EyeBOX

Authorization

FDA created a new Class II eye-movement assessment type for adjunctive concussion evaluation within one week of injury.

Submission lineageDEN170091; no predicate submission.

Intended use

Measure and analyze eye movements as an aid in diagnosing concussion within one week of head injury, together with a standard neurological assessment.

Input

Binocular gaze coordinates recorded for 220 seconds while viewing a moving video stimulus

Output

EyeBOX score from 0 to 20; Positive or negative EyeBOX classification; Eye-tracking plots in the clinical report

Use boundary

Adjunct only: a positive result may occur with or without concussion, a negative result does not exclude concussion, and the device is not a substitute for CT or standard neurological assessment.

AI disclosure and function

Algorithm / classifier

The summary describes normalization of high-frequency binocular gaze data into a 0-20 score and threshold classification, without naming a machine-learning architecture.

The device is algorithmic and appears on FDA's AI-enabled list, but the reviewed materials do not establish a foundation model, LLM or generative interface.

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

Validation

Pivotal clinical study of enrolled head-injured subjects comparing device classification with a trial-specific clinical reference standard; the reference definition was revised during the study while device output remained blinded.

Comparator

A trial-specific clinical concussion classification based on alteration of consciousness or mental status plus SCAT3 cognitive and symptom thresholds.

Ground truth

The revised clinical reference standard replaced an initial three-clinician panel after 84.4% of the first 199 adjudications included at least one uncertain judgment.

Representation

293 eligible subjects were screened and 282 were analyzed; the cohort included pediatric and adult patients, but redacted subgroup counts and no complete race or site table prevent a full representation assessment.

Reported performance

Sensitivity

80.4% (37/46; 95% CI 66.1%-91.9%)Classification performance against the revised clinical reference; the prespecified lower-bound sensitivity goal was not met.

Specificity

66.1% (156/236; 95% CI 59.7%-72.1%)Classification performance against the revised clinical reference; the prespecified lower-bound specificity goal was not met.

Negative predictive value

94.5% (156/165; 95% CI 89.9%-97.5%)Observed at 16.3% concussion prevalence and not a standalone rule-out result.

Positive predictive value

31.6% (37/117; 95% CI 23.3%-40.9%)A positive classification occurred frequently without meeting the revised clinical concussion definition.

Test-retest equivalent classification

79.3% (95% CI 67.8%-87.5%; n=63)Repeat classifications showed material variability that FDA addressed through labeling and benefit-risk controls.

Limitations

• The prespecified sensitivity and specificity performance goals were not met.
• The clinical reference standard was revised during the study after extensive uncertainty in panel adjudication.
• Only 263 analyzed subjects were assessed within one week, limiting effectiveness evidence beyond that window.
• Test-retest classifications were equivalent in 79.3% of the 63 repeated assessments.
• The study did not establish improved recovery, reduced repeat injury or another patient outcome.

It does not independently diagnose or exclude concussion, replace CT, show stable classification in every repeat test or improve clinical outcomes.

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

FDA Class II recall Z-2398-2019 involved three EyeBOX Model OCL 02 units whose user manual omitted two precaution statements; the recall is terminated.

This historical record is model- and unit-specific and should not be generalized to every EyeBOX unit or used as a patient-event rate.

Primary FDA sources

1. FDA AI-enabled device list — Submission DEN170091Confirms inclusion, list display name, company, lead panel, product code and decision date only.
2. FDA submission database — De Novo DEN170091Confirms grant, product code, decision date and PCCP database field.
3. FDA decision summary — De Novo summary pp. 1-14; clinical study pp. 6-9; benefit-risk pp. 11-14Supports intended use, algorithm description, pivotal cohort, revised reference standard, missed goals, repeatability, labeling and benefit-risk boundary.
4. FDA recall record — Recall Z-2398-2019Supports the three-unit Model OCL 02 labeling recall and its terminated status.

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.