Brainomix 360 Hyperdensity

Authorization

FDA cleared a narrower output set than the predicate and authorized controlled performance optimization without continuous in-field learning.

Submission lineageK260406 → predicate K232363; PCCP authorized in K260406.

Intended use

Automatically label, visualize and volumetrically quantify intracranial hyperdensities on a non-contrast head CT acquired at one time point.

Input

Adult non-contrast head CT DICOM series from a single time point

Output

Annotated DICOM segmentation series with color overlays; DICOM summary series with total hyperdensity volume; Editable local viewer display

Use boundary

The physician must review the output with the original images; segmentation and volumetry assist image processing and do not independently establish diagnosis or treatment.

AI disclosure and function

AI/ML explicit

The summary explicitly identifies a locked deep-learning AI/ML algorithm for intracranial-hyperdensity segmentation and volumetry.

Future PCCP updates may retrain weights or adjust subtype thresholds, but every modified version remains trained, tuned and locked before release; continuous in-field learning is excluded.

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

Validation

Clinical image-set comparison of automated segmentation and volumetry with manual radiologist ground truth, with subgroup stratification and software verification.

Comparator

Manual intracranial-hyperdensity segmentations and performance criteria aligned with the K232363 predicate.

Ground truth

Manual measurements established by trained radiologists.

Representation

Performance was stratified by clinical site, sex, age, race or ethnicity, slice thickness, scanner manufacturer and estimated volume, but the public summary provides no subgroup counts.

Reported performance

Hyperdensity-volume mean absolute error

upper 95% CI bound <7.5 mLThe prespecified volumetric precision goal was met; the exact observed MAE is not disclosed in the summary.

Hyperdensity Dice

lower 95% CI bound >70%The prespecified segmentation-overlap goal was met; the exact observed Dice score is not disclosed in the summary.

Limitations

• The public summary does not disclose the sample size, site count, exact observed primary metrics or subgroup results.
• It does not establish prospective workflow benefit, clinical decisions or patient outcomes.
• The device does not report the predicate's lateral-ventricle volume or midline-shift outputs.

It does not autonomously diagnose hemorrhage, establish prospective utility, improve outcomes or continuously learn after deployment.

PCCP / Change control

Authorized in K260406. Retraining model weights with expanded subtype-enriched datasets and adjusting probability thresholds for subtype classes without changing indications, workflow, clinical role or fundamental algorithmic approach.

• Predefined verification and validation framework and acceptance criteria
• Impact assessment for overfitting, bias, over-segmentation and under-segmentation
• Train, tune and lock every modified model before release
• Release-note communication to users before deployment
• No continuously learning adaptive algorithm in the field

A PCCP authorizes bounded changes under predefined controls. It is not permission for unrestricted or continuously self-learning behavior.

Postmarket check

No exact-name openFDA device-enforcement result was found in the targeted query.

No exact-name openFDA result found as of 2026-09-08; this is not proof that no recall exists.

Primary FDA sources

1. FDA AI-enabled device list — Submission K260406Confirms inclusion, list display name, company, lead panel, product code and decision date only.
2. FDA submission database — 510(k) K260406Confirms pathway, review panel, decision and Yes PCCP database field.
3. FDA decision summary — 510(k) summary pp. 1-8; performance pp. 3-4; PCCP pp. 6-8Supports intended use, locked deep-learning architecture, clinical validation boundaries, predicate and authorized PCCP scope and controls.

Sources checked Sep 8, 2026 · Next scheduled review Oct 8, 2026. “No exact-name result” is a bounded search finding, not proof that no safety event or recall exists.