Cross-case comparison · Atlas v1.3
Regulatory Evidence Matrix
Compare pathway, disclosed AI function, validation design, change control, and postmarket boundaries across ten selected FDA case studies.
- Public basis
- 10 selected cases
- Jurisdiction
- United States
- Outcome benefit
- 0 established
- Selection reviewed
- Sep 9, 2026
What this comparison does
Places six regulatory dimensions beside one another so differences in pathway, disclosure, evidence, change control, and postmarket scope remain visible.
What it does not do
It is not a complete FDA market census, vendor ranking, purchasing guide, or proof that an authorized device improves patient outcomes.
Ten teaching cases
Compare the claim with its boundary.
Labels summarize the full case records. Follow a device name for the intended use, detailed validation, source links, and version-specific limits.
On narrow screens, scroll horizontally to compare all six columns.
| Device / function | FDA pathway | Public AI disclosure | Strongest validation design | PCCP / change control | Postmarket boundary |
|---|---|---|---|---|---|
| SafeOp 3: Neural Informatix SystemK252842 · Monitoring | 510(k)Substantial equivalence | Algorithm / classifierControlled updates under a PCCP | NonclinicalSoftware, electrical-safety, EMC, and nonclinical performance verification. | PCCP authorizedIn K252842 | Component / model recall matchScoped record; not a product-wide event rate |
| Brainomix 360 HyperdensityK260406 · Imaging | 510(k)Substantial equivalence | AI/ML explicitControlled updates under a PCCP | Retrospective clinicalClinical image-set comparison against radiologist ground truth. | PCCP authorizedIn K260406 | No exact-name recall resultBounded search; not proof that no recall exists |
| Mazor X System / Mazor X Stealth EditionK251316 · Planning | 510(k)Substantial equivalence | AI/ML explicitLocked model or algorithm | Retrospective clinicalRetrospective imaging evaluation plus expert assessment of suggested plans. | No PCCP identifiedReviewed submission and FDA database record | Product-family recall matchNot directly attributed to the indexed submission |
| SpineAR SNAP (SyncAR Spine)K252054 · Navigation | 510(k)Substantial equivalence | AI/ML explicitAdaptivity not established publicly | Retrospective clinicalHeld-out CT and MR vertebral-segmentation evaluation. | No PCCP identifiedReviewed submission and FDA database record | Product-family recall matchNot directly attributed to the indexed submission |
| Acorn 3D Software (AC-SEG-4009); Acorn 3DP Model (AC-101-XX)K260322 · Planning | 510(k)Substantial equivalence | AI/ML explicitControlled updates under a PCCP | NonclinicalSoftware, end-user, and bench testing of correction planning. | Inherited PCCPAuthorized in K234009, not the indexed submission | No exact-name recall resultBounded search; not proof that no recall exists |
| JLK-NCCTK252421 · Imaging | 510(k)Substantial equivalence | AI/ML explicitAdaptivity not established publicly | Retrospective clinicalIndependent retrospective standalone and reader-performance evaluation for LVO. | No PCCP identifiedReviewed submission and FDA database record | No exact-name recall resultBounded search; not proof that no recall exists |
| BrainSeeDEN220066 · Prognosis | De NovoNew device type | Algorithm / classifierLocked model or algorithm | Retrospective clinicalRetrospective pivotal validation using longitudinal clinical diagnoses. | No PCCP identifiedReviewed submission and FDA database record | Required performance studyNo exact-name recall result in the targeted search |
| BrainScope Ahead 100DEN140025 · Assessment | De NovoNew device type | Algorithm / classifierLocked model or algorithm | Prospective clinicalProspective, blinded, 11-site US pivotal study. | No PCCP identifiedReviewed submission and FDA database record | No exact-name recall resultBounded search; not proof that no recall exists |
| EyeBOXDEN170091 · Assessment | De NovoNew device type | Algorithm / classifierLocked model or algorithm | Prospective clinicalPivotal head-injury study; the clinical reference standard changed during the study. | No PCCP identifiedReviewed submission and FDA database record | Component / model recall matchScoped record; not a product-wide event rate |
| xvision Spine systemK251639 · Navigation | 510(k)Substantial equivalence | AI function not established publiclyAdaptivity not established publicly | NonclinicalSoftware, accuracy, phantom, cadaver, and human-factors testing. | No PCCP identifiedReviewed submission and FDA database record | Component / model recall matchScoped record; not a product-wide event rate |
Reading notes
The columns are related, but not interchangeable.
- Pathway
- 510(k) and De Novo answer different regulatory questions; neither is an evidence grade.
- Disclosure
- FDA-list inclusion does not by itself identify a learning architecture, generative interface, or foundation model.
- Validation
- The matrix reports the strongest disclosed design, not a quality score. Nonclinical, retrospective, and prospective evidence support different claims.
- Change control
- An authorized PCCP permits bounded changes. It does not imply unrestricted or continuously self-learning operation.
- Postmarket
- Recall matches are version- and component-sensitive; a negative targeted search is not proof that no recall or event exists.