ECGWmark
A deep watermarking framework for electrocardiogram (ECG) signals. It employs a fully convolutional, quantization-free design and uses a continuous full-resolution latent space, a frame-level localization mechanism, and differentiable physiological loss functions to enable secure embedding of patient metadata and extraction robust to temporal desynchronization.
Timeline (last 90 days)
ECGWmark is a fully convolutional, quantization-free deep watermarking framework for medical time-series signals.
ECGWmark operates in a continuous, full-resolution latent space, avoiding the information loss caused by quantization.
ECGWmark introduces a localized frame-level mechanism to address temporal desynchronization.
ECGWmark introduces a differentiable physiological loss function guided by a clinical classifier to constrain the model to preserve the diagnostic semantics of the signal.
ECGWmark was validated on the MIT-BIH database.
Under the cyclic temporal shift conditions evaluated, ECGWmark maintains a channel bit extraction accuracy above 90%.
In cross-patient diagnostic evaluation, Class-V sensitivity improved from 71.97% to 76.50%.
In cross-patient diagnostic evaluation, Macro-F1 slightly decreased from 35.25% to 34.71%.
In cross-patient diagnostic evaluation, Class-S sensitivity dropped sharply from 8.89% to 1.39%.
Watermark embedding had inconsistent effects across different diagnostic classes, with recognition capability improving for some classes while sensitivity declined for others.
1 more timeline events
All Facts (11)
ECGWmark is a fully convolutional, quantization-free deep watermarking framework for medical time-series signals.
50%UnverifiedECGWmark operates in a continuous, full-resolution latent space, avoiding the information loss caused by quantization.
50%UnverifiedECGWmark introduces a localized frame-level mechanism to address temporal desynchronization.
50%UnverifiedECGWmark introduces a differentiable physiological loss function guided by a clinical classifier to constrain the model to preserve the diagnostic semantics of the signal.
50%UnverifiedECGWmark was validated on the MIT-BIH database.
50%UnverifiedUnder the cyclic temporal shift conditions evaluated, ECGWmark maintains a channel bit extraction accuracy above 90%.
50%UnverifiedIn cross-patient diagnostic evaluation, Class-V sensitivity improved from 71.97% to 76.50%.
50%UnverifiedIn cross-patient diagnostic evaluation, Macro-F1 slightly decreased from 35.25% to 34.71%.
50%UnverifiedIn cross-patient diagnostic evaluation, Class-S sensitivity dropped sharply from 8.89% to 1.39%.
50%UnverifiedWatermark embedding had inconsistent effects across different diagnostic classes, with recognition capability improving for some classes while sensitivity declined for others.
50%UnverifiedECGWmark provides a valuable new paradigm for secure watermarking of medical time-series signals, but further work on class-level balance is still needed before clinical deployment.
50%