(102 days)
MuscleView 2.0 is a magnetic resonance diagnostic software device used in adults and pediatrics aged 18 and older which automatically segments muscle, bone, fat and other anatomical structures from magnetic resonance imaging. After segmentation, it enables the generation, display and review of magnetic resonance imaging data. The segmentation results need to be reviewed and edited using appropriate software. Other physical parameters derived from the images may also be produced. This device is not intended for use with patients who have tumors in the trunk, arms and/or lower limb(s). When interpreted by a trained clinician, these images and physical parameters may yield information that may assist in diagnosis.
MuscleView 2.0 is a software-only medical device which performs automatic segmentation of musculoskeletal structures. The software utilizes a locked artificial intelligence/machine learning (AI/ML) algorithm to identify and segment anatomical structures for quantitative analysis. The input to the software is DICOM data from magnetic resonance imaging (MRI), but the subject device does not directly interface with any devices. The output includes volumetric and dimensional metrics of individual and grouped regions of interest (ROIs) (such as muscles, bones and adipose tissue) and comparative analysis against a Virtual Control Group (VCG) derived from reference population data.
MuscleView 2.0 builds upon the predicate device, MuscleView 1.0 (K241331, cleared 10/01/2024), which was cleared for the segmentation and analysis of lower extremity structures (hips to ankles). The subject device extends functionality to include:
- Upper body regions (neck to hips)
- Adipose tissue segmentation (subcutaneous, visceral, intramuscular, and hepatic fat)
- Quantitative comparison with a Virtual Control Group
- Additional derived metrics including Z-scores and composite scores (e.g., muscle-bone score)
The submission includes a Predetermined Change Control Plan which details the procedure for retraining AI/ML algorithms or adding data to the Virtual Control Groups in order to improve performance without negatively impacting the safety or efficacy of the device.
Here's a breakdown of the acceptance criteria and the study proving the device meets them, based on the provided FDA 510(k) clearance letter for MuscleView 2.0:
1. Table of Acceptance Criteria and Reported Device Performance
The acceptance criteria for MuscleView 2.0 were based on the device's segmentation accuracy, measured by Dice Similarity Coefficient (DSC) and absolute Volume Difference (VDt), remaining within the interobserver variability observed among human experts. The study demonstrated the device met these criteria. Since the text explicitly states the AI model's performance was "consistently within these predefined interobserver ranges," and "passed validation," the reported performance for all ROIs was successful in meeting the acceptance criteria.
| Metric | Acceptance Criteria | Comment on Reported Performance |
|---|---|---|
| Dice Similarity Coefficient (DSC) | DSC values where the 95% confidence interval for each ROI (across all subgroup analyses) indicates performance at or below interobserver variability (meaning higher DSC, closer to 1.0, is better). Specifically, a desired outcome was "a mean better than or equal to the acceptance criteria." | Consistently within predefined interobserver ranges and passed validation for all evaluated ROIs and subgroups. (See Table 1 for 95% CIs of individual ROIs across subgroups). |
| Absolute Volume Difference (VDt) | VDt values where the 95% confidence interval for each ROI (across all subgroup analyses) indicates performance at or below interobserver variability (meaning lower VDt, closer to 0, is better). Specifically, a desired outcome was "a mean better than or equal to the acceptance criteria." | Consistently within predefined interobserver ranges and passed validation for all evaluated ROIs and subgroups. (See Table 2 for 95% CIs of individual ROIs across subgroups). |
2. Sample Sizes Used for the Test Set and Data Provenance
| AI Setting | Number of Unique Scans | Number of Unique Subjects | Data Provenance |
|---|---|---|---|
| AI Setting 1 (Lower Extremity) | 148 | 148 | Retrospective, "diverse population," multiple imaging sites and MRI manufacturers (GE, Siemens, Philips, Canon, Toshiba/Other). Countries of origin not explicitly stated, but "regional demographics" are provided implying a mix of populations. |
| AI Setting 2 & 3 (Upper Extremity and Adipose Tissue) | 171 | 171 | Retrospective, "diverse population," multiple imaging sites and MRI manufacturers (GE, Siemens, Philips, Canon, Other/Unknown). Countries of origin not explicitly stated, but "regional demographics" are provided implying a mix of populations. |
- Overall Test Set: 148 unique subjects (for AI Setting 1) + 171 unique subjects (for AI Settings 2 & 3) = 319 unique subjects.
