shift: Expert consensus, calculated as the average shift distance of all annotators.
Segmentation (Lateral Ventricles, Whole Brain): Algorithmic consensus, calculated using Simultaneous Truth and Performance Level Estimation (STAPLE) based on expert annotations.
General: It is based on expert annotations of images acquired from the Hyperfine Swoop portable MRI system.

8. Sample Size for the Training Set

Sample Size for Training Set: The exact numerical sample size for the training set is not explicitly stated. The document only mentions that the data collection for the training and validation datasets was done at "multiple sites."

9. How the Ground Truth for the Training Set Was Established

The data collection for the training and validation datasets was done at multiple sites.
The datasets were annotated by multiple experts.
The "entire group of training image sets was divided into segments and each segment was given to a single expert."
"The expert's determination became the ground truth for each image set in their segment." This implies a form of single-reader ground truth for each segmented batch, rather than multi-reader consensus for every single case within the training set.

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 is a blue square with the letters "FDA" in white. To the right of the blue square is the text "U.S. FOOD & DRUG ADMINISTRATION" in blue.

December 16, 2022

Hyperfine, Inc. % Christine Kupchick Sr. Regulatory Specialist 351 New Whitfield St. GUILFORD CT 06437

Re: K223268

Trade/Device Name: BrainInsightTM Regulation Number: 21 CFR 892.2050 Regulation Name: Medical image management and processing system Regulatory Class: Class II Product Code: QIH Dated: October 21, 2022 Received: October 24, 2022

Dear Christine Kupchick:

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 (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 located 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.

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 803) for

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

devices or postmarketing safety reporting (21 CFR 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 4, Subpart A) for combination products; and, if applicable, the electronic product radiation control provisions (Sections 531-542 of the Act); 21 CFR 1000-1050.

Also, please note the regulation entitled, "Misbranding by reference to premarket notification" (21 CFR Part 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.

D. Ryk

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

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

Indications for Use

510(k) Number (if known) K223268

Device Name BrainInsight

Indications for Use (Describe)

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."

FORM FDA 3881 (6/20)

PSC Publishing Services (301) 443-6740 EFF

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

HYPERFINE

BrainInsight™ 510(k) SUMMARY K223268

510(k) Submitter

Company Name:	Hyperfine, Inc.
Company Address:	351 New Whitfield StGuilford. CT 06437

Contact

Name:	Christine Kupchick
Telephone:	(203) 343-3404
Email:	ckupchick@hyperfine.io

Date Prepared:

December 13, 2022

Device Identification

Trade Name:	BrainInsight™
Common Name:	Automated Radiological Image Processing Software
Regulation Number:	21 CFR 892.2050
Product Code:	QIH
Regulatory Class:	Class II

Predicate Device Information

The subject BrainInsight is substantially equivalent to the predicate BrainInsight (K220815). The predicate device has not been subject to a design-related recall.

Device Description

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

Indications for Use

Intended Patient Population

The table below shows the intended patient population.

Application	Patient Population	T1 Gray/White	T2	T2-Fast	FLAIR
Midline Shift	Ages 2+	V	V	V	V
Lateral Ventricles	Ages 2+	V	V	V	V
Whole Brain	Ages 18+	V	V	N/A	N/A

Technological Characteristics

The subject device has the same indications for use, fundamental technology, and operating principles, as the predicate (K220815). Therefore, the subject device is substantially equivalent to the predicate.

Substantial Equivalence Discussion

The table below compares the subject device to the predicate.

Attribute	Subject BrainInsight	Predicate BrainInsight (K220815)
Indications for Use	BrainInsight is intended for automatic labeling,spatial measurement, and volumetricquantification of brain structures from a set oflow-field MR images and returns annotated andsegmented images, color overlays and reports.	Same
Target AnatomicalSites	Brain	Same
Patient Population	Adult and pediatric (≥ 2 years) - Lateralventricles and midline shift applicationsAdult (≥ 18 years) - Whole brain application	Adult (≥ 18 years) - Lateral ventricles,midline shift, and whole brain applications
Technology	Automated measurement of brain tissuevolumes and structures of Al-reconstructedlow-field MR imagesAutomatic segmentation and quantificationof brain structures of Al-reconstructed low-field MR images using machine learningtools	Same
Method of Use	MR images are automatically sent toBrainInsight, and processed images areautomatically returned in ~7 minutes	Same
User Interface /PhysicalCharacteristics	No software requiredOperates in a serverless cloud environmentUser interface through PACS (multiplevendors)	Same
Operating System	Supports Linux	Same

