DeepXray Spina
Device Facts
| Record ID | K253192 |
|---|---|
| Device Name | DeepXray Spina |
| Applicant | Alpha Intelligence Manifolds, Inc. |
| Product Code | SAO · Radiology |
| Decision Date | Jun 10, 2026 |
| Decision | SESE |
| Submission Type | Traditional |
| Regulation | 21 CFR 892.1171 |
| Device Class | Class 2 |
| Attributes | AI/ML, Software as a Medical Device, Real-World Evidence |
Real-World Evidence
| Submission | Device | Sponsor | RWD Sources | RWE Use Summary | Key Tags |
|---|---|---|---|---|---|
| K253192 · Jun 10, 2026 | DeepXray Spina | Alpha Intelligence Manifolds, Inc. | Retrospective multi-center clinical imaging archives; Routine clinical DXA scan results | Retrospective clinical data were used to validate the performance of the AI algorithm against DXA-derived ground truth across diverse patient populations and clinical sites. | Retrospective study; Multi-center cohort; DXA ground truth; Clinical performance validation |
Clinical Evidence
| Study Design | Population | Comparator | Key Endpoints |
|---|---|---|---|
| Retrospective Multi-Center Clinical Performance Study; Retrospective, multi-center observational study; Follow-up/Duration: Not applicable | Adults aged 50 years and older; Sample Size: 1,328 total patients (Test Set-1: 577; Test Set-2: 159; Test Set-3: 592); Number of Sites: Multi-center (Taiwan and U.S.) | Not applicable for this study | AUC, sensitivity, and specificity for identifying patients with T-score ≤ -2.5 |
AI Performance
| Output | Algorithm | Acceptance | Observed | Dev DS | Dev Readers | Test DS | Test Readers |
|---|---|---|---|---|---|---|---|
| Low bone mineral density (BMD) notification | Locked AI/ML algorithms using radiogrammetry | 95% CI lower bound of IoU ≥ 0.80 for ROI localization | AUC = 0.955, sensitivity = 77.0%, specificity = 93.6% | — | — | Test Set-1 (Taiwanese Multi-Center Cohort): 577 subjects; Test Set-2 (Far Eastern Memorial Hospital, Taiwan): 159 subjects; Test Set-3 (U.S. Multi-Center Cohort): 592 subjects. | — |
Indications for Use
DeepXray Spina is a software application intended for use opportunistically with standard frontal radiographs of the lumbar spine or KUB (kidney, ureter, and bladder) performed in patients aged 50 years and older. DeepXray Spina provides a notification in the form of a report to aid radiologists and/or physician interpreters in identifying patients with possible low bone mineral density (BMD) at L1–L4 to prompt a clinical assessment of bone health. DeepXray Spina should not be used to rule out low BMD. Radiologists and referring clinicians should follow recommended practices for screening and assessment, regardless of the absence of a DeepXray Spina report. Input Restriction: DeepXray Spina is intended for use with standard frontal radiographs of the lumbar spine or KUB that are suitable for L1–L4 BMD analysis. The device is not intended for use with images in which one or more L1–L4 vertebrae contains: (a) metallic spinal implants or prostheses (e.g., pedicle screws, rods, cages, or total disc replacements); (b) vertebral cement augmentation (e.g., kyphoplasty or vertebroplasty); (c) severe vertebral deformity, fracture, or other structural abnormality that would preclude a valid DXA-based L1–L4 BMD measurement; or (d) extraspinal medical implants overlying the L1–L4 measurement region (e.g., abdominal aortic stent grafts or vascular grafts).
Device Story
Software-only device (SaMD) for opportunistic BMD screening; inputs standard frontal lumbar spine or KUB radiographs (CR/DR). Employs locked AI/ML algorithms to perform radiogrammetry; estimates L1–L4 BMD based on structural/textural features; converts to T-score using NHANES reference; thresholds at T ≤ -2.5. Operates as add-on to DeepXray system (K223621) within institutional Linux server/Docker containers. Outputs binary classification ('Low BMD' or 'Not Suspected') via web report or DICOM-format report to PACS. Includes automated QC for anatomical/technical exclusion conditions. Radiologists use report and ROI overlay to verify findings and prompt clinical bone health assessment. Benefits include early identification of osteoporosis risk in patients undergoing routine imaging.