- Data Provenance: Retrospective, curated collection of MRI datasets from a diverse patient population (age, BMI, biological sex, ethnicity) from multiple imaging sites and MRI manufacturers (GE, Siemens, Philips, Canon, Toshiba/Other/Unknown). Independent from training datasets. De-identified.
3. Number of Experts Used to Establish Ground Truth for the Test Set and Qualifications
- Number of Experts: Not explicitly stated, but referred to as "expert segmentation analysts" and "expert human annotation." The study mentions "consensus process by expert segmentation analysts" for training data, and for testing, "manual segmentation performed by experts" and that the "interobserver variability range observed among experts" was used as a benchmark. The document does not specify the exact number of experts or their specific qualifications (e.g., years of experience or board certification).
4. Adjudication Method for the Test Set
- Adjudication Method: The ground truth for both training and testing datasets was established through a "consensus process by expert segmentation analysts" for training data and "manual segmentation performed by experts" for the test set. It does not explicitly state a 2+1 or 3+1 method; rather, it implies a consensus was reached among the experts. The key here is the measurement of "interobserver variability," suggesting that multiple experts initially segmented the data, and their agreement (or discordance) defined the benchmark, from which a final consensus might have been derived.
5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study
- No MRMC study was performed. The performance testing was a standalone study comparing the AI segmentation to expert manual segmentation (ground truth) rather than comparing human readers with and without AI assistance. The text states: "Performance results demonstrated segmentation accuracy within the interobserver variability range observed among experts." This indicates a comparison of the AI's output against what multiple human experts would agree upon, not an evaluation of human performance improvement with AI.
6. Standalone Performance Study
- Yes, a standalone study was done. The document states: "To evaluate the performance of the MuscleView AI segmentation algorithm, a comprehensive test was conducted using a test set that was fully independent from the training set. The AI was blinded to the ground truth segmentation labels during inference, ensuring an unbiased comparison." This clearly describes a standalone performance evaluation of the algorithm.
7. Type of Ground Truth Used
- Expert Consensus / Expert Manual Segmentation: The ground truth was established by "manual segmentation performed by experts" and through a "consensus process by expert segmentation analysts." This is a form of expert consensus derived from detailed manual annotation. The benchmark for acceptance was the "interobserver variability range observed among experts."
8. Sample Size for the Training Set
| AI Setting | Number of Unique Scans | Number of Unique Subjects |
|---|---|---|
| AI Setting 1 (Lower Extremity) | 1658 | 1294 |
| AI Setting 2 & 3 (Upper Extremity and Adipose Tissue) | 392 | 209 |
| Total Unique Subjects for Training: 1294 + 209 = 1503 (Note: Some subjects might be present in both sets if they had both lower and upper extremity scans, but the table specifies "unique subjects" per AI setting) |
- Total Training Set: 1658 (scans for AI Setting 1) + 392 (scans for AI Settings 2 & 3) = 2050 unique scans.
- Total Unique Subjects: 1294 (for AI Setting 1) + 209 (for AI Settings 2 & 3) = 1503 unique subjects.
9. How Ground Truth for the Training Set Was Established
- The ground truth for the training set was established through a "consensus process by expert segmentation analysts" on a "curated collection of retrospective MRI datasets."
FDA 510(k) Clearance Letter - MuscleView 2.0
Page 1
U.S. Food & Drug Administration
10903 New Hampshire Avenue
Silver Spring, MD 20993
www.fda.gov
Doc ID # 04017.08.00
September 9, 2025
Springbok, Inc. (dba Springbok Analytics)
Scott Magargee
Chief Executive Officer
110 Old Preston Avenue
Charlottesville, Virginia 22902
Re: K251682
Trade/Device Name: MuscleView 2.0
Regulation Number: 21 CFR 892.1000
Regulation Name: Magnetic Resonance Diagnostic Device
Regulatory Class: Class II
Product Code: LNH
Dated: August 1, 2025
Received: August 1, 2025
Dear Scott Magargee:
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.
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
Page 2
K251682 - Scott Magargee Page 2
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 device, then 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 (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-reporting-combination-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 requirements, please see the UDI System webpage at https://www.fda.gov/medical-devices/device-advice-comprehensive-regulatory-assistance/unique-device-identification-system-udi-system.
