patent.fyi
Applications  /  US 2026/0354802
United States Patent Application Publication

US 2026/0354802 A1

Published
Nov 12, 2026
Filed
Jul 20, 2026
Earliest date on the face
Aug 30, 2024
Classification
A61B 8/5284
First published here
Nov 19, 2026

Generated automatically. Not reviewed by a person. Not legal advice or an opinion.

Section 1 · machine-generated

What if?

Questions about how individual parts of claim 1 might be changed. A question may describe something different from what the claim language recites, or a version of it; no view is offered on which.

1.1Ventricular ultrasound imaging

What ifWhat if the system acquired cardiac ultrasound images and calculated ejection fraction from ventricular volumes?
Claim language“obtaining electrocardiogram (ECG) data”

A phased-array transducer records apical four-chamber and two-chamber cine loops. A workstation traces the left ventricular cavity at end diastole and end systole, estimates both volumes with a biplane disk method, and divides their difference by the end-diastolic volume.

1.2Fixed ECG scoring rules

What ifWhat if a fixed ECG point table replaced the trained models?
Claim language“a plurality of trained artificial intelligence (AI) models”

A signal processor identifies QRS onset and offset and measures QRS duration, Q-wave duration, and R-wave amplitude by lead. A lookup table assigns points to specified intervals of those measurements, and an accumulator sums the points into a low-EF screening score.

1.4Signal quality selects one model

What ifWhat if ECG signal quality selected the single model output reported for each recording?
Claim language“a combination of the scores”

A quality module measures clipping and baseline drift and checks which leads are present. A routing table selects a 12-lead classifier or a classifier configured for the available lead subset, and the system reports the selected classifier's score.

1.5Shared cutoff for all patients

What ifWhat if every patient received the same operating point threshold?
Claim language“based on a plurality of patient characteristics”

During validation, a threshold sweep identifies the cutoff that yields a prespecified sensitivity on a pooled validation set with known EF. The inference service stores the cutoff as a single configuration value and loads it for every recording.

1.6Score ordered imaging queue

What ifWhat if the overall score set a patient's position in an imaging queue?
Claim language“comparing the overall score of low EF to the operating point threshold”

A scheduler sorts pending patients by descending low-EF score and uses referral time to break ties. It fills available echocardiography slots from the top of the queue, and the imaging examination supplies the EF measurement.

Section 2 · machine-generated

Variations

Further arrangements of the same subject matter, described here as of the publication date at the foot of this page.

Other field of use

Canine screening in veterinary clinics

A veterinary recorder acquires a six-lead electrocardiogram from a resting dog. Three models trained on canine recordings paired with echocardiographic measurements each score the recording for low left ventricular ejection fraction. Software averages the scores and retrieves a threshold from a table indexed by body mass band and thoracic conformation. The workstation displays a low ejection fraction indication when the average meets or exceeds the threshold.

Component substitution

Logistic regression for score fusion

A logistic regression model replaces the arithmetic mean used to combine scores from trained electrocardiogram models. Each model receives the same recording, and its score occupies a fixed input position in the regression. The regression is fitted on held-out model scores paired with low ejection fraction labels from echocardiography and outputs one probability. The software compares that probability with a threshold indexed by rhythm class and ventricular conduction category.

Parameter or scale variation

Thirty minute ambulatory analysis windows

A two-lead ambulatory recorder samples both channels at 250 samples per second. Every 30 minutes, software selects up to six 10-second segments that pass lead contact and motion artifact checks. Four models trained on ambulatory recordings each score every selected segment, and the software averages the four scores per segment before taking the median across segments. It compares the median with a threshold indexed by resting heart rate band and rhythm class and reports insufficient signal when fewer than three segments pass.

Combination with existing systems

Integration with an echocardiography worklist

An electrocardiogram archive supplies the waveform and encounter identifier to an inference service when a recording is finalized. The service combines scores from three trained models and compares the result with a threshold selected by rhythm class and recorded heart failure history. For an above-threshold result, the service matches the encounter identifier to an existing echocardiography order. It adds the score and recording timestamp to the corresponding worklist entry for review.

Automated or data-driven variant

Thresholds selected from waveform measurements

A rhythm classifier and a beat delineation algorithm process the same electrocardiogram supplied to the ejection fraction models. The classifier labels the rhythm, and the delineation algorithm measures the median duration of the ventricular depolarization complexes. Software uses both measurements to index a threshold table derived from recordings paired with echocardiographic measurements. It compares the combined model score with the selected threshold and records the rhythm label, measured duration, and resulting indication.

Signal quality weighted model scores

Three trained models receive the same 12-lead recording, with each model using a different group of leads. A signal quality estimator measures clipping, baseline drift, and high-frequency noise in each lead and assigns each model a weight from the quality of its lead group. Software forms a weighted mean of the model scores and reports insufficient signal if all weights are zero. It compares the weighted mean with a threshold selected by rhythm and ventricular pacing status.

Specific version of a part

Continuous ejection fraction estimate

A regression model takes a ten-second ECG waveform and uses echocardiographic EF percentages as training targets. A calibration module uses held-out prediction errors to set an uncertainty interval around the estimated percentage, and the report displays the estimate and interval.