Clinical Evidence
Retrospective multi-center study (n=1,328 total across three cohorts). Ground truth: DXA-derived L1–L4 BMD (± 6 months). Primary endpoint (T-score ≤ -2.5): AUC 0.955 (95% CI: 0.942–0.967), sensitivity 77.0%, specificity 93.6%. Subgroup analyses performed for age, sex, BMI, ethnicity, and X-ray equipment. Reproducibility (n=194) showed 3.4% CV and kappa 0.703–0.706.
Technological Characteristics
Software-only (SaMD); locked AI/ML algorithm; radiogrammetry principle. Deployed in Docker containers on local Linux server. Integrates with existing DICOM/PACS infrastructure. Automated QC function for image exclusion. No hardware components.
Indications for Use
Indicated for patients aged 50+ undergoing standard frontal lumbar spine or KUB radiographs to identify potential low bone mineral density (BMD) at L1–L4. Contraindicated for patients with metallic spinal implants, vertebral cement augmentation, severe vertebral deformity/fracture, or extraspinal implants overlying the L1–L4 region.
Regulatory Classification
Identification
Radiology software for opportunistic evaluation of low bone mineral density. This device is software which opportunistically assesses radiological images to estimate bone mineral density (BMD) intended to assist in a healthcare professional's decision to evaluate patients for possible low BMD within a bone health screening program. The software employs an algorithm that estimates BMD using eligible radiological image data obtained for other clinical purposes.
Special Controls
In combination with the general controls of the FD&C Act, radiology software for opportunistic evaluation of low bone mineral density is subject to the following special controls:
Predicate Devices
- 16 Bit Rho (DEN230023)
Reference Devices
- DeepXray system (K223621)
Submission Summary (Full Text)
{0}
**FDA** **U.S. FOOD & DRUG**
ADMINISTRATION
June 10, 2026
Alpha Intelligence Manifolds, Inc.
% Qingzong TSENG
VP, Data Science
2f, # 170, Zhonghe Rd., Zhonghe Dist.
NEW TAIPEI CITY, 235068
TAIWAN
Re: K253192
Trade/Device Name: DeepXray Spina
Regulation Number: 21 CFR 892.1171
Regulation Name: Radiology software for opportunistic evaluation of low bone mineral density
Regulatory Class: Class II
Product Code: SAO
Dated: September 26, 2025
Received: May 11, 2026
Dear Qingzong TSENG:
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}
K253192 - Qingzong TSENG
Page 2
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 13485 clause 8.3 (Nonconforming product), ISO 13485 clause 8.5.2 (Corrective action), and ISO 13485 clause 8.5.3 (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 ISO 13485 clause 7.5) and document changes and approvals in the Medical Device File (ISO 13485 clause 4.2.3).
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 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-
{2}
K253192 - Qingzong TSENG
Page 3
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,
Lu Jiang, Ph.D.
Assistant Director
Diagnostic X-Ray Systems Team
DHT8B: Division of Radiological Imaging
Devices and Electronic Products
OHT8: Office of Radiological Health
Office of Product Evaluation and Quality
Center for Devices and Radiological Health
Enclosure
{3}
DEPARTMENT OF HEALTH AND HUMAN SERVICES
Food and Drug Administration
Form Approved: OMB No. 0910-0120
Expiration Date: 07/31/2026
See PRA Statement below.
# Indications for Use
510(k) Number (if known)
K253192
Device Name
DeepXray Spina
Indications for Use (Describe)
DeepXray Spina is a software application intended for use opportunistically with standard frontal radiographs of the lumbar spine or KUB (kidney, ureter, and bladder) performed in patients aged 50 years and older. DeepXray Spina provides a notification in the form of a report to aid radiologists and/or physician interpreters in identifying patients with possible low bone mineral density (BMD) at L1–L4 to prompt a clinical assessment of bone health. DeepXray Spina should not be used to rule out low BMD. Radiologists and referring clinicians should follow recommended practices for screening and assessment, regardless of the absence of a DeepXray Spina report.
Input Restriction: DeepXray Spina is intended for use with standard frontal radiographs of the lumbar spine or KUB that are suitable for L1–L4 BMD analysis. The device is not intended for use with images in which one or more L1–L4 vertebrae contains: (a) metallic spinal implants or prostheses (e.g., pedicle screws, rods, cages, or total disc replacements); (b) vertebral cement augmentation (e.g., kyphoplasty or vertebroplasty); (c) severe vertebral deformity, fracture, or other structural abnormality that would preclude a valid DXA-based L1–L4 BMD measurement; or (d) extraspinal medical implants overlying the L1–L4 measurement region (e.g., abdominal aortic stent grafts or vascular grafts).
