patent.fyi
Applications  /  US 2026/0354727
United States Patent Application Publication

US 2026/0354727 A1

Published
Nov 12, 2026
Filed
Sep 20, 2023
Earliest date on the face
Sep 20, 2022
Classification
A61B 5/308
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.

1.1Record from the esophagus

What ifWhat if the system acquired cardiac electrical signals from the esophageal wall through a bipolar catheter?
Claim language“from a human skin surface”

Two ring electrodes on the catheter contact the esophageal mucosa at a position that produces a clear ventricular deflection. A differential amplifier and an analog-to-digital converter record the voltage difference for ventricular beat detection.

1.2Filter forward and backward

What ifWhat if the system extracted cardiac frequency bands with forward and backward digital filtering?
Claim language“a stationary wavelet transform”

Separate bandpass filters spanning 5 to 15 Hz and 15 to 35 Hz process each buffered segment in the forward direction and then in the reverse direction. The two passes cancel the filter phase response and produce two time-aligned sample streams for beat detection.

1.2Timestamp beats after causal filtering

What ifWhat if the system used a causal bandpass filter and located each beat timestamp in a buffer of the input samples?
Claim language“a zero-phase filter”

A causal infinite impulse response bandpass filter from 8 to 25 Hz supplies candidate events from peaks in the squared filtered samples. For each candidate, the detector searches a buffered input window that extends back across the filter delay and timestamps the largest absolute excursion from the local baseline.

1.3Detect on combined detail energy

What ifWhat if the detector combined the detail coefficient layers into one energy trace before detecting any QRS complexes?
Claim language“detecting a QRS wave complex for each layer”

The processor scales each selected layer by its median absolute deviation, with a positive floor on the scaling denominator, then sums the squared coefficients at each sample. A peak detector applies an adaptive threshold to the resulting trace and enforces a 200 ms refractory interval.

1.3Merge detections by unanimity

What ifWhat if a beat required a matching detection in every monitored detail layer?
Claim language“by a weighted voting scheme”

Each layer independently generates a list of candidate timestamps. A logical coincidence gate emits a heartbeat when every monitored layer supplies a candidate within the same 50 ms window and assigns the median candidate timestamp to that beat.

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

Surface cardiac monitoring in horses

Three adhesive electrodes on clipped skin form a base-apex cardiac lead, with two sensing electrodes and one reference electrode. An isolated differential amplifier digitizes the lead for a zero-phase stationary wavelet filter bank that produces time-aligned detail and approximation coefficients. Each detail layer flags peaks in its rectified coefficient envelope, and a weighted vote groups candidates within 60 ms to report QRS timestamps.

Component substitution

Centered finite impulse response filter bank

A parallel bank of symmetric finite impulse response filters replaces the stationary wavelet stage and receives digitized surface cardiac samples. The bank uses 129-tap bandpass filters with passbands of 5–12 Hz, 10–22 Hz, and 18–35 Hz at 500 samples per second, plus a symmetric low-pass filter for the baseline component. The processor assigns each output to the center sample of its input window. Each bandpass channel detects local energy peaks, and a weighted vote combines candidates within 40 ms.

Parameter or scale variation

Three detail layers at 250 hertz

The acquisition circuit samples one skin-surface cardiac lead at 250 samples per second, and a zero-phase stationary wavelet stage produces three detail layers and one approximation layer. Each detail detector squares its coefficients, sums them over a 20-sample moving window, and marks local maxima above an adaptive threshold. Candidates within 10 samples form a cluster, with weights of 1, 2, and 3 from the finest to the coarsest layer. The processor reports a heartbeat when the cluster scores at least 4, using the highest-weight candidate as its timestamp.

Combination with existing systems

Cardiac timing for pulse transit measurement

A skin-surface cardiac lead and a finger optical pulse sensor acquire samples on a shared clock. The cardiac channel uses a zero-phase stationary wavelet stage, QRS detection in each detail layer, and weighted voting to produce a heartbeat timestamp. The optical channel marks pulse onset at the intersection of the maximum-upstroke tangent and the preceding trough level. The processor pairs each heartbeat with the first optical onset 80–600 ms later and records the interval as pulse arrival time.

Automated or data-driven variant

Motion conditioned voting across detail layers

A three-axis accelerometer mounted on the electrode housing records motion alongside the skin-surface cardiac signal. A zero-phase stationary wavelet stage produces detail and approximation layers, and each detail detector submits timestamped QRS candidates. Calibration on labeled recordings assigns each detail layer a weight for each of three motion ranges using the fraction of its candidates that match reference beats. During operation, software selects the weight set from the root-mean-square acceleration after gravity removal over the preceding second and reports candidate clusters whose summed weights exceed a calibrated threshold.