Synchronous twelve-lead ECG capture

Ten surface electrodes feed a multichannel ECG front end that acquires the signals synchronously at 500 samples per second. The recorder stores ten seconds of 12-lead waveforms with lead identifiers and amplitude scaling in microvolts.

Two waveform classifier architectures

A one-dimensional convolutional network and a recurrent network each receive the same ten-second matrix of lead voltages. Each network trains independently on ECG records labeled by whether echocardiographic EF falls below 40 percent.

Weighted mean of calibrated scores

A model-specific calibration map converts each classifier output to an estimated probability of low EF. The inference service multiplies the calibrated scores by fixed nonnegative weights that sum to one and adds the weighted values.

Age and rhythm cutoff table

A lookup table stores operating point thresholds indexed by age band and rhythm category. Each entry holds a cutoff selected to meet a prespecified sensitivity target on validation records from that group, and the inference service retrieves the entry matching the patient's age band and recorded rhythm.

Section 3 · quoted from USPTO records

Other work

Passages reproduced word for word from two earlier-published U.S. patent documents, placed beside the language of claim 1. No characterization is offered. Each passage is marked A or B for the document it comes from; ¶ numbers are the paragraph numbers in the USPTO’s published full text of that document.

Document AUS 2017/0055870 A1
Published
Mar 2, 2017
Filed
Aug 24, 2016
Class
A61B 5/04288
Document BUS 2020/0012948 A1
Published
Jan 9, 2020
Filed
Jul 9, 2018
Class
G06N 3/126
Claim 1
Document A
Document B
1.1obtaining electrocardiogram (ECG) data for a patient;
The present invention relates to electrocardiogram (ECG) acquisition and viewing systems and, more specifically, to an ECG module receiving 12 lead ECG data from a patient and communicating wirelessly with a computer device for displaying the ECG data and uploading ECG data to a network patient database.
A · US 2017/0055870 A1 · ¶ 3
1.2providing the ECG data as input to a plurality of trained artificial intelligence (AI) models, wherein each trained AI model is trained to determine a score of low ejection fraction (EF) based on the ECG data;
Embodiments of the present invention recognize that an ensemble, indicating a plurality of models learnt from training data, are utilized for predicting an output from inputs, in the context of predictive analytics.
B · US 2020/0012948 A1 · ¶ 15
1.3determining, using each trained AI model, a score of low EF based on the ECG data;
An instance of a base model operates on input data to generate a score corresponding to the base model.
B · US 2020/0012948 A1 · ¶ 22
1.4determining an overall score of low EF based on a combination of the scores of low EF determined by each of the trained AI models;
As ensemble scoring employs scores corresponding to more than one model, conventional ensemble scoring requires more computing and storage resources than scoring based on single model, in exchange of enhanced accuracy from single model scoring.
B · US 2020/0012948 A1 · ¶ 34
1.5obtaining an operating point threshold for the patient based on a plurality of patient characteristics;
1.6comparing the overall score of low EF to the operating point threshold; and
The end condition 131 is a threshold condition set for a score predicted based on a set of selected base models.
B · US 2020/0012948 A1 · ¶ 22
1.7determining an indication of low EF for the patient based on the comparison.

An empty cell means only that no passage was reproduced there.

Statements in these documents
While many 12-lead ECGs are capable of making longer recordings, requiring adequate electronic storage and on/off functionality, these systems are typically unable to handle the extended data in a way which assists the physicians in making improved 12-lead ECG diagnostic statement capabilities.
A · US 2017/0055870 A1 · ¶ 6
The present invention provides an electrocardiogram (ECG) acquisition and viewing system receiving extended, e.g. five□ minute, ECG recordings and automatically extracting segments with heart arrhythmias allowing overreading cardiologists to quickly analyze the extracted segments from the larger data set.
A · US 2017/0055870 A1 · ¶ 7
As in calculating the outputs based on prediction, for example, credit scores for individual financial trustworthiness, in order for the prediction to be reliably accurate, the inputs representing numerous factors are gathered from various sources and are heavily computed in predicting the outputs.
B · US 2020/0012948 A1 · ¶ 2
An ensemble, referring to a set of models, by combining a number of selected models formulating respective hypothetical correlations, attempts to form a comprehensive correlation between the inputs and the outputs, and accordingly, to produce an accurate prediction that is improved from other predictions produced based on one model.
B · US 2020/0012948 A1 · ¶ 2
Cite aspatent.fyi, US 2026/0354802 A1, claim 1, first published 2026-11-19T04:13:02-05:00, https://patent.fyi/US20260354802/, SHA-256 b99b236ec5fac34cea8f2903a5ed439ba314d1844fedbc2cd4d41d5b575bba54
First published
Permanent addresshttps://patent.fyi/US20260354802/
Content fingerprint (SHA-256)b99b236ec5fac34cea8f2903a5ed439ba314d1844fedbc2cd4d41d5b575bba54
Methodv2026-11-18.2 · automated, not reviewed by a person
Correctionscorrections@patent.fyi