Page 3
K251682 - Scott Magargee Page 3
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-devices/medical-device-safety/medical-device-reporting-mdr-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/medical-devices/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-devices/device-advice-comprehensive-regulatory-assistance/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
Page 4
Indications for Use
Please type in the marketing application/submission number, if it is known. This textbox will be left blank for original applications/submissions.
Please provide the device trade name(s).
MuscleView 2.0
Please provide your Indications for Use below.
MuscleView 2.0 is a magnetic resonance diagnostic software device used in adults and pediatrics aged 18 and older which automatically segments muscle, bone, fat and other anatomical structures from magnetic resonance imaging. After segmentation, it enables the generation, display and review of magnetic resonance imaging data. The segmentation results need to be reviewed and edited using appropriate software. Other physical parameters derived from the images may also be produced. This device is not intended for use with patients who have tumors in the trunk, arms and/or lower limb(s). When interpreted by a trained clinician, these images and physical parameters may yield information that may assist in diagnosis.
Please select the types of uses (select one or both, as applicable).
☑ Prescription Use (Part 21 CFR 801 Subpart D)
☐ Over-The-Counter Use (21 CFR 801 Subpart C)
MuscleView 2.0 Page 8 of 44
Page 5
18.1 – 510(k) Summary
A better view of health
18.1 – 510(k) Summary Page 1 of 12
09/02/2025
Applicant
Scott Magargee
Chief Executive Officer
Springbok, Inc.
110 Old Preston Ave
Charlottesville, VA 22902 USA
Phone: 1-215-680-9078
Email: scott.magargee@springbokanalytics.com
Device Information
Trade Name: MuscleView 2.0
Regulation: 892.1000 – Magnetic Resonance Diagnostic Device
Product Code: LNH (Class 2) – System, Nuclear Magnetic Resonance Imaging
Intended Use
MuscleView uses Artificial Intelligence/Machine Learning (AI/ML) algorithms to automatically segment anatomical structures in magnetic resonance imaging data. This segmentation and the associated metrics are intended to assist healthcare professionals in the visualization and quantification of anatomical structures.
Indications for Use
MuscleView 2.0 is a magnetic resonance diagnostic software device used in adults and pediatrics aged 18 and older which automatically segments muscle, bone, fat and other anatomical structures from magnetic resonance imaging. After segmentation, it enables the generation, display and review of magnetic resonance imaging data. The segmentation results need to be reviewed and edited using appropriate software. Other physical parameters derived from the images may also be produced. This device is not intended for use with patients who have tumors in the trunk, arms and/or lower limb(s). When interpreted by a trained clinician, these images and physical parameters may yield information that may assist in diagnosis.
Page 6
A better view of health
Device Description
MuscleView 2.0 is a software-only medical device which performs automatic segmentation of musculoskeletal structures. The software utilizes a locked artificial intelligence/machine learning (AI/ML) algorithm to identify and segment anatomical structures for quantitative analysis. The input to the software is DICOM data from magnetic resonance imaging (MRI), but the subject device does not directly interface with any devices. The output includes volumetric and dimensional metrics of individual and grouped regions of interest (ROIs) (such as muscles, bones and adipose tissue) and comparative analysis against a Virtual Control Group (VCG) derived from reference population data.
MuscleView 2.0 builds upon the predicate device, MuscleView 1.0 (K241331, cleared 10/01/2024), which was cleared for the segmentation and analysis of lower extremity structures (hips to ankles). The subject device extends functionality to include:
- Upper body regions (neck to hips)
- Adipose tissue segmentation (subcutaneous, visceral, intramuscular, and hepatic fat)
- Quantitative comparison with a Virtual Control Group
- Additional derived metrics including Z-scores and composite scores (e.g., muscle-bone score)
The submission includes a Predetermined Change Control Plan which details the procedure for retraining AI/ML algorithms or adding data to the Virtual Control Groups in order to improve performance without negatively impacting the safety or efficacy of the device.
