← Regulatory Atlas

Planning · K251316

Mazor X System / Mazor X Stealth Edition

Two locked AI planning functions inside a robotic spine platform, with expert plan acceptance kept separate from placement accuracy and patient outcomes.

Mazor Robotics Ltd.

Pathway
Traditional 510(k)
Decision
Sep 11, 2025
Product code
OLO
Evidence
Retrospective clinical
First published
Sep 9, 2026
Page updated
Sep 17, 2026
Evidence search
Sep 9, 2026
Professional review
Not claimed

Anonymous editorial source review; independent regulatory, legal, or professional peer review is not claimed. Page updates do not reset evidence dates.

What FDA established

FDA accepted two bounded locked-AI planning functions in a robotic spine platform after retrospective measurement and expert plan-acceptability validation.

What FDA did not establish

It does not prove autonomous planning, screw-placement accuracy, fewer complications, improved outcomes or safety of every product-family software version.

Authorization

K251316 added AI-supported spinopelvic measurements and initial pedicle-screw Plan Assist while preserving surgeon review and approval of the operative plan.

Submission lineageK251316 → predicate K230064; ongoing recall Z-1122-2026 is scoped to the Mazor X product family and specified software versions, not directly to this submission.

Intended use

Precisely position surgical instruments or spinal implants during open, minimally invasive or percutaneous general spine surgery, generate volumetric three-dimensional images from C-arm projections, and track instruments relative to patient anatomy and images.

Input
Preoperative CT and intraoperative fluoroscopic, O-arm or supported three-dimensional image data; AP and lateral radiographs for spinopelvic measurements; Surgeon planning inputs and user preferences
Output
Registered images and navigated instrument position; Robotic trajectory guidance; AI-generated spinopelvic landmark measurements; AI-suggested initial pedicle-screw positions for surgeon review
Use boundary
The surgeon defines and approves the operative plan. AI measurements and suggested screw positions are initial planning aids and do not autonomously select, place or verify an implant.

AI disclosure and function

AI/ML explicit

The summary explicitly identifies two locked ML landmark-detection algorithms for 2D automatic measurements and a locked Plan Assist algorithm that suggests initial pedicle-screw positions.

The AI functions generate measurements and initial plans within defined image and anatomy boundaries; they are not self-learning systems or autonomous robotic control.

No foundation-model, LLM, or generative-AI function was identified in the reviewed FDA materials.

Validation

Retrospective image-based evaluation of thirteen spinopelvic measurements against radiologist annotations and expert assessment of AI-generated pedicle-screw plans, with development and validation datasets separated at the site level.

Comparator
Three US board-certified radiologists for spinopelvic measurements and three expert assessments of clinical acceptability for each AI-generated screw plan; K230064 served as the system predicate.
Ground truth
Radiologist-created endplate, endpoint and femoral-head annotations for measurements; expert clinical-acceptability judgments for screw plans.
Representation
Measurement validation reported 67% female, 30% male and 3% unknown, with pediatric/adolescent, adult and geriatric cases. Plan Assist reported 28% female, 12% male and 60% unknown, with 60% age group unknown. Poor-quality and specified complex conditions were excluded.

Reported performance

Spinopelvic measurement agreement
ICC greater than 0.75 with 95% confidence for all tested measurementsAgreement with radiologist annotations in retrospective radiographs, not intraoperative accuracy or outcome evidence.
Plan Assist clinical acceptability
true-positive rate exceeded 80% with 95% confidenceExpert acceptance of 326 generated screw plans from 25 spine images, not observed implant placement accuracy.

Limitations

  • The two AI functions used different validation datasets and cannot be represented by one patient count.
  • Plan Assist testing included only 25 spine images and 326 plans, with substantial unknown demographic data.
  • Poor images, implants and several clinically important complex conditions were excluded.
  • No prospective surgical workflow, placement-accuracy, complication or patient-outcome evidence was reported.

It does not prove autonomous planning, screw-placement accuracy, fewer complications, improved outcomes or safety of every product-family software version.

PCCP / Change control

No authorized PCCP was identified for this submission in the reviewed FDA database record or decision material.

Absence of an identified PCCP does not prevent every future device change; it means no authorized plan was established in the reviewed public record.

Postmarket check

Multiple openFDA enforcement records match the Mazor X product family. Class II recall Z-1122-2026 is ongoing and covers specified Mazor X software versions 5.0.1, 5.1.2 and 5.1.3 for errors that may cause incorrect surgical-instrument positioning.

The enforcement record is matched to the Mazor X product family and specified software versions, not directly to submission K251316 or every Mazor X installation. It is not a patient-event rate.

Primary FDA sources

  1. FDA AI-enabled device list — Submission K251316Confirms inclusion, list display name, company, lead panel, primary product code and decision date only.
  2. FDA submission database — 510(k) K251316Confirms pathway, Orthopedic review panel, decision and No PCCP database flag.
  3. FDA decision summary — 510(k) summary pp. 1-16; intended use pp. 1-2; AI functions and validation pp. 12-15Supports intended use, locked AI functions, dataset boundaries, reference standards, performance criteria, exclusions and predicate comparison.
  4. FDA recall record — Recall Z-1122-2026Supports the product-family and specified-software-version recall; does not attribute the recall directly to K251316 or unlisted versions.

Sources checked Sep 9, 2026 · Next scheduled review Oct 9, 2026. “No exact-name result” is a bounded search finding, not proof that no safety event or recall exists.