AccurKardia

The company presents its work as AI enabled ECG, and the cleared product's own filings describe no learned component. AccurECG performs single lead, beat by beat analysis, detects ventricular and supraventricular ectopic beats, measures heart rate and interprets 13 rhythm classifications, and the company sells that interpretation as software. The 510(k) summary for the first clearance (K223013, 23 June 2023, AccurKardia, Inc., New York) describes a "server-side automated proprietary ECG interpretation support algorithm" written in LabVIEW, with a web interface coded in Java, and uses no machine learning, deep learning, neural network or AI language. The summary for version 2.0 (K252361, 22 December 2025) describes "advanced signal processing and image-based techniques" for waveform delineation and arrhythmia interpretation, again without a trained model. The FDA's list of AI enabled devices, as published through December 2025, does not show K252361.

The AI language on the company's pages attaches most clearly to two investigational ECG based screening tools, AK+ Guard and AK-AVS, which hold Breakthrough Device Designation and are research use only, not sold. For the product that is sold, no learned component can be found: the interpretation may be a strong rule and signal processing system, but it is not described as a model estimated from data.

Artificial intelligence is claimed and the product's filings describe signal processing and a proprietary algorithm rather than a learned model.

The limits are stated in the clearance and the handling of uncertain output is not. AccurECG produces fully automated interpretation for qualified healthcare professionals to assess; the cleared indication covers adults 22 and older prescribed outpatient cardiac monitoring for up to ten days, single lead recordings from wet silver chloride electrodes in named file formats, and excludes life supporting systems and ECG monitor and alarm devices, with results not to be the sole means of diagnosis.

That sets what the system may and may not do. What is not published is the threshold at which it declines to interpret: how noisy or out of scope recordings are flagged, and how a reader's corrections are captured.

The approach and the versions are identified, and the training data is not. The version 2.0 FDA summary describes a client side web interface and a server side algorithm using advanced signal processing and image based techniques for QRS and T wave delineation and arrhythmia interpretation, written in Python on cloud virtual machines, and lists the tests run: unit, code review, module and system verification, performance evaluation and cybersecurity testing.

Versions are traceable through the clearances, v1.3.0 under K223013 and v2.0 under K252361, which ties a change to a regulatory review. The company does not say whether any component is trained on data or what data it was developed on, and it gives no per class performance behind the 99 percent figure.

The interpretation layer is the company's own cleared algorithm, which cannot be quietly swapped, and the rest of the chain is unnamed. AccurECG is developed by AccurKardia and each version is cleared by FDA, so the component that produces an interpretation is fixed and attributable. Around it nothing is named: the cloud provider hosting the Ubuntu virtual machines, any subprocessors, and the device makers and monitoring services whose systems send recordings and receive results. Because the product sits between recording devices and reading facilities, a patient's recording passes through several organizations, and no participant publishes the map.

Two clearances and a vendor reported accuracy figure, with no study published behind it. AccurECG holds FDA clearances K223013 (June 2023) and K252361 for version 2.0 (December 2025). Neither 510(k) summary publishes performance results: the first cites the ANSI/AAMI EC57 and IEC 60601-2-47 standards without databases, record counts or per class figures, and the second reports testing to the same standards and statistical equivalence with the prior version, again without figures. The company cites 99 percent accuracy across the cleared arrhythmia classifications from a retrospective study reported to FDA, and results in minutes rather than days, with no database, record count, reference standard or denominator on its pages.

No peer reviewed evaluation in deployed settings, comparison against cardiologist over read, or named customer outcome is published.

A vendor reported percentage without method or denominator.

Nothing is published on how patient recordings move through the service. The company's privacy policy (AccurKardia, Inc., updated 12 December 2024) covers the website and related services and expressly excludes protected health information governed by HIPAA and separate agreements such as business associate agreements; its retention, de-identification and safeguards language therefore describes website and business contact data, not ECG recordings. No retention period, training use or de-identification standard for recordings is published anywhere on the company's pages.

The data path crosses several organizations. The platform is cloud based and ingests recordings made on other companies' patches and Holter monitors, returning interpretations to monitoring services, hospitals and diagnostic testing facilities, so one recording can pass through a device maker, a monitoring service, this interpretation layer and a reading facility, each with its own terms. Continuous rhythm data at volume is the corpus the company's investigational ECG screening tools need, and nothing says whether customer recordings feed them or whether a customer can decline. Safety engineering, such as how a corrupted or misattributed recording is caught, is not published.

The company states that patient data runs under HIPAA and business associate agreements, and publishes no document that reaches the service. Its privacy policy (updated 12 December 2024) covers the website and related services and expressly excludes protected health information governed by HIPAA and separate agreements such as business associate agreements, which is a statement of status; the agreements themselves, their tier and subcontractor terms are not published.

Customers include hospitals and independent diagnostic testing facilities, which are covered entities, so an interpretation service processing their patients' recordings works under business associate terms. The chain decides what those terms must cover: a hospital may contract with a monitoring service that in turn contracts with AccurKardia, making it a subcontractor reached only by flow down, or the relationship may be direct, and the two allocate breach notification and liability differently. The patient wearing the patch has a relationship with the ordering clinician and none with the company interpreting the recording.

A status stated with nothing published that reaches the service.