Predicate Device Information
Trade Name: MuscleView 1.0
510K Number: K241331
Regulation: 892.1000 – Magnetic Resonance Diagnostic Device
Product Code: LNH (Class 2) – System, Nuclear Magnetic Resonance Imaging
Reference Device Information
Trade Name: AMRA Profiler
510K Number: K211983
Regulation: 892.1000 – Magnetic Resonance Diagnostic Device
Product Code: LNH (Class 2) – System, Nuclear Magnetic Resonance Imaging
Page 7
A better view of health
AI/ML Model Cards
| Model | Release Date | Anatomical Region | Model Function |
|---|---|---|---|
| MuscleView 2.0 | 05/30/2025 | Full Body | Assistive AI |
Model Training and Testing
The MuscleView 2.0 – Setting 1 model underwent supervised machine learning using a curated collection of retrospective MRI datasets from a diverse population (n = 1294 unique subjects). Each dataset was labeled through a consensus process by expert segmentation analysts to establish ground truth.
The model was tested on diverse datasets (n = 148 unique subjects) independent from the training datasets and blinded to ground truth segmentations. Performance results demonstrated segmentation accuracy within the interobserver variability range observed among experts.
Model Settings
All AI/ML inferences are performed automatically, with no configuration, calibration or exposed modifiable parameters. The model behavior is locked and deterministic.
Privacy
Patient data is kept within the user's network and is not used to train or test future models.
Predicate Device Comparison
| Attribute | Subject Device | Predicate Device | Reference Device | Comment |
|---|---|---|---|---|
| Device Name | MuscleView 2.0 | MuscleView 1.0 | AMRA Profiler | n/a |
| Manufacturer | Springbok, Inc. | Springbok, Inc. | AMRA | n/a |
| 510(k) Number | TBD | K241331 | K211983 | n/a |
| Regulation | 892.1000 | 892.1000 | 892.1000 | No difference |
| Product Code | LNH | LNH | LNH | No difference |
Page 8
A better view of health
| Attribute | Subject Device | Predicate Device | Reference Device | Comment |
|---|---|---|---|---|
| Indications for Use | MuscleView is a magnetic resonance diagnostic software device which automatically segments muscle, bone, fat and other anatomical structures from magnetic resonance imaging. After segmentation, it enables the generation, display and review of magnetic resonance imaging data. Other physical parameters derived from the images may also be produced. When interpreted by a trained clinician, these images and physical parameters may yield information that may assist in diagnosis. | MuscleView is intended for use in adults and pediatric patients aged 18 years and older to automatically segment muscle and bone structures of the lower extremities from magnetic resonance imaging using a machine learning-based approach. Following segmentation, it provides derived metrics including muscle volume, bone volume, intramuscular fat percentage, and left/right asymmetry. The software is intended to be used by physicians trained in the interpretation of MRI images and serves as an initial method for segmenting muscle and bone structures from one or more study series. Segmentation results must be reviewed and, if necessary, edited using appropriate software. MuscleView is intended solely to provide segmentation and derived metrics for muscle and bone structures and is not intended to directly support the diagnosis of any disease. This device is not intended for use in patients with tumors in the lower limbs. | Indicated for use as a magnetic resonance diagnostic device software application for non-invasive fat and muscle evaluation, this device enables the generation, display, and review of 2D magnetic resonance medical image data. It is designed to utilize DICOM 3.0 compliant magnetic resonance image datasets acquired from compatible MR systems to display the internal structure of the body, including the liver. Other physical parameters derived from the images may also be produced. The software provides a number of quantification tools, such as Region of Interest (ROI) placements, to be used for the assessment of regions within an image to quantify liver tissue characteristics, including the determination of fat fraction in the liver, T2*, and muscle volume. These images and the physical parameters derived from them, when interpreted by a trained clinician, yield information that may assist in diagnosis. | The subject device expands upon the capabilities of the predicate device by supporting segmentation and analysis of a broader range of anatomical regions of interest (ROIs), including ROIs in the upper body and different types of adipose tissue. Furthermore, the subject device includes a comparative function similar to that of the reference device, enabling comparison of patient-specific metrics to a virtual control group. |
| Anatomy | Muscles, bones and fat structures in the legs, trunk and arms. | Muscles, bones and fat structures in the legs. | Muscles, bones and fat structures in the legs | |
