SubtleHD-PET (1.x)
Device Facts
| Record ID | K254013 |
|---|---|
| Device Name | SubtleHD-PET (1.x) |
| Applicant | Subtle Medical, Inc. |
| Product Code | LLZ · Radiology |
| Decision Date | May 14, 2026 |
| Decision | SESE |
| Submission Type | Traditional |
| Regulation | 21 CFR 892.2050 |
| Device Class | Class 2 |
| Attributes | AI/ML, Software as a Medical Device, PCCP, Real-World Evidence, Pediatric |
Real-World Evidence
| Submission | Device | Sponsor | RWD Sources | RWE Use Summary | Key Tags |
|---|---|---|---|---|---|
| K254013 · May 14, 2026 | SubtleHD-PET (1.x) | Subtle Medical, Inc. | Retrospective clinical exams; Research study image datasets | Retrospective clinical data was used to validate the device's noise reduction performance (PSNR, SNR, RMSE, SSIM, SUVmax) and to conduct a reader study assessing image quality, diagnostic confidence, and artifact presence. | Retrospective clinical data; Performance validation; Reader study; PET imaging |
Clinical Evidence
| Study Design | Population | Comparator | Key Endpoints |
|---|---|---|---|
| Performance Validation Study; Retrospective analysis of clinical images; Follow-up/Duration: Not applicable | Patients undergoing PET scans (Brain and Whole Body); ages 2-88; various clinical conditions (e.g., Cancer, Epilepsy, Cardiomyopathy); Sample Size: 203 studies (PET-only model); 122 studies (PET/CT model) | Standard-of-care (SOC) PET images | PSNR, SNR, RMSE, SSIM, SUVmax |
| Reader Study; Retrospective reader study; Follow-up/Duration: Not applicable | Patients undergoing PET scans (Brain and Whole Body); Sample Size: 203 studies (PET-only model); 122 studies (PET/CT model) | Standard-of-care (SOC) PET images | Image quality/diagnostic confidence (5-point Likert), artifact assessment (3-point Likert), false positives/negatives |
AI Performance
| Output | Algorithm | Acceptance | Observed | Dev DS | Dev Readers | Test DS | Test Readers |
|---|---|---|---|---|---|---|---|
| PET image noise reduction | CNN with cascaded layers for convolutional filtering and nonlinearity operations | Average PSNR increase >= 1dB (p < 0.05); SNR increase significant (p < 0.05) | PET-only model: 1.935 dB average PSNR increase; PET/CT model: 1.215 dB average PSNR increase | — | — | Performance Validation and Reader Study: 203 studies (PET-only model) and 122 studies (PET/CT model) | >1 (board-certified nuclear medicine physicians) |
Indications for Use
SubtleHD-PET is an image processing software intended for noise reduction of PET images (including PET/CT and PET/MR) obtained with any kind of radiotracer (e.g. 18F-FDG).
Device Story
SubtleHD-PET is an AI/ML-based radiological image processing SaMD; enhances PET images to reduce noise. Inputs: DICOM images from PET/CT (PET and CT DICOMs) or PET/MR/PET-only (PET DICOMs). Operation: Uses convolutional neural network (CNN) with cascaded layers to perform numerical operations (convolutional filtering, linear combination, nonlinearity) to separate noise from structural components. Output: Enhanced DICOM images. Used in hospitals, clinics, and imaging centers; operated by radiologists, nuclear medicine physicians, and technologists. Software integrates via MDDS to retrieve/send images. Benefits: Improved image quality/diagnostic confidence; enables noise reduction for standard-of-care and accelerated PET exams. Healthcare providers view enhanced images on existing PACS workstations to assist in clinical decision-making.
Clinical Evidence
Performance validation and reader study using retrospective clinical data (203 studies for PET-only, 122 for PET/CT). Metrics: PSNR, SNR, RMSE, SSIM, SUVmax. Results: Average PSNR increase ≥ 1dB (p<0.05); significant SNR increase (p<0.05). Reader study (board-certified nuclear medicine physicians) confirmed image quality/diagnostic confidence scores ≥ 3 (5-point scale) and no introduced artifacts (score ≥ 2, 3-point scale). No false positives/negatives introduced.