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 (8/23)
Page 1 of 1
PSC Publishing Services (301) 443-6740 EF
{4}
K253192
# 510(k) SUMMARY
# **DeepXray Spina
Alpha Intelligence Manifolds, Inc.**
| **Applicant:** | Alpha Intelligence Manifolds, Inc. 2F, No.170, Zhonghe Road, Zhonghe District, New Taipei City, 235068, Taiwan Telephone: +886-2-2240-6570 |
| --- | --- |
| **Date Prepared:** | Jun 9, 2026 |
| **Device Name:** | DeepXray Spina |
| **Regulation Number:** | 892.1171 |
| **Product Code:** | SAO |
| **Classification Name:** | Radiology software for opportunistic evaluation of low bone mineral density |
| **Device Class:** | Class II |
| **Review Panel:** | Radiology |
| **Predicate Devices:** | 16 Bit Rho (DEN230023) |
# **Device Description**
DeepXray Spina is a software-only medical device (SaMD) that identifies patients with Low BMD (T-score $\leq -2.5$) from standard frontal radiographs of the lumbar spine or KUB (kidney, ureter, and bladder), acquired as Computed Radiography (CR) or Digital Radiography (DR). The device does not interact with patients directly, does not modify the acquisition system, and does not control any life-sustaining devices.
The scientific principle underlying DeepXray Spina is radiogrammetry: bone mineral density (BMD) is inferred from the structural and textural features of standard radiographs rather than measured by absorptiometric principles. The device employs locked artificial intelligence / machine-learning (AI/ML) algorithms, trained and validated against dual-energy X-ray absorptiometry (DXA) ground truth, to estimate the aggregate L1–L4 BMD, matching the same
{5}
anatomical region and aggregation convention used in standard DXA lumbar spine measurements. The estimated BMD is then converted internally to a T-score against the U.S. National Health and Nutrition Examination Survey (NHANES) young-adult non-Hispanic white female reference population, as recommended by the International Society for Clinical Densitometry (ISCD), and thresholded at −2.5 to produce the user-facing binary classification. The intermediate L1–L4 BMD and T-score values are not displayed.
The device output is a binary classification: “Low BMD” (internal L1–L4 T-score ≤ −2.5) or “Not Suspected” (internal L1–L4 T-score > −2.5). A DeepXray Spina report is generated for positive cases. For cases where the algorithm outputs a negative result, no report is generated, and neither the radiologist nor the referring physician receives any device output. Each generated report displays a vertebral region-of-interest (ROI) overlay on the original radiograph, enabling the clinician to visually verify anatomical localization and the absence of severe artifacts before acting on the result. An automated Quality-Control (QC) function detects technical and anatomical exclusion conditions within the L1–L4 region and flags input images with potential quality issues.
DeepXray Spina is installed as an add-on module to the cleared DeepXray system (K223621). The software is deployed within Docker containers on a Linux server inside the institution’s local network and shares the same Input & Report Server (IRS) as DeepXray. The IRS manages DICOM inputs, performs basic tag and image-type filtering, and delivers outputs through either a web-based report accessible from a browser or a static DICOM-format report returned to the institution’s PACS. The DeepXray Spina pipeline runs inside the Inference Engine (IE) of the existing DeepXray system and includes anatomical ROI localization, landmark identification, BMD estimation, automated QC checks, and report generation.
## Indications for Use
DeepXray Spina is a software application intended for use opportunistically with standard frontal radiographs of the lumbar spine or KUB (kidney, ureter, and bladder) performed in patients aged 50 years and older. DeepXray Spina provides a notification in the form of a report to aid radiologists and/or physician interpreters in identifying patients with possible low bone mineral density (BMD) at L1–L4 to prompt a clinical assessment of bone health. DeepXray Spina should not be used to rule out low BMD. Radiologists and referring clinicians should follow recommended practices for screening and assessment, regardless of the absence of a DeepXray Spina report.
