Evidence-Informed Development
Grounded in published evidence
Pacslens triage models are designed against the published literature on AI-assisted radiology for critical findings. We cite our sources.
Our Approach
How we interpret the evidence
AI triage claims in radiology require published sensitivity and specificity benchmarks from peer-reviewed sources. We do not make unsupported accuracy claims. The citations below represent the published literature that informs our model development targets — not proprietary unpublished claims.
At angel-funded, Pacslens has not yet published its own prospective trial data. The evidence table below references published studies on FDA-cleared competitor devices as the landscape context within which we design and position our models. This is an honest framing: competitor clearances validate the indication categories; our own clinical validation work is ongoing.
Important: Sensitivity and specificity figures cited below come from published peer-reviewed literature and FDA 510(k) submissions for cleared third-party devices. These are not Pacslens performance claims. Pacslens is not FDA-cleared. See our regulatory pathway section.
Literature Review
Published literature by indication
| Indication | Context & Reference | Published Finding |
|---|---|---|
| LVO | Viz.ai LVO detection — FDA cleared K193482; multiple published validation studies (Stroke, 2021) | Sensitivity 91–95%, Specificity 88–94% in published retrospective cohorts. Referenced by Viz.ai in 510(k) K193482 predicate context. |
| ICH | Aidoc ICH — FDA cleared K183327; NEJM Evidence 2023 AI triage study on acute ICH pathway | Significant time-to-read reduction in acute ICH pathway. Study demonstrated reduced door-to-read time for STAT ICH findings in community hospital setting. |
| PE | Multiple CTPA AI detection studies (Radiology 2020, European Radiology Supplements 2022) | AUC 0.90–0.94 for central PE detection in retrospective series. Performance decreases for subsegmental PE — an acknowledged limitation in the literature. |
| Aortic Dissection | Emerging literature; FDA 510(k) clearances in progress from multiple vendors at time of writing | Limited prospective data available in public literature. Pacslens model currently validated on retrospective holdout dataset only — not cleared for clinical use. |
| Pneumothorax | Multiple CXR/CT AI studies; RapidAI (K201992) and Aidoc (K203804) published validation data | High sensitivity for moderate-large pneumothorax on CT. Limited prospective data for tension PTX specifically — differentiated from simple PTX in few published series. |
FDA K-numbers reference the public FDA 510(k) database for third-party cleared devices. These are cited for literature context only, not as Pacslens clearance claims. Pacslens has no FDA clearance at this time.
Regulatory Status
Our regulatory pathway
Pacslens is preparing 510(k) submissions for selected indications under FDA Class II medical device regulations (21 CFR Part 892). We expect to reference published predicate devices including Aidoc (K203804), Viz.ai (K193482), and RapidAI (K193346) in our submissions.
Until FDA clearance is obtained, Pacslens is available for evaluation and pilot deployment in non-diagnostic-primary-read contexts — as a prioritization aid, not a diagnostic device. The radiologist remains the responsible party for all clinical reads.
CURRENT STATUS
FDA 510(k) submission in progress for LVO and ICH indications. No clearance yet.
PREDICATE DEVICES
K203804 · K193482 · K193346 · K183327
REGULATORY CLASS
FDA Class II medical device, 21 CFR Part 892 (radiology AI software).