Technological Characteristics
Software-only device operating on off-the-shelf hardware (Linux). Uses convolutional neural network (CNN) for image enhancement. Processes DICOM-compliant image data. Two models: PET-only and PET/CT. Connectivity via MDDS/PACS integration. ISO 13485:2016 and MDSAP certified QMS.
Indications for Use
Indicated for noise reduction of PET images (including PET/CT and PET/MR) in patients of all ages undergoing PET imaging with any radiotracer.
Regulatory Classification
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.
Special Controls
*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).
Predicate Devices
Submission Summary (Full Text)
{0}
FDA U.S. FOOD & DRUG ADMINISTRATION
May 14, 2026
Subtle Medical, Inc.
Adam Heroux
Consultant
883 Santa Cruz Ave.
Suite 205
Menlo Park, California 94025
Re: K254013
Trade/Device Name: SubtleHD-PET (1.x)
Regulation Number: 21 CFR 892.2050
Regulation Name: Medical Image Management And Processing System
Regulatory Class: Class II
Product Code: LLZ
Dated: April 13, 2026
Received: April 13, 2026
Dear Adam Heroux:
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.
U.S. Food & Drug Administration
10903 New Hampshire Avenue
Silver Spring, MD 20993
www.fda.gov
{1}
K254013 - Adam Heroux
Page 2
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 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 Management System Regulation (QMSR) (21 CFR Part 820), which includes, but is not limited to, ISO 13485 clause 7.3 (Design controls), ISO 13484 clause 8.3 (Nonconforming product), and ISO 13485 clause 8.5 (Corrective and preventative action). Please note that regardless of whether a change requires premarket review, the QMSR requires device manufacturers to review and approve changes to device design and production (ISO 13485 clause 7.3 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 Management System Regulation (QMSR) (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}
K254013 - Adam Heroux
Page 3
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.
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
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 | | |
| --- | --- | --- |
| Please type in the marketing application/submission number, if it is known. This textbox will be left blank for original applications/submissions. | K254013 | ? |
| Please provide the device trade name(s). | | ? |
| SubtleHD-PET (1.x) | | |
| Please provide your Indications for Use below. | | ? |
| SubtleHD-PET is an image processing software intended for noise reduction of PET images (including PET/CT and PET/MR) obtained with any kind of radiotracer (e.g. 18F-FDG). | | |
| Please select the types of uses (select one or both, as applicable). | ☑ Prescription Use (21 CFR 801 Subpart D) ☐ Over-The-Counter Use (21 CFR 801 Subpart C) | ? |
{4}
SubtleHD-PET 510(k) Summary
K254013
Table 1. Contact Details and Device Name
| Date Summary Prepared: | 13 April 2026 |
| --- | --- |
| Contact Details: | |
| Company Name: | Subtle Medical, Inc. |
| Company Address: | 883 Santa Cruz Ave, Suite 205 Menlo Park, CA 94025 United States |
| Applicant Name: | Mr Ajit Shankaranarayanan |
| Application Telephone: | 650-448-4285 |
| Applicant Email: | ajit@subtlemedical.com |
| Device Name: | |
| Device Trade Name | SubtleHD-PET (1.x) |
| Common Name: | Medical image management and processing system |
| Classification Name: | System, Image Processing, Radiological |
| Regulation Number: | 892.2050 |
| Product Code: | LLZ |
| Device Class: | Class II |
| Legally Marketed Predicate Devices: | Primary Predicate #: K213140 Primary Predicate Name: Claritas iPET Primary Predicate Legal Manufacturer: Claritas Healthtech Pte, Ltd. Secondary Predicate #: K211964 Secondary Predicate Name: SubtlePET Secondary Predicate Legal Manufacturer: Subtle Medical, Inc. |
# Device Description Summary
SubtleHD-PET is an AI/ML radiological image processing Software as a Medical Device (SaMD). SubtleHD-PET uses deep learning algorithms to enhance medical images acquired by PET scanners to reduce image noise. As it only processes images for the end user, the device has no user interface. SubtleHD-PET is intended to be operated by radiologists, nuclear medicine physicians, and technologists at an imaging center, clinic, or hospital through an Medical Device Data System (MDDS) to prepare and send input DICOM images and retrieve
{5}
output DICOM images. SubtleHD-PET does not have a user interface or interact with the PET scanner. The SubtleHD-PET software can be used with PET images acquired as part of standard-of-care and accelerated PET exams as the input. SubtleHD-PET has two models: PET/CT model which takes the PET and CT DICOMs (from PET/CT) as the input and PET-only model which takes the PET DICOMs (from PET/MR or PET/CT) only as the input. The outputs are the corresponding images with enhanced image quality. Original DICOM images are passed onto the SubtleHD-PET software as an input argument and the enhanced images are saved in the designated location prescribed when running the SubtleHD-PET software.