Input Restriction: DeepXray Spina is intended for use with standard frontal radiographs of the lumbar spine or KUB that are suitable for L1–L4 BMD analysis. The device is not intended for use with images in which one or more L1–L4 vertebrae contains: (a) metallic spinal implants or prostheses (e.g., pedicle screws, rods, cages, or total disc replacements); (b) vertebral cement augmentation (e.g., kyphoplasty or vertebroplasty); (c) severe vertebral deformity, fracture, or
{6}
other structural abnormality that would preclude a valid DXA-based L1–L4 BMD measurement; or (d) extraspinal medical implants overlying the L1–L4 measurement region (e.g., abdominal aortic stent grafts or vascular grafts).
## Comparison of Technological Characteristics
DeepXray Spina has the same general intended use and similar indications for use as the predicate device, Rho (16 Bit, Inc., DEN230023). Both are software-only applications that employ locked AI/ML algorithms trained on DXA reference data to identify patients who may have low BMD, and both generate a single binary output. The principal differences are: (a) DeepXray Spina operates exclusively on the frontal lumbar-spine / KUB view, whereas Rho accepts a broader range of planar radiographs; (b) DeepXray Spina is image-only and does not incorporate demographic variables, whereas Rho does; and (c) DeepXray Spina's binary output is at the WHO osteoporosis threshold (T ≤ −2.5), whereas Rho's binary output is at the WHO low-bone-mass threshold (T < −1). Neither difference alters the intended use or principle of operation, and neither raises new or different questions of safety and effectiveness. A side-by-side comparison of key features is provided in Table 1.
Table 1. Device–Predicate Comparison.
| Device | DeepXray Spina (Subject Device) | Rho (DEN230023, Predicate) |
| --- | --- | --- |
| Classification Name & Product Code | Radiology software for opportunistic evaluation of low BMD (SAO) | Radiology software for opportunistic evaluation of low BMD (SAO) |
| Regulation Number | 21 CFR 892.1171 | 21 CFR 892.1171 |
| Device Class | Class II | Class II (De Novo) |
| Intended Use | Opportunistic aid to identify patients who may have low BMD from frontal L-spine / KUB radiographs | Opportunistic aid to identify patients who may have low BMD from standard planar radiographs |
| Intended Patient Population | Adults ≥ 50 years | Adults ≥ 50 years |
| Input Imaging Region(s) | Frontal lumbar spine or KUB (anatomically matched to L1–L4 reference) | Lumbar spine, thoracic spine, chest, pelvis, knee, or hand/wrist |
| Algorithm Type | Locked AI/ML; image-only (no demographic inputs) | Locked AI/ML; incorporates demographic inputs |
| Cleared Device Output | Binary: suspected Low BMD (L1–L4 T-score ≤ −2.5) vs. not suspected (L1–L4 T-score > −2.5) | Binary: suspected low BMD vs. not suspected, at T-score < −1 |
{7}
## Nonclinical Performance Data
The following non-clinical performance testing was conducted in support of substantial equivalence, in accordance with 21 CFR 807.92(b)(1):
- Software Verification and Validation Testing: conducted in accordance with the FDA guidance Content of Premarket Submissions for Device Software Functions and IEC 62304:2006/AMD 1:2015. All unit, integration, and system-level tests met pre-specified acceptance criteria. The device is released as a locked AI/ML model.
- Cybersecurity Testing: conducted in accordance with the FDA guidance Cybersecurity in Medical Devices: Quality System Considerations and Content of Premarket Submissions. Testing included static code analysis, dynamic code analysis, fuzz testing and penetration testing.
## Clinical Performance Data
A retrospective, multi-center clinical performance study was conducted to validate the safety and effectiveness of DeepXray Spina against DXA-derived ground truth. Three independent test datasets, fully disjoint from the algorithm-development dataset, were analyzed: Test Set-1 (Taiwanese Multi-Center Cohort, n = 577), Test Set-2 (Far Eastern Memorial Hospital, Taiwan, n = 159), and Test Set-3 (U.S. Multi-Center Cohort from OneMedNet and Gradient Health, n = 592). The ground truth was the L1–L4 BMD measured by DXA within ± 6 months of the radiograph; T-scores were derived from the NHANES young-adult non-Hispanic white female reference.