| Intended Users | Trained clinicians | Trained clinicians | Trained clinicians | No difference |
| Target Population | Adults and pediatric patients aged 18 and older | Adults and pediatric patients aged 18 and older | Adults | No difference |
| Contraindications | Tumors in the trunk, arms and/or lower limbs | Tumors in the lower extremities | None, software only | Subject device contraindications reflect the expanded indications for use from the predicate device. |
| Imaging modality | Magnetic Resonance Imaging | Magnetic Resonance Imaging | Magnetic Resonance Imaging | No difference |
| MRI Scanners | GE, Siemens, Philips | GE, Siemens, Philips | GE, Siemens, Philips | No difference |
| MR Field Strength | 1.5T, 3.0T | 1.5T, 3.0T | 1.5T, 3.0T | No difference |
Page 9
A better view of health
Software Testing
MuscleView 2.0 has been tested in alignment with the FDA's guidance on software functions ("Content of Premarket Submissions for Device Software Functions", June 14, 2023) and cybersecurity ("Cybersecurity in Medical Devices: Quality System Considerations and Content of Premarket Submissions", September 27, 2023). It has successfully met the predetermined acceptance criteria for the following tests:
- Unit Testing
- Integration Testing
- Cybersecurity Testing
- Software Verification
- Software Validation
Performance Testing
To evaluate the performance of the MuscleView AI segmentation algorithm, a comprehensive test was conducted using a test set that was fully independent from the training set. The AI was blinded to the ground truth segmentation labels during inference, ensuring an unbiased comparison. De-identified subject scans were included in the test cohort. The cohort was demographically diverse, including both male and female participants across a range of age groups and body mass indices (BMI), and scans acquired from multiple imaging sites and MRI manufacturers (Siemens, GE, Philips) to ensure clinical applicability.
| MRI Manufacturers | Biological Sex & Age | Patient Population |
|---|---|---|
| GE | Males 18-21 | Healthy Adult/Athlete |
| Siemens | Males over 21 | Patient |
| Philips | Females 18-21 | |
| Cannon | Females over 21 | |
| Toshiba (AI Model Setting 1 only) |
Performance testing of segmentation and associated metrics was performed by comparing the results obtained to a reference standard developed by manual segmentation performed by experts. Validation Tests met the acceptance criteria if the dice similarity coefficient (DSC) or volume difference (VDt) was below interobserver variability, establishing boundaries that reflect clinically acceptable variance among expert human reviewers. The AI model's performance was consistently within these predefined interobserver ranges for nearly all ROIs evaluated.
These results demonstrate that the AI segmentation system performs at a level equivalent to expert human annotation, with robustness confirmed across imaging vendors, patient populations, and anatomical diversity. This supports the generalizability of the model for clinical deployment in a wide variety of real-world imaging environments and patient demographics.
Page 10
A better view of health
Summary of AI Training & Testing Dataset
The AI responsible for muscle segmentation within MuscleView was trained and tested on MRI scans generated across varying patients, demographics (including – but not limited to – gender, age and ethnicity), scan sites, and MRI parameters (including manufacturer, magnetic field strength, series settings/types, matrix size, field of view, and scan resolution). See the table below detailing variation in the data.
| Factor | Groups | AI Setting 1 (Lower Extremity) | AI Setting 2 & 3 (Upper extremity and Adipose Tissue) | ||
|---|---|---|---|---|---|
| Training data | Validation data | Training data | Validation data | ||
| Number of unique scans | n/a | 1658 | 148 | 392 | 171 |
| Number of unique subjects | n/a | 1294 | 148 | 209 | 171 |
| Gender | Male | 1192* | 102 | 230* | 96* |
| Female | 374* | 46 | 156* | 74* | |
| Age* | Mean | 29 | 31.7 | 41.8 | 39.3 |
| Standard Deviation | 13.4 | 15.5 | 19.4 | 17.7 | |
| Ethnicity (based on regional demographics) | % Non-Hispanic White | 52 | 52 | 50 | 50 |
| % Hispanic/Latino | 18 | 18 | 17 | 15 | |
| % Black/African American | 14 | 14 | 11 | 14 | |
| % Asian | 10 | 10 | 13 | 7 | |
| % Australian | 2 | 2 | 3 | 6 | |
| % American Indian / Alaska Native | <1 | <1 | <1 | <1 | |
| % Native Hawaiian / Pacific Islander | <1 | <1 | <1 | <1 | |
| % Australian Aboriginal | <1 | <1 | <1 | 1 | |
| % Other | 2 | 2 | 2 | 3 | |
| MRI Manufacturer | Siemens | 680 | 74 | 293 | 116 |
| GE Medical Systems | 435 | 28 | 54 | 26 | |
| Philips Medical Systems | 82 | 29 | 41 | 25 | |
| Canon | 1 | 4 | 0 | 3 | |
| Other (Toshiba) or Unknown | 460 | 13 | 4 | 1 |
*Demographic data were not available for some scans.