The SubtleHD-PET software implements an image enhancement algorithm using a convolutional neural network. Input images are enhanced by running through the CNN that consists of cascaded layers to perform numerical operations, including convolutional filtering, linear combination, and nonlinearity operations.
## Intended Use / Indications for Use
SubtleHD-PET is an image processing software intended for noise reduction of PET images (including PET/CT and PET/MR) obtained with any kind of radiotracer (e.g. 18F-FDG).
## Intended Use / Indications for Use Comparison
SubtleHD-PET and its predicates are all intended to denoise PET images of various input protocols, imaging scanners, and radiotracers.
## Technological Comparison
SubtleHD-PET 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 neural 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. 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.
SubtleHD-PET v1.x is classified as Class II with product code LLZ through the FDA using Claritas iPET and SubtlePET as predicates. Below is a preliminary comparison of SubtleHD-PET and the predicates.
{6}
Table 2. Comparison of Technological Characteristics
| Comparison | SubtleHD-PET v1.x (Subject Device) | Claritas iPET (Predicate) (R213140) | SubtlePET v2.x (Predicate) (R211964) | Noted Differences |
| --- | --- | --- | --- | --- |
| Indications for Use | SubtleHD-PET is an image processing software intended for noise reduction of PET images (including PET/CT and PET/MR) obtained with any kind of radiotracer (e.g. fluorodeoxyglucose (FDG)). | Claritas iPET is an image processing software intended for use by radiologists and nuclear medicine physicians for noise reduction, sharpening, and resolution improvement of PET images (including PET/CT and PET/MRI) obtained with any kind of radionuclides, e.g. fluorodeoxyglucose (FDG). Enhanced images will be saved in DICOM files and exist in conjunction with original images. | SubtlePET is an image processing software intended for use by radiologists and nuclear medicine physicians for transfer, storage, and noise reduction of fluorodeoxyglucose (FDG), amyloid, 18F-DOPA, 18F-DCFPyL, Ga-68 Dotatate, and Ga-68 PSMA radiotracer PET images. | Claritas iPET has sharpening and resolution enhancement. SubtleHD-PET v1.x claims any radiotracer, similar to Claritas iPET rather than listing the radiotracers like SubtlePET v2.x. Subject and predicate devices are both image enhancement algorithms for denoising. Substantially equivalent. |
| Intended Use | Image enhancement system which is an image processing software for image enhancement of PET images including PET/CT and PET/MRI. | Same | Same | Same |
| Workflow | 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, PET device, workstation, and more). Enhanced images coexist with the original images. | Same in case of the PACS integrated version. The stand-alone version can input the slices of the PET, MR or CT scans as DICOM files, interactively visualizes the input and the output data, and saves the enhanced volume in DICOM files. | Same | Claritas iPET has a user interface for their stand-alone version. Substantially equivalent. |
| Product Code | LLZ | LLZ | KPS, LLZ | Same |
| Physical Characteristics | Software package that operates on off-the-shelf hardware | Software package that operates on a virtual machine (VM) or deployed on a local computer. | Software package that operates on off-the-shelf hardware | Same |
| Intended User | Radiologists and nuclear medicine physicians | Same | Same | Same |
{7}
| Comparison | SubtleHD-PET v1.x (Subject Device) | Claritas iPET (Predicate) (K213140) | SubtlePET v2.x (Predicate) (K211964) | Noted Differences |
| --- | --- | --- | --- | --- |
| Intended Location | Medical facility (hospitals, clinics, imaging center, etc.) | Same in case of the PACS integrated version. The stand-alone version runs on the client computer. | Medical facility (hospitals, clinics, imaging center, etc.) | Same |
| Modalities | Multi-modality; specifically processes PET, PET/CT and PET/MR images | Same | Same | Same |
| Input(s) | Requires either (1) PET and CT DICOM series from PET/CT scanner, (2) PET DICOM series from only PET or PET/MR scanner | Processes PET DICOM from either PET/MR or PET/CT. | Processes PET DICOM from either PET/MR or PET/CT. | SubtlePET only processed the PET DICOM series from PET/CT scanner. SubtleHD-PET uses input from the PET and CT DICOM series to process from a PET/CT scanner. Substantially equivalent. |