The outside check published is FDA's premarket review of the software's cybersecurity, and no attestation of how the company runs security is. The version 2.0 clearance lists cybersecurity testing among the verification activities submitted to FDA, which examines the security of the product.

It does not examine whether the company controls employee access, monitors for intrusion or would detect a breach, and no SOC 2 report, HITRUST or ISO/IEC 27001 certification, trust page or vulnerability disclosure policy is published. The holding is longitudinal rhythm data on identified patients at volume, accumulated from other companies' devices.

Two clearances are identifiable with their versions and indication, and the uses the company shows run past that indication. The AccurECG Analysis System was cleared under K223013 (23 June 2023) and version 2.0 under K252361 (22 December 2025, announced January 2026), both product code DPS, Class II, applicant AccurKardia, Inc., and the home page names K252361 beside the FDA cleared label. The version 2.0 indication covers qualified healthcare professionals assessing recorded ambulatory ECG data from adults 22 and older prescribed outpatient cardiac monitoring for up to ten days, from single lead recordings with wet silver chloride electrodes in named formats, and excludes life supporting systems and ECG monitor and alarm devices; results are not to be the sole means of diagnosis.

The company's site shows mobile cardiac telemetry devices and event recorders alongside patches and Holter monitors as recording sources. Mobile cardiac telemetry runs as a monitor with near real time alerting and often beyond ten days, which the indication does not reach, so a buyer using AccurECG in that service has to match the use to the cleared scope. The cleared label sits on the analysis software, which is the product the clearance covers. Two investigational ECG based screening tools, AK+ Guard and AK-AVS, hold Breakthrough Device Designation and are research use only.

Clearances named, with the marketed sources outside the scope they cover.

The algorithm is tested against recognized standards and no subgroup results are published. The version 2.0 clearance reports performance evaluation to the ANSI/AAMI EC57 and IEC 60601-2-47 standards and statistical equivalence with the prior version, and the company attributes its 99 percent accuracy figure to a retrospective study reported to FDA; the summary reports no subgroup analysis.

ECG interpretation varies by sex, age and body habitus, and a device agnostic product adds variation by recording source: patches and Holter monitors differ in sampling rate, filtering, electrode design and noise before any patient variation, so performance can follow procurement rather than physiology without the clinician seeing it. Nothing describes how the system flags a recording it should not interpret.

A qualified professional reviews every interpretation, and no error rate with a denominator is published. Both clearances indicate AccurECG for qualified healthcare professionals assessing recorded ambulatory ECG data and state that automated results are not intended to be the sole means of diagnosis; the version 2.0 indication also excludes life supporting systems and ECG monitor and alarm devices. A reader therefore signs off before a result reaches the chart.

Neither filing publishes the error rate. The K223013 summary (23 June 2023) cites compliance with ANSI/AAMI EC57:2012 and IEC 60601-2-47:2012 and reports no databases, record counts, sensitivity or positive predictive value; the K252361 summary (22 December 2025) reports EC57 testing and statistical equivalence with the prior version, also without per class figures or case counts. The company cites 99 percent accuracy across the cleared arrhythmia classifications from a retrospective study reported to FDA, with no denominator, database or reference standard on its pages. No warranty, indemnity or remedy for a wrong interpretation is published.

Every result gets professional review, and nothing stands behind the output itself.

Interoperability here means device data rather than the chart, and the formats are named. The version 2.0 clearance names the recording formats it accepts: WFDB .DAT and .HEA files, ISHNE formatted Holter .ECG files and EDF, from single lead wet electrode recordings, and the company describes the platform as device agnostic across patches and Holter monitors without device specific tuning, with flexible customer integrations for monitoring companies and device makers. No record system integration, HL7 or FHIR interface, or named device partner is published on the pages describing the product.

A single cloud deployment, described in the clearance, with no location committed. The version 2.0 summary describes a client side web interface used from a PC workstation with high speed internet and a server side algorithm running on Ubuntu Linux virtual machines in the cloud, and the company describes the platform as built for high ECG volumes with flexible customer integrations. No hosting provider, region, tenancy model, residency commitment or on premises option is published.

No pricing is published, and the buyer set makes the unit the question. The company sells interpretation software to cardiac monitoring companies, device manufacturers, hospitals and independent diagnostic testing facilities, and describes version 2.0 as built to process high volume ECG streams at lower cost; its site links terms of use.

Those buyers are different businesses: a monitoring company reselling interpretation, a device maker embedding it and a hospital consuming it directly would each expect a different commercial shape, and nothing published says whether one model or several is offered. Where interpretation reaches a clinician inside a partner's product, the clinician may not know whose algorithm produced it.

The validated scope is named precisely in the clearance, and the marketed sources run past it. The version 2.0 indication covers adults 22 and older prescribed outpatient cardiac monitoring for up to ten days, single lead recordings from wet silver chloride electrodes, assessed by qualified healthcare professionals in office or clinical settings, across 13 rhythm classifications, which is narrow and supported by FDA review.

The company's site shows mobile cardiac telemetry devices, event recorders, patches and Holter monitors as sources, and patches and telemetry commonly run past ten days or alert in real time, which the indication does not cover. The investigational screening tools extend ECG toward other conditions and are research use only.