| Characteristic | Test Set-1 (Taiwanese Multi-Center Cohort) | Test Set-2 (FEMH Cohort) | Test Set-3 (U.S. Multi-Center Cohort) |
| --- | --- | --- | --- |
| Number of Subjects (n) | 577 | 159 | 592 |
| Age: Mean (SD), years | 61.61 (11.27) | 65.96 (10.99) | 67.55 (10.30) |
| Age: Range, years | (21.39, 96.01) | (31.00, 91.00) | (40.16, 89.83) |
| Sex: (Female/Male), n (%) | F: 490 (84.92%), M: 87 (15.08%) | F: 129 (81.13%), M: 30 (18.87%) | F: 488 (82.43%), M: 104 (17.57%) |
| Body Mass Index: Mean (SD), kg/m² | 24.02 (3.80) | 23.74 (4.30) | 27.00 (5.69) |
| Ethnic Groups: n (%) | Asian: 577 (100.00%) | Asian: 159 (100.00%) | White: 327 (55.24%), Hispanic: 120 (20.27%), Black: 78 (13.18%), Asian: 61 (10.30%), unknown: 6 (1.01%) |
{8}
| Characteristic | Test Set-1 (Taiwanese Multi-Center Cohort) | Test Set-2 (FEMH Cohort) | Test Set-3 (U.S. Multi-Center Cohort) |
| --- | --- | --- | --- |
| T-score ≤ -2.5 n (%) | 149 (25.82%) | 63 (39.62%) | 101 (17.06%) |
| -2.5 < T-score < -1 n (%) | 219 (37.95%) | 60 (37.74%) | 213 (35.98%) |
| T-score ≥ -1 n (%) | 209 (36.22%) | 36 (22.64%) | 278 (46.96%) |
| #Xray DICOM (Xray paired to DXA) | 718 | 159 | 749 |
For the primary classification task (identifying patients with T-score ≤ -2.5, the "Low BMD" screening result), DeepXray Spina demonstrated the following combined performance across all three test sets: AUC = 0.955 (95% CI: 0.942–0.967), sensitivity = 77.0% (95% CI: 72.0%–81.9%), and specificity = 93.6% (95% CI: 92.1%–95.1%).
Table 2. Primary Classification (L1–L4 T-score ≤ -2.5), Intended Use Population, Age ≥ 50
| Endpoint | Test Set-1 | Test Set-2 | Test Set-3 | Combined |
| --- | --- | --- | --- | --- |
| *Scenario 1 – “Report with QC Warning” as reportable result; counted in performance* | | | | |
| #Patients | 505 | 148 | 566 | 1,219 |
| AUC | 0.961 (0.942, 0.977) | 0.968 (0.942, 0.988) | 0.942 (0.919, 0.963) | 0.955 (0.942, 0.967) |
| Sensitivity | 0.771 (0.699, 0.841) | 0.934 (0.857, 0.985) | 0.684 (0.581, 0.775) | 0.770 (0.720, 0.819) |
| Specificity | 0.954 (0.931, 0.974) | 0.874 (0.802, 0.941) | 0.931 (0.909, 0.953) | 0.936 (0.921, 0.951) |
| *Scenario 2 – “Report with QC Warning” treated as no output; considered as negative result* | | | | |
| #Patients | 505 | 148 | 566 | 1,219 |
| AUC | 0.961 (0.942, 0.977) | 0.968 (0.942, 0.988) | 0.942 (0.919, 0.963) | 0.955 (0.942, 0.967) |
| Sensitivity | 0.760 (0.688, 0.832) | 0.918 (0.836, 0.982) | 0.684 (0.581, 0.775) | 0.762 (0.712, 0.809) |
| Specificity | 0.954 (0.931, 0.974) | 0.874 (0.802, 0.941) | 0.933 (0.910, 0.954) | 0.937 (0.922, 0.952) |
{9}
# Subgroup analysis
Additional performance metrics were assessed across predefined subgroups, including age, sex, ethnicity, BMI, X-ray equipment manufacturer, and image protocol, to evaluate the consistency and generalizability of the device's performance. These results support the device's effectiveness in reliably identifying high-risk individuals in real-world clinical settings.
Note: Age subgroup analysis includes subjects aged <50 years to account for all enrolled subjects.
Table 3: Classification Performance (T-score ≤ -2.5) by age group
| Group | #Patients | #Patients having T-score ≤ -2.5 by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- |
| < 50 | 109 | 10 | 0.862 (0.737, 0.993) | 0.429 (0.143, 0.875) | 0.983 (0.957, 1.000) |
| 50-59 | 335 | 59 | 0.976 (0.962, 0.988) | 0.758 (0.648, 0.869) | 0.952 (0.925, 0.975) |
| 60-69 | 466 | 137 | 0.948 (0.926, 0.969) | 0.784 (0.720, 0.855) | 0.937 (0.912, 0.962) |
| 70-79 | 296 | 81 | 0.936 (0.906, 0.963) | 0.725 (0.626, 0.830) | 0.920 (0.884, 0.952) |
| >=80 | 122 | 26 | 0.970 (0.937, 0.991) | 0.842 (0.687, 0.970) | 0.923 (0.865, 0.969) |
Note: All subgroup analyses in Table 4 to Table 14 are performed on the Intended Use Population (Age ≥ 50). Subjects aged <50 years are excluded from these analyses.