| Operating System / Computer | Linux | Windows/Linux | Linux | Same |
| Rx or OTC | Rx | Same | Same | Same |
| User Interface | None | None - when integrated into existing PACS workstations, viewed on existing PACS workstation. A user interface for stand-alone version visualizing 2D slices and 3D rendering for demo and research purposes. | None | Claritas iPET has a user interface for their stand-alone version. Substantially equivalent. |
| DICOM Standard Compliance | The software processes DICOM compliant image data | Same | Same | Same |
{8}
| Comparison | SubtleHD-PET v1.x (Subject Device) | Claritas iPET (Predicate) (K213140) | SubtlePET v2.x (Predicate) (K211964) | Noted Differences |
| --- | --- | --- | --- | --- |
| Image Enhancement Algorithm Description | The software employs 2 different convolutional neural network models in a pixel's neighborhood to compute the value for each pixel. One model uses both PET and CT as model inputs if PET and CT images are available. The other model uses PET as the only model input if CT image is not available. The software predicts the noise components and structural components in PET images. The software separates these components, which enhances the structure while simultaneously reducing the noise. | The image enhancement algorithm is a modification of the non-local means algorithm where the filtering weights can be obtained from higher resolution and lower noise voxel arrays obtained with other modalities, i.e. CT or MR. The resolution of the target is at least the maximum of the combined modalities, but may be higher. | The software employs a convolutional neural network-based method in a pixel's neighborhood to generate the value for each pixel. Using a residual learning approach, the software predicts the noise components and structural components. The software separates these components, which enhances the structure while simultaneously reducing the noise | Subject and predicate devices use image enhancement algorithms as their core technology. Substantially equivalent. |
| Model Architecture | Two Models: PET-CT model, PET-only model | Unknown | One Model: PET model | SubtleHD-PET has two models while SubtlePET only had one model. Subject and predicate devices use PET DICOMS as the model architecture. Substantially equivalent. |
## Predetermined Change Control Plan (PCCP)
SubtleHD-PET 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-PET PCCP plans for changes to add additional denoising mix factor levels to allow for configuration that best fits each user's perceived image quality preferences. Additionally, SubtleHD-PET plans for changes to add additional data to the performance and training datasets for generalizability. This includes additional ages, scanner vendors/models, anatomies, detector types, and radiotracers.
{9}
These modifications 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 Instructions for Use (IFU). 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.
- Performance Validation testing utilizing retrospective clinical data to demonstrate the software enhanced image quality in PET images via a reduction of noise. These tests passed.
- A Reader Study utilizing retrospective clinical data to demonstrate the software enhanced image quality in PET images via a reduction of noise. These tests passed.
## Performance Validation & Reader Study Dataset Characteristics:
To represent the patient population and use of PET in the field, the SubtleHD-PET performance and reader study validation test dataset consists of:
- Brain and Whole Body anatomical regions.
- PET Only Model: Brain (29%), Whole Body (71%)
- PET/CT Model: Brain (20%), Whole Body (80%)
- BGO, LBS, LSO, LYSO, and SiPM detector types.
- 18F-Choline, 18F-DCFPyL, 18F-DOPA, 18F-FDG, 18F-Florbetaben, 18F-Florbetapir, 68Ga-Dotatate, 18F-Flutemetamol, and 68Ga-PSMA radiotracers.
- For 18F-FDG, at least 20% of the dataset has a high Body Mass Index (BMI) (>= 30).
- GE, Philips, Siemens, and United Imaging Healthcare imaging scanner vendors.
- PET Only Model: GE (54%), Philips (7%), Siemens (24%), and United Imaging Healthcare (15%)
- PET/CT Model: GE (54%), Philips (7%), Siemens (14%), and United Imaging Healthcare (25%)
- Pediatric and Adult subject ages (ranging from 2 to 88 years old).