Independence of test data from training data was ensured in several ways. (1) MRI Data used to train the AI was explicitly separate from validation data (both as MRI scans and as subjects), (2) the majority (70% - 84%) of datasets in the validation set came from imaging centers and organizations not used in the train datasets (19 – 22 new sites), and (3) the personnel involved in establishing the reference standard for the AI were not involved in the algorithm's development to ensure the independence of training and testing.
The three AI settings responsible for the segmentation of 172 ROIs - setting 1 (lower extremity), setting 2 (upper extremity), and setting 3 (adipose tissue) - were validated, demonstrating their accuracy across many varying scan and patient cases as related to the gold-standard of label generation – interobserver
Page 11
A better view of health
repeatability of manually vetted labels. AI setting 1 consisted of a separate training and testing set than settings 2 & 3. The segmentation performance and derived metrics of all ROIs passed the predetermined acceptance criteria in both healthy and patient population subgroups, i.e. a one-sample T-test between the results of the validation set and the acceptance criteria threshold (the 95% confidence interval from the interobserver repeatability of manually vetted labels with a buffer if applicable). A desired outcome to "approve" a ROI passed validation was a mean better than or equal to the acceptance criteria (a significant T-test showing the sample differed significantly better than the acceptance criteria threshold). Each musculoskeletal structure either used dice similarity coefficient (DSC) or volume difference (VDt) as its comparison metric, dependent on what best captured the structure's segmentation accuracy. A healthy and patient subgroup analysis split the validation dataset into two groups, healthy (athlete and healthy control) and patients (amputees, muscular dystrophy patients, and patient post-surgery). The subgroup analysis across different manufacturers; GE, Siemens, Phillips, Toshiba (AI setting 1), Canon; also showed consistent performance, indicating high generalizability of our product. The age/biologic sex subgroup analysis (males 18 – 21 years old, females 18 – 21 years old, males >21 years old, and females >21 years old) also demonstrated that each subgroup passed the predetermined acceptance criteria. For the 95% confidence interval for all structures and subgroups, see Table 1 (DSC) and Table 2 (VDt).
Table 1: The 95% confidence interval for the ROIs in AI settings 1-3 for the DICE (DSC) metric.
| ROI | Descriptive Statistics (DSC) 95% Confidence Interval | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Pathology Level Subgroup Analysis | MRI Manufacturer Subgroup Analysis | Age/Biologic Sex Subgroup Analysis | |||||||||
| Healthy | Patient | Canon | GE | Philips | Siemens | Toshiba | Male 18-21 Years | Male Over 21 Years | Female 18-21 years | Female Over 21 years | |
| adductor brevis | 0.962 -0.965 | 0.909 -0.932 | 0.978 -0.984 | 0.947 -0.967 | 0.937 -0.954 | 0.941 -0.954 | 0.956 -0.96 | 0.956 -0.969 | 0.95 -0.959 | 0.958 -0.963 | 0.911 -0.941 |
| adductor longus | 0.978 -0.979 | 0.934 -0.952 | 0.982 -0.989 | 0.965 -0.98 | 0.954 -0.972 | 0.962 -0.971 | 0.971 -0.975 | 0.97 -0.982 | 0.968 -0.975 | 0.974 -0.978 | 0.936 -0.96 |
| adductor magnus | 0.988 -0.989 | 0.963 -0.974 | 0.994 -0.997 | 0.981 -0.992 | 0.976 -0.986 | 0.979 -0.984 | 0.981 -0.985 | 0.984 -0.991 | 0.982 -0.987 | 0.987 -0.99 | 0.968 -0.978 |
[Table continues with many more rows of ROI data...]