Table 4: Classification Performance (T-score ≤ -2.5) by Sex
| Group | #Patients | #Patients having T-score ≤ -2.5 by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- |
| Female | 1017 | 280 | 0.957 (0.944, 0.970) | 0.787 (0.738, 0.836) | 0.934 (0.917, 0.952) |
| Male | 202 | 23 | 0.924 (0.868, 0.969) | 0.571 (0.387, 0.789) | 0.942 (0.905, 0.971) |
{10}
*Table 5: Classification Performance (T-score $\leq -2.5$) by BMI group*
| Group | #Patients | #Patients having T-score $\leq -2.5$ by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- |
| <18.5 | 37 | 18 | 0.974 (0.920, 1.000) | 1.000 (1.000, 1.000) | 0.750 (0.556, 0.933) |
| 18.5-25 | 503 | 156 | 0.944 (0.918, 0.963) | 0.851 (0.794, 0.903) | 0.888 (0.851, 0.919) |
| >=25 | 450 | 72 | 0.971 (0.957, 0.983) | 0.655 (0.548, 0.759) | 0.972 (0.955, 0.987) |
| unreported | 229 | 57 | 0.953 (0.919, 0.978) | 0.658 (0.540, 0.780) | 0.967 (0.940, 0.988) |
*Table 6: Classification Performance (T-score $\leq -2.5$) by Ethnic group*
| Group | #Patients | #Patients having T-score $\leq -2.5$ by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- |
| Asian | 712 | 219 | 0.956 (0.939, 0.971) | 0.809 (0.755, 0.862) | 0.936 (0.915, 0.955) |
| Black | 76 | 9 | 0.972 (0.919, 1.000) | 0.700 (0.333, 1.000) | 0.949 (0.892, 0.988) |
| Hispanic | 118 | 24 | 0.950 (0.902, 0.987) | 0.571 (0.379, 0.786) | 0.976 (0.945, 1.000) |
| White | 308 | 49 | 0.953 (0.926, 0.975) | 0.746 (0.611, 0.860) | 0.917 (0.880, 0.950) |
| Unknown | 5 | 2 | 0.667 (0.000, 1.000) | 0.000 (0.000, 0.000) | 1.000 (1.000, 1.000) |
\* 5 patients with unknown ethnic group are included in the table above
{11}
*Table 7: Classification Performance (T-score $\leq -2.5$) by Xray Manufacturer*
| Group | #Patient | #DICOM | #DICOM having T-score $\leq -2.5$ by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- | --- |
| Canon Inc. | 159 | 189 | 56 | 0.985 (0.970, 0.995) | 0.857 (0.754, 0.943) | 0.962 (0.926, 0.992) |
| FUJIFILM | 362 | 438 | 115 | 0.952 (0.923, 0.974) | 0.774 (0.689, 0.853) | 0.938 (0.904, 0.966) |
| GE Healthcare | 62 | 82 | 9 | 0.939 (0.862, 0.996) | 0.778 (0.411, 1.000) | 0.863 (0.771, 0.947) |
| KONICA MINOLTA | 204 | 231 | 51 | 0.955 (0.923, 0.979) | 0.765 (0.629, 0.870) | 0.928 (0.886, 0.963) |
| PZMedical | 130 | 147 | 36 | 0.981 (0.960, 0.996) | 0.833 (0.710, 0.941) | 0.964 (0.922, 0.992) |
| Philips | 32 | 36 | 6 | 0.961 (0.857, 1.000) | 0.667 (0.000, 1.000) | 1.000 (1.000, 1.000) |
| SAMSUNG | 130 | 171 | 38 | 0.934 (0.892, 0.968) | 0.711 (0.552, 0.861) | 0.895 (0.830, 0.950) |
| SIEMENS | 136 | 164 | 39 | 0.923 (0.866, 0.974) | 0.692 (0.545, 0.852) | 0.944 (0.904, 0.982) |
| Others | 27 | 37 | 7 | 1.000 (1.000, 1.000) | 0.571 (0.200, 1.000) | 1.000 (1.000, 1.000) |
{12}
Table 8: Classification Performance (T-score ≤ -2.5) by Modality type
| Group | #Patient | #DICOM | #DICOM having T-score ≤ -2.5 by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- | --- |
| CR | 999 | 1215 | 301 | 0.958 (0.943, 0.971) | 0.761 (0.703, 0.815) | 0.951 (0.935, 0.965) |
| DX | 222 | 280 | 56 | 0.950 (0.921, 0.973) | 0.821 (0.702, 0.933) | 0.875 (0.824, 0.922) |
Table 9: Classification Performance (T-score ≤ -2.5) by DXA Manufacturer