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

ProcessingArchitecture	Automated internal pipeline that performs:• segmentation• volume calculation• distance measurement• numerical information display	Same
Data Source	• MRI Scanner: Hyperfine Swoop FSE MRI T1-Gray/White Contrast, T2, T2-Fast and FLAIRscans acquired with specified protocols• Supports DICOM format as input	• MRI Scanner: Hyperfine Swoop FSEMRI T1 and T2 scans acquired withspecified protocols• Supports DICOM format as input
Output	Provides volumetric measurements of brainstructures:• Includes segmented color overlays andmorphometric reports• Supports DICOM format as output of resultsthat can be displayed on DICOMworkstations and PACS	Same
Safety	Automated quality control functions:• Tissue contrast check• Scan protocol verification• Atlas alignment check• Results must be reviewed by a trainedphysician• LV segmentation output quality check	Automated quality control functions:• Tissue contrast check• Scan protocol verification• Atlas alignment check• Results must be reviewed by a trainedphysician

PERFORMANCE

As part of demonstrating substantial equivalence to the predicate, a risk analysis was completed to identify the risks associated with the software modifications. Software verification as related to the modifications was performed per IEC 62304:2006 and as recommended in the FDA Guidance for the Content of Premarket Submissions for Software Contained in Medical Devices." The subject device passed all the testing in accordance with internal requirements and applicable standards to support substantial equivalence.

Each model was trained using a training dataset to optimize parameters and a separate validation dataset to select the best set of parameters. Comparing the training and validation metrics helps to monitor and prevent overfitting. The datasets were augmented to improve robustness using standard techniques that apply transformations to the input data. The testing dataset was separate from training and validation datasets. Each subject was assigned a unique identifier and all subjects in training and validation data were excluded from the test set.

The data collection for the training and validation datasets were done at multiple sites. Each site used the T1-Gray/White contrast and T2, T2-Fast and FLAIR sequences from the FDA cleared Hyperfine Swoop Portable MR imaging system. The datasets were annotated by multiple experts. The entire group of training image sets was divided into segments and each segment was given to a single expert. The expert's determination became the ground truth for each image set in their segment. Each model and application were validated using an appropriate sample size to yield statistically significant results. All test images were acquired using Swoop software versions 8.3 and 8.4. The test set had the following distribution:

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

Category	Data Distribution
Age	>2 to 12 years - 20.6% >12 to <18 years - 8.8% >18 to 90 years - 70.6%
Gender	33% F / 41% M / 25% Anonymized
Pathology	Stroke (Infarct) Hydrocephalus Hemorrhage (SAH, SDH, IVH, IPH) Mass/Edema Tumor

Quantitative evaluation was performed to validate performance using software. The performance of the model and the annotators to the consensus-based annotation was computed to ensure that the model performance is no worse than the average annotator. The acceptance criteria were defined based on non-inferiority testing, in which the model discrepancy to the annotators can be no worse than the average annotator discrepancy.

Midline ShiftDiscrepancy	T1	T2	T2-Fast	FLAIR
Model	0.99	0.76	1.00	0.90
Mean Annotator	1.42	1.00	1.38	1.21

Lateral Ventricle LeftDiscrepancy	T1	T2	T2-Fast	FLAIR
Model	0.17	0.20	0.16	0.12
Mean Annotator	0.18	0.24	0.18	0.12

Lateral Ventricle RightDiscrepancy	T1	T2	T2-Fast	FLAIR
Model	0.19	0.22	0.15	0.13
Mean Annotator	0.19	0.24	0.16	0.13

The mean absolute error was used to calculate the error range for midline shift. Ground truth for midline shift was determined based on the average shift distance of all annotators.

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

Application	T1 Error	T2 Error	T2-Fast Error	FLAIR Error
Midline Shift	1.01 mm	0.80 mm	0.89 mm	0.75 mm

The mean Dice coefficient was used to calculate the error range for the lateral ventricles and whole brain. Ground truth for segmentation is calculated using Simultaneous Truth and Performance Level Estimation (STAPLE).

Application	Dice Overlap [%]		Volume Differences [%]
T1	Device	Annotator	Device	Annotator
Left Ventricle	85	90	25	9
Right Ventricle	83	90	26	11
Whole Brain	95	97	3	2

Application	Dice Overlap [%]		Volume Differences [%]
T2	Device	Annotator	Device	Annotator
Left Ventricle	84	88	27	21
Right Ventricle	82	87	26	20
Whole Brain	96	97	5	5

Application	Dice Overlap [%]		Volume Differences [%]
T2-Fast	Device	Annotator	Device	Annotator
Left Ventricle	86	91	26	17
Right Ventricle	86	92	23	13

Application	Dice Overlap [%]		Volume Differences [%]
FLAIR	Device	Annotator	Device	Annotator
Left Ventricle	89	93	9	7
Right Ventricle	88	94	11	8

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

The test results show high accuracy of BrainInsight performance as compared to the reference and annotators and the subject device met all acceptance criteria.

Conclusion

Based on the indications for use, technological characteristics, performance results, and comparison to the predicate, the subject BrainInsight has been shown to be substantially equivalent to the predicate and does not present any new issues of safety or effectiveness.

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).