- PET Only Model: 87% Adult, 13% Pediatric
- PET/CT Model: 85% Adult, 15% Pediatric
- Subject sex.
- PET Only Model: 43% Male
- PET/CT Model: 48% Male
- Standard-of-care PET data and Low-count PET data from accelerated scans (25%-90% acceleration equating to 10%-75% count PET data).
Various clinical conditions are present in the performance dataset, as per reports accompanying the images from the sources, including: Cancer Follow-Up, Cardiomyopathy, Epilepsy, H&N
{10}
Cancer, Lymphoma, and Seizure. This selection criteria represents a well characterized clinically-relevant reference dataset.
To show the performance of the device was not hindered by site variability, in the validation dataset, data was selected from sources not included in the training dataset. The majority of performance data comes from sources in the United States. Independence between the training and test datasets was ensured by using non-overlapping datasets, with no shared subjects, studies, or images between training and testing.
Even though all the primary endpoints metrics for performance have passed, analysis of the device's performance across the various subgroups in the performance dataset showed slight variations. As an example, Figure 1 below shows the differences in average PSNR and SNR improvements between the different subgroups of the PET-only model results. While all the subgroups demonstrate significant improvements from SubtleHD-PET processing, subgroups like Brain, 18F-Flutemetamol & 18F-DOPA (for average PSNR) and 68-Ga-PSMA & 18F-DOPA (for SNR increase) showed lesser performance gains. Therefore, we advise the users to expect slight performance differences based on various confounding factors.
Figure 1: Plot showing the device's performance differences in average PSNR and SNR improvements between the different subgroups of the PET-only model results.
{11}
We observed statistically significant differences in performance metrics (PSNR, SNR, RMSE, SSIM) between the "Training Site" and "Not Training Site" subgroups, but these differences are reduced and become non-significant when we balance the datasets of the two cohorts. Therefore, we contend that this difference is not a result of model bias toward specific clinical sites, but rather a reflection of the imbalance in anatomy and radiotracer types between the two cohorts.
Various reconstruction methods are present in the performance dataset, as per the metadata accompanying the images from the sources, including: BLOB-OS-TF, FORE - OSEM 3D, OSEM 2D, OSEM 3D, OSEM 3D + PSF, OSEM 3D + PSF + TOF, OSEM 3D + TOF, Q.Clear, VPFX, VPFX + PSF, VPHD and VPHD + PSF. While some reconstruction methods receive greater benefits from SubtleHD-PET processing, all 12 reconstruction methods showed a significant improvement across PSNR, RMSE, SSIM, and SNR. In the $\chi^2$ analysis, reduced performance was observed with VPFX + PSF reconstruction. Therefore, we suggest caution while using SubtleHD-PET for VPFX+PSF.
{12}
Distribution by Reconstruction Method
Figure 1. Box plot of device performance metrics with respect to reconstruction method. Across all reconstruction methods, quantitative metrics showed a significant improvement after processing with SubtleHD-PET.
# Performance Validation:
This study utilized retrospectively acquired human data acquired as part of research studies or clinical exams from various institutions. For each dataset, four slices were selected from the upper quartile of slices, 2nd quartile, 3rd quartile, and lowest quartile by a Subtle Medical employee with an MD and/or a PhD in a clinically-relevant field for the calculation of PSNR, RMSE, and SSIM. For the calculation of SUVmax and SNR, 2 rectangular ROIs are drawn in an area with focal uptake (lesion) and background regions, respectively, by a Subtle Medical
{13}
employee with an MD and/or a PhD in a clinically-relevant field. In total 203 studies were used to validate the PET-only model and 122 studies used to validate the PET/CT model acceptance criteria in Table 3 at three different mix factors (low, default, high).