| Group | #Patient | #DICOM | #DICOM having T-score ≤ -2.5 by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- | --- |
| GE | 1089 | 1333 | 328 | 0.955 (0.940, 0.966) | 0.780 (0.729, 0.831) | 0.931 (0.914, 0.947) |
| Hologic | 130 | 162 | 29 | 0.961 (0.912, 0.990) | 0.655 (0.440, 0.840) | 0.970 (0.938, 0.993) |
Table 10: Classification Performance (T-score ≤ -2.5) by X-ray Protocol
| Group | #Patient | #DICOM | #DICOM having T-score ≤ -2.5 by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- | --- |
| KUB | 571 | 678 | 166 | 0.950 (0.931, 0.966) | 0.735 (0.667, 0.809) | 0.934 (0.910, 0.953) |
| LSPINE | 754 | 817 | 191 | 0.958 (0.939, 0.973) | 0.801 (0.736, 0.856) | 0.938 (0.917, 0.957) |
{13}
*Table 11: Classification Performance (T-score $\leq -2.5$) by X-ray Tube Voltage (kVp)*
| Group | #Patient | #DICOM | #DICOM having T-score $\leq -2.5$ by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- | --- |
| $\leq 80$ | 256 | 327 | 77 | 0.942 (0.908, 0.971) | 0.714 (0.604, 0.836) | 0.924 (0.887, 0.958) |
| $>80$ | 319 | 346 | 96 | 0.972 (0.958, 0.984) | 0.875 (0.807, 0.944) | 0.932 (0.900, 0.960) |
| Unknown | 658 | 822 | 184 | 0.950 (0.930, 0.967) | 0.739 (0.670, 0.805) | 0.942 (0.920, 0.962) |
*Table 12: Classification Performance (T-score $\leq -2.5$) by X-ray Tube Current (mAs)*
| Group | #Patient | #DICOM | #DICOM having T-score $\leq -2.5$ by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- | --- |
| $<30$ | 423 | 500 | 147 | 0.946 (0.921, 0.967) | 0.810 (0.733, 0.879) | 0.907 (0.876, 0.938) |
| $30-40$ | 146 | 163 | 36 | 0.976 (0.953, 0.993) | 0.750 (0.588, 0.889) | 0.961 (0.921, 0.992) |
| $>40$ | 101 | 109 | 20 | 0.966 (0.929, 0.989) | 0.800 (0.619, 0.947) | 0.910 (0.844, 0.966) |
| unknown | 584 | 723 | 154 | 0.954 (0.932, 0.972) | 0.734 (0.658, 0.810) | 0.953 (0.933, 0.971) |
{14}
*Table 13: Classification Performance (T-score $\leq -2.5$) by DXA – Xray Interval (days)*
| Group | #Patient | #DICOM | #DICOM having T-score $\leq -2.5$ by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- | --- |
| 0 | 316 | 347 | 103 | 0.959 (0.938, 0.976) | 0.806 (0.721, 0.884) | 0.947 (0.915, 0.973) |
| 1-60 | 698 | 803 | 175 | 0.953 (0.932, 0.970) | 0.771 (0.700, 0.836) | 0.930 (0.907, 0.950) |
| 61-180 | 276 | 345 | 79 | 0.951 (0.923, 0.973) | 0.722 (0.618, 0.821) | 0.940 (0.906, 0.970) |
*Table 14: Classification Performance (T-score $\leq -2.5$) by X-ray pixel size (mm)*
| Group | #Patient | #DICOM | #DICOM having T-score $\leq -2.5$ by DXA | AUC | Sensitivity | Specificity |
| --- | --- | --- | --- | --- | --- | --- |
| <0.14 | 266 | 332 | 73 | 0.962 (0.932, 0.984) | 0.726 (0.618, 0.829) | 0.958 (0.928, 0.982) |
| 0.14-0.16 | 423 | 513 | 128 | 0.947 (0.925, 0.969) | 0.766 (0.690, 0.850) | 0.927 (0.898, 0.955) |
| >0.16 | 553 | 650 | 156 | 0.956 (0.936, 0.973) | 0.795 (0.724, 0.860) | 0.931 (0.905, 0.953) |
### *Reproducibility analysis*
Reproducibility on 194 cases with repeated acquisitions yielded a coefficient of variation of 3.4% and Cohen's kappa of 0.703–0.706 at the $T \leq -2.5$ operating point. There are 118 subjects having 2 X-ray images on the same day, 76 subjects having 2 X-rays on different days (but still paired to the same DXA study).