Page 12
A better view of health
[Continuation of Table 1 with additional ROI measurements and confidence intervals]
Page 13
A better view of health
Table 2: The 95% confidence interval for the ROIs in AI settings 1-3 for the absolute volume difference metric.
| ROI | Descriptive Statistics (DSC) 95% Confidence Interval | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Pathology Level Subgroup Analysis | MRI Manufacturer Subgroup Analysis | Age/Biologic Sex Subgroup Analysis | |||||||
| Healthy | Patient | Healthy | GE | Healthy | Siemens | Healthy | Male 18-21 Years | Healthy | |
| adductor brevis | 1.09 -1.74 | 1.51 -2.29 | 0 -0.09 | 0.7 -1.52 | 0.98 -1.98 | 1.54 -2.41 | 0.6 -0.94 | 1.21 -2 | 1.19 -1.77 |
| adductor longus | 1.5 -2.26 | 1.52 -2.59 | 0.08 -2.17 | 1.08 -2.52 | 1.93 -3.18 | 1.48 -2.47 | 0.62 -1.08 | 1.37 -2.54 | 1.69 -2.45 |
[Table continues with many more rows of volume difference data...]
Page 14
A better view of health
[Continuation of Table 2 with additional volume difference measurements]
Page 15
A better view of health
[Final continuation of Table 2 with remaining volume difference measurements]
Page 16
A better view of health
Predetermined Change Control Plan
A Predetermined Change Control Plan (PCCP) has been developed in accordance with the FDA Guidance "Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions" (issued December 4, 2024). Modifications to the MuscleView AI/ML models will be made in accordance with the PCCP summarized below:
| Modification | Rationale | Modification Protocol | Impact Assessment |
|---|---|---|---|
| #1 - Retraining the AI/ML model for one or more ROIs using additional training data | To improve robustness and generalizability of the AI/ML model by training on additional annotated datasets. | Retraining of the AI/ML model with a mixture of old and new test datasets, followed by performance testing to compare the retrained model to the original version. | Benefit: Improved model generalizability across diverse cases. Risk: Introduction of bias, overfitting Mitigation: Testing data sequestration and testing on new data will ensure proper evaluation and mitigate risks of overfitting. |
| #2 - Adjustment of AI Model Settings | To improve robustness and generalizability of the AI/ML model used to identify anatomical regions by modifying a predefined set of training hyperparameters and pre/post processing methods during re-training. | Adjustment of preprocessing parameters using existing datasets, followed by performance verification against baseline metrics to ensure output consistency. | Benefit: Improved model generalizability across diverse cases. Risk: Introduction of bias, overfitting Mitigation: Predefined parameter boundaries and comparison to the original metrics mitigate risks of overfitting. |
| #3 - Addition of data points to the Virtual Control Groups | To improve the statistical significance of the Virtual Control Group (VCG) outputs used to generate scored metrics. | Recalculation of linear regression equations using additional subject data, followed by performance evaluation to ensure equivalent or improved correlation strength. | Benefit: Improved predictive accuracy Risk: Introduction of bias error due to skewed sampling Mitigation: Data collection protocols align with those contained within this submission |
Conclusion
The subject device and the predicate device have similar indications for use and are intended for similar applications in musculoskeletal imaging. The technological differences between the devices do not alter the intended use, do not raise new or different questions of safety or effectiveness, and are supported by performance data demonstrating that the subject device is at least as safe and effective as the predicate device.
18.1 – 510(k) Summary Page 12 of 12
§ 892.1000 Magnetic resonance diagnostic device.
(a)
Identification. A magnetic resonance diagnostic device is intended for general diagnostic use to present images which reflect the spatial distribution and/or magnetic resonance spectra which reflect frequency and distribution of nuclei exhibiting nuclear magnetic resonance. Other physical parameters derived from the images and/or spectra may also be produced. The device includes hydrogen-1 (proton) imaging, sodium-23 imaging, hydrogen-1 spectroscopy, phosphorus-31 spectroscopy, and chemical shift imaging (preserving simultaneous frequency and spatial information).(b)
Classification. Class II (special controls). A magnetic resonance imaging disposable kit intended for use with a magnetic resonance diagnostic device only is exempt from the premarket notification procedures in subpart E of part 807 of this chapter subject to the limitations in § 892.9.