The following are the endpoints, acceptance criteria, results, and conclusions from the SubtleHD-PET:
Table 3. Performance Validation Summary
| Endpoint | Acceptance Criteria | Result /Conclusion |
| --- | --- | --- |
| Denoising (Peak Signal-to-Noise Ratio) Primary Endpoint | The average increase in PSNR is ≥ 1dB and the PSNR increase is significant at the p < 0.05 level based on two-tailed paired t-test. | PASS: The average increase in PSNR was ≥ 1dB. PSNR increase was significant at the p < 0.05 level based on two-tailed paired t-test. |
| Denoising (Peak Signal-to-Noise Ratio) Exploratory Endpoint | Report the range and average PSNR improvement. | Average PSNR improvement for the PET-only model in the default configuration is 1.935 dB, with a range of up to 5.964 dB. Average PSNR improvement for the PET/CT model in the default configuration is 1.215 dB, with a range of up to 6.065 dB. |
| Denoising (Signal-to-Noise Ratio) Primary Endpoint | SNR increase is significant at the p < 0.05 level based on one-tailed paired t-test. | PASS: The average increase in SNR was significant at the p < 0.05 level based on one-tailed paired t-test. |
| Denoising (Signal-to-Noise Ratio) Exploratory Endpoint | Report the range and average SNR improvement. | Average SNR improvement for the PET-only model in the default configuration is 3.179 dB, with a range of up to 26.938 dB. Average SNR improvement for the PET/CT model in the default configuration is 4.590 dB, with a range of up to 29.876 dB. |
{14}
| Endpoint | Acceptance Criteria | Result /Conclusion |
| --- | --- | --- |
| Denoising (Root Mean Square Error) Exploratory Endpoint | Report the range and average reduction in RMSE. | The average reduction in RMSE for the PET-only model for the default configuration is -0.040, with a range of -0.172 to 0.009. The average reduction in RMSE for the PET/CT model for the default configuration is -0.028, with a range of -0.150 to 0.070. |
| Image Quality (Structural Similarity Index Measure) Exploratory Endpoint | Report the range and average increase in SSIM. | The average SSIM for the PET-only model for the default configuration is 0.019, with a range of -0.018 to 0.106. The average SSIM for the PET/CT model for the default configuration is 0.009, with a range of -0.064 to 0.057. |
| Image Quality (SUVmax) Exploratory Endpoint | Report difference in SUVmax. | The following is the average difference in SUVmax for the PET-only model for the default configuration with each tracer: • 18F-FDG: 0.94% • 18F-DCFPyL: -0.66% • 68Ga-Dotatate: -1.53% • 18F-Choline: -1.03% • 68Ga-PSMA: 2.32% • 18F-DOPA: 2.25% The following is the average difference in SUVmax for the PET/CT model for the default configuration with each tracer: • 18F-FDG: -2.17% • 18F-DCFPyL: -6.74% • 68Ga-Dotatate: 1.00% |
## Reader Study:
This study utilized retrospectively acquired human data obtained as part of research studies or clinical exams from various institutions. The data set included 203 studies (203 SubtleHD-PET, 203 SOC) used to validate the PETonly model and 122 studies (122 SubtleHD-PET, 122 SOC) used to validate the PET/CT model. The datasets were presented side-by-side with standard-of-care and SubtleHD-PET enhanced images.
Board-certified nuclear medicine physicians evaluated SubtleHD-PET-enhanced images for image quality/diagnostic confidence using a 5-point Likert scale, assessed artifacts using a 3-point Likert scale, and evaluated false positives and false negatives using lesion counts.
The following are the endpoints, acceptance criteria, results, and conclusions from the SubtleHD-PET Reader Study:
{15}
Table 4. Reader Study Summary
| Endpoint | Acceptance Criteria | Result / Conclusion |
| --- | --- | --- |
| Image Quality (IQ) / Diagnostic Confidence Primary Endpoint | The average scores across all readers of the SubtleHD-PET enhanced images are an average of 3 or better (enhanced images are the same or better than original). | PASS: SubtleHD-PET image quality / diagnostic confidence is the same or better than the standard-of-care images. |
| FP/FN Primary Endpoint | No false positives or false negatives are detected in the SubtleHD-PET images. | PASS: SubtleHD-PET does not introduce any false positives or false negatives. |
| Artifacts Evaluation Secondary Endpoint | SubtleHD-PET enhanced images have an average of 2 or better (SubtleHD-PET does not introduce artifacts that could impact diagnosis in any image) | PASS: SubtleHD-PET does not introduce artifacts that could impact diagnosis in any image. |
# Conclusion:
These results demonstrate that SubtleHD-PET is substantially equivalent to the predicate devices.