Learned QRS detection in detail layers

A zero-phase stationary wavelet processor decomposes a digitized skin-surface cardiac lead into time-aligned detail and approximation layers. A separate one-dimensional convolutional classifier slides a 160 ms window across each detail layer and assigns its center sample a QRS probability. Training labels mark centers within 20 ms of annotated QRS times as positive, and each classifier uses the same coefficient normalization in training and operation. The processor groups local probability peaks within 40 ms, sums the peak probabilities using fixed layer weights, and reports a beat when the sum crosses a threshold set on held-out recordings.

Specific version of a part

Gel electrodes on the torso

Two silver–silver chloride electrodes with conductive hydrogel attach below the right clavicle and on the left lower torso. A third torso electrode provides a common-mode reference, and a differential front end digitizes the sensing pair at 500 samples per second.

Centered undecimated wavelet stages

A stationary wavelet transform convolves the input with symmetric low-pass and high-pass analysis kernels centered on each sample. Each successive stage doubles the tap spacing and filters the preceding approximation without decimation, producing detail and approximation sequences on the original sample grid.

Adaptive peaks in each layer

A detector squares the coefficients in each detail layer and averages the result over a centered 80 ms window. Each layer uses its own threshold derived from recent signal and background peak estimates and accepts candidate peaks at least 200 ms apart.

Weighted votes within time windows

Three detail layers receive voting weights of 0.5, 0.3, and 0.2, and each layer contributes at most one vote per 50 ms window. A weight sum of at least 0.7 confirms a beat, and the weighted mean of the contributing timestamps sets the reported beat time.

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 11,311,245 B2
Published
Apr 26, 2022
Filed
Nov 30, 2020
Class
A61B 5/7214
Document BUS 11,389,101 B2
Published
Jul 19, 2022
Filed
Jun 21, 2019
Class
A61B 5/364
Claim 1
Document A
Document B
1.1step 1, acquiring surface electrophysiological signals from a human skin surface;
Electrocardiography (ECG) is a medical diagnostic technology in which the electronic physiological activity of the heart is recorded through the electrodes on the skin and recorded in time.
A · US 11,311,245 B2 · ¶ 3
For the card-type, the left hand and the right hand are required to contact the electrodes at the same time when acquiring signals, and for the form of a ring, a bracelet, a watch or the like, the finger not wearing anything is required to be placed on the electrode sensor, such as the ring, the bracelet, the watch or the like, when acquiring signals.
B · US 11,389,101 B2 · ¶ 58
1.2step 2, processing the surface electrophysiological signals using a zero-phase filter based on a stationary wavelet transform, to obtain a number of layers of wavelet detail coefficients and approximate coefficients;
In the following, taking wavelet decomposition filtering as an example, the wavelet decomposition can divide the ECG signal into 10 layers, and respectively sets the detail coefficients for the first and second layer corresponding to the high-frequency noise and the detail coefficients for the ninth and tenth layer corresponding to the low-frequency baseline drift to zero. Then, the processed wavelet detail coefficients is reconstructed to acquire the denoised ECG signal.
B · US 11,389,101 B2 · ¶ 60
1.3step 3, detecting a QRS wave complex for each layer of wavelet detail coefficients, and obtaining a heartbeat detection result from each layer of wavelet detail coefficients by a weighted voting scheme.
After the SWT algorithm is performed for the second ECG signal S2 received in the initial period, the first calculation circuit 121 may obtain the second ECG signal S2 received in the initial period in the lowest level, and perform a QRS detection for the second ECG signal S2 received in the initial period in the lowest level to generate a QRS template signal SQRS.
A · US 11,311,245 B2 · ¶ 31
Taking a quadratic spline wavelet decomposition method as an example, the ECG signal is decomposed by four scales with the quadratic spline wavelet, and the corresponding QRS complex energy distribution is the largest on the 23 scale. On the 23 scale, the R wave peak position may be obtained by zero-crossing between a pair of positive and negative maximum values.
B · US 11,389,101 B2 · ¶ 62

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

Statements in these documents
Therefore, when the attenuation rate of the filter is too high, the real ECG signal may be also be filtered.
A · US 11,311,245 B2 · ¶ 4
The calculation circuit is coupled to the filter to receive the second ECG signal and processes the second ECG signal to generate a noise signal, and subtracts the noise signal from the second ECG signal to filter the noise signal in the first ECG signal.
A · US 11,311,245 B2 · ¶ 6
The present application provides a method for identifying arrhythmia, a device for identifying arrhythmia, and a computer readable medium.
B · US 11,389,101 B2 · ¶ 5
Cite aspatent.fyi, US 2026/0354727 A1, claim 1, first published 2026-11-19T02:47:19-05:00, https://patent.fyi/US20260354727/, SHA-256 8e039021581c5abfac2c2cd1fd1a324e7ad4b0a78655b67689e129a5cec2dfc7
First published
Permanent addresshttps://patent.fyi/US20260354727/
Content fingerprint (SHA-256)8e039021581c5abfac2c2cd1fd1a324e7ad4b0a78655b67689e129a5cec2dfc7
Methodv2026-11-18.2 · automated, not reviewed by a person
Correctionscorrections@patent.fyi