{15}
Table 15: Reproducibility analysis
| | X-rays on the same day | X-rays on different days (1-266 day)* |
| --- | --- | --- |
| Subject | 118 | 76 |
| % low bone mass (T-score < -1) | 55.98% | 57.89% |
| % osteoporosis (T-score ≤ -2.5) | 18.64% | 26.32% |
| CV% of Repeated BMD estimation by DeepXray | 3.44% | 3.38% |
| Classification for T-score ≤ -2.5 | | |
| % agreement | 91.53% | 90.79% |
| kappa | 0.703 | 0.706 |
| z statistic | 7.815 | 6.66 |
| p-value | <0.0001 | <0.0001 |
| Classification for T-score < -1 | | |
| % agreement | 88.98% | 88.16% |
| kappa | 0.764 | 0.743 |
| z statistic | 12.379 | 9.349 |
| p-value | <0.0001 | <0.0001 |
* The same patient having 2 instances of X-ray study on different dates but still paired to the same DXA study. Since the DXA pairing criterion is ± 6 months (180 days), one X-ray study might be before DXA and another after, thus total interval between two X-ray studies could be more than 180 days.
### Vertebral ROI Localization
ROI localization was directly validated on Test Set-1 (Asian cohort) against manual reference annotations; all per-vertebra (L1–L4) mean Intersection-over-Union (IoU) values and the overall L-spine ROI bounding-box IoU met the pre-specified acceptance criterion (95% CI lower bound ≥ 0.80). ROI localization was not independently validated with IoU metrics on Test Set-2 and Test Set-3; however, it was indirectly verified across all test sets as an integral intermediate step of the end-to-end AI pipeline, and overall pipeline performance met pre-specified acceptance criteria across all cohorts.
{16}
# Automated Quality-Control (QC) Function
Evaluated on a dedicated problematic-image set (n = 297; images bearing IFU-excluded conditions against which the QC function serves as a second-tier automated safeguard) and on all clinically acceptable images from Test Sets 1, 2, and 3 (n = 1,626). Two QC performance scenarios are reported. Scenario 1 (“Report with Warning” included as reportable result): QC sensitivity = 76.4% (95% CI: 71.2%–81.1%), QC specificity = 100.0% (95% CI: 99.8%–100.0%). Scenario 2 (“Report with Warning” excluded; treated as no output): QC sensitivity = 93.6% (95% CI: 90.2%–96.1%), QC specificity = 98.2% (95% CI: 97.4%–98.8%).
# Substantial Equivalence
DeepXray Spina has the same general intended use (opportunistic identification of patients who may have low BMD), the same intended patient population (adults ≥ 50 years), the same principle of operation (locked AI/ML radiogrammetry with DXA as the training reference), and the same cleared output form (binary classification) as its predicate device, Rho (16 Bit, Inc., DEN230023). The use of the anatomically matched L-spine / KUB input, the image-only inference, and the T ≤ −2.5 operating point (vs. Rho’s T < −1) do not alter the intended use and do not raise new or different questions of safety and effectiveness. At the predicate’s own operating point (T < −1), DeepXray Spina’s supporting classification performance (combined AUC = 0.939) is comparable to or exceeds the reported performance of the predicate.
The non-clinical and clinical performance data summarized above demonstrate that DeepXray Spina performs as intended and is as safe and effective as the predicate device for its intended use. The results also fulfill the applicable special controls for product code SAO under 21 CFR 892.1171. DeepXray Spina is therefore substantially equivalent to the predicate device.
