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.
No entries.
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
Signal quality scoring during equine exercise
A girth-mounted recorder acquires a differential electrocardiogram from electrodes on the horse’s thorax during exercise. Veterinary reviewers label each segment according to whether they can identify and time individual ventricular complexes. A one-dimensional convolutional network trains on the labeled segments and outputs a probability that each incoming segment supports reliable beat timing.
Component substitution
Quality scoring with gradient boosted trees
Gradient boosted decision trees replace the waveform classifier and receive one feature vector per electrocardiogram segment. A feature extractor measures clipping fraction, baseline drift, beat detection agreement and the median correlation between detected beats and a segment template. Training pairs each feature vector with its reader-assigned quality label, and a calibration function fitted on held-out recordings converts the classifier output to a probability of readability.
Parameter or scale variation
Quality scoring with short overlapping windows
A single-lead recorder samples the electrocardiogram at 250 samples per second and forms 4-second windows of 1,000 samples with a 1-second stride. Reviewers label each window according to whether they can time the ventricular complexes throughout it. A temporal convolutional classifier trains on the labeled windows using recording-level splits and emits a new quality probability every second after the first window fills.
Combination with existing systems
Bedside telemetry with lead quality overlays
A bedside telemetry gateway mirrors incoming electrocardiogram samples to a quality service and attaches a lead identifier and timestamp to each segment. A classifier trained on reader-labeled segments scores each lead independently. The central monitoring station aligns the scores with the displayed waveforms and highlights lead intervals whose scores fall below a threshold selected on validation recordings.
Automated or data-driven variant
Motion informed selection for reader review
A wearable recorder timestamps electrocardiogram samples and three-axis accelerometer samples on a shared clock. Software groups synchronized segments into low, medium and high motion bands using the root mean square of high-pass-filtered acceleration and selects segments with classifier probabilities closest to 0.5 within each band. Reviewers label the selected waveforms for readability, and the training service adds those labels to the training set for the next model update.
Automated labeling from beat morphology
A labeling program runs two beat detectors on each electrocardiogram segment, aligns their matched beats at the detected peaks and forms a median beat template. The program assigns a readable label when it finds at least three matched beats, detector agreement reaches 90 percent, and median beat-to-template correlation reaches 0.85. It assigns an unreadable label to segments with sustained clipping and sends all remaining segments to a review queue. A temporal classifier trains on the automatically labeled segments and uses independently reviewed recordings for validation.
Cite aspatent.fyi, US 2026/0354768 A1, claim 1, first published 2026-11-19T03:31:31-05:00, https://patent.fyi/US20260354768/, SHA-256 61f267d1cc0e217ec3d095b24c908fef9bead3c0efa595232a861ce54f76d5f4