Application of Convolutional Neural Network for Decoding of 12-Lead Electrocardiogram from a Frequency-Modulated Audio Stream (Sonified ECG).

Research of novel biosignal modalities with application to remote patient monitoring is a subject of state-of-the-art developments. This study is focused on sonified ECG modality, which can be transmitted as an acoustic wave and received by GSM (Global System for Mobile Communications) microphones. Thus, the wireless connection between the patient module and the cloud server can be provided over an audio channel, such as a standard telephone call or audio message. Patients, especially the elderly or visually impaired, can benefit from ECG sonification because the wireless interface is readily available, facilitating the communication and transmission of secure ECG data from the patient monitoring device to the remote server. The aim of this study is to develop an AI-driven algorithm for 12-lead ECG sonification to support diagnostic reliability in the signal processing chain of the audio ECG stream. Our methods present the design of two algorithms: (1) a transformer (ECG-to-Audio) based on the frequency modulation (FM) of eight independent ECG leads in the very low frequency band (300-2700 Hz); and (2) a transformer (Audio-to-ECG) based on a four-layer 1D convolutional neural network (CNN) to decode the audio ECG stream (10 s @ 11 kHz) to the original eight-lead ECG (10 s @ 250 Hz). The CNN model is trained in unsupervised regression mode, searching for the minimum error between the transformed and original ECG signals. The results are reported using the PTB-XL 12-lead ECG database (21,837 recordings), split 50:50 for training and test. The quality of FM-modulated ECG audio is monitored by short-time Fourier transform, and examples are illustrated in this paper and supplementary audio files. The errors of the reconstructed ECG are estimated by a popular ECG diagnostic toolbox. They are substantially low in all ECG leads: amplitude error (quartile range RMSE = 3-7 µV, PRD = 2-5.2%), QRS detector (Se, PPV > 99.7%), P-QRS-T fiducial points' time deviation (<2 ms). Low errors generalized across diverse patients and arrhythmias are a testament to the efficacy of the developments. They support 12-lead ECG sonification as a wireless interface to provide reliable data for diagnostic measurements by automated tools or medical experts.

Audio-Conversion of Biomedical Signals - A Possible Approach to Improve Remote Monitoring of Elderly and Visually Impaired People.

Long-term remote patients monitoring implies minimal discomfort and reliability throughout the study period. These requirements are fulfilled by portable (wearable) patient devices, with low consumption, which transmit data wirelessly, at a short distance, to a mobile communication device (GSM) and through it, to a remote end recipient - doctor, medical center or a hospital server. The data transfer technology requires the monitored person to perform a sequence of actions, such as: selecting the appropriate application on the mobile phone, establishing a connection between the patient module and the phone, recording the data in the phone's memory, starting the data transfer from the phone to the final receiver. Practice shows that often this sequence of activities is difficult for elderly people and especially for visually impaired people, which as a result compromises the remote monitoring process. In this paper are presented an approach and conceptual implementation of a system for remote monitoring of cardiac activity, using the most popular way of remote connectivity - voice (sound) communication. In addition to the ease of use, this type of communication does not require special data protection, due to the lack of RF interfaces for short-distance data transmission. The presented results of laboratory studies, as well as conducted tests under medical supervision of patients in a cardiology clinic, confirm the workability of the proposed approach for remote monitoring of patients by audio conversion of the ECG signal.

Electrochemical Properties of PEDOT:PSS/Graphene Conductive Layers in Artificial Sweat.

Electrodes based on PEDOT:PSS are gaining increasing importance as conductive electrodes and functional layers in various sensors and biosensors due to their easy processing and biocompatibility. This study investigates PEDOT:PSS/graphene layers deposited via spray coating on flexible PET substrates. The layers are characterized in terms of their morphology, roughness (via AFM and SEM), and electrochemical properties in artificial sweat using electrochemical impedance spectroscopy (EIS) and cyclic voltammetry (CV). The layers exhibit dominant capacitive behavior at low frequencies, with cut-off frequencies determined for thicker layers at 1 kHz. The equivalent circuit used to fit the EIS data reveals a resistance of about three orders of magnitude higher inside the layer compared to the charge transfer resistance at the solid/liquid interface. The capacitance values determined from the CV curves range from 54.3 to 122.0 mF m-2. After 500 CV cycles in a potential window of 1 V (from -0.3 to 0.7 V), capacitance retention for most layers is around 94%, with minimal surface changes being observed in the layers. The results suggest practical applications for PEDOT:PSS/graphene layers, both for high-frequency impedance measurements related to the functioning of individual organs and systems, such as impedance electrocardiography, impedance plethysmography, and respiratory monitoring, and as capacitive electrodes in the low-frequency range, realized as layered PEDOT:PSS/graphene conductive structures for biosignal recording.

Online digital filter and QRS detector applicable in low resource ECG monitoring systems.

The present work describes fast computation methods for real-time digital filtration and QRS detection, both applicable in autonomous personal ECG systems for long-term monitoring. Since such devices work under considerable artifacts of intensive body and electrode movements, the input filtering should provide high-quality ECG signals supporting the accurate ECG interpretation. In this respect, we propose a combined high-pass and power-line interference rejection filter, introducing the simple principle of averaging of samples with a predefined distance between them. In our implementation (sampling frequency of 250 Hz), we applied averaging over 17 samples distanced by 10 samples (Filter10x17), thus realizing a comb filter with a zero at 50 Hz and high-pass cut-off at 1.1 Hz. Filter10x17 affords very fast filtering procedure at the price of minimal computing resources. Another benefit concerns the small ECG distortions introduced by the filter, providing its powerful application in the preprocessing module of diagnostic systems analyzing the ECG morphology. Filter10x17 does not attenuate the QRS amplitude, or introduce significant ST-segment elevation/depression. The filter output produces a constant error, leading to uniform shifting of the entire P-QRS-T segment toward about 5% of the R-peak amplitude. Tests with standardized ECG signals proved that Filter10x17 is capable to remove very strong baseline wanderings, and to fully suppress 50 Hz interferences. By changing the number of the averaged samples and the distance between them, a filter design with different cut-off and zero frequency could be easily achieved. The real-time QRS detector is designed with simplified computations over single channel, low-resolution ECGs. It relies on simple evaluations of amplitudes and slopes, including history of their mean values estimated over the preceding beats, smart adjustable thresholds, as well as linear logical rules for identification of the R-peaks in real-time. The performance of the QRS detector was tested with internationally recognized ECG databases (AHA, MIT-BIH, European ST-T database), showing mean sensitivity of 99.65% and positive predictive value of 99.57%. The performance of the presented QRS detector can be highly rated, comparable and even better than other published real-time QRS detectors. Examples representing some typical unfavorable conditions in real ECGs, illustrate the common operation of Filter10x17 and the QRS detector.

Real-time detection of pathological cardiac events in the electrocardiogram.

The development of accurate and fast methods for real-time electrocardiogram (ECG) analysis is mandatory in handheld fully automated monitoring devices for high-risk cardiac patients. The present work describes a simple software method for fast detection of pathological cardiac events. It implements real-time procedures for QRS detection, interbeat RR-intervals analysis, QRS waveform evaluation and a decision-tree beat classifier. Two QRS descriptors are defined to assess (i) the RR interval deviation from the mean RR interval and (ii) the QRS waveform deviation from the QRS pattern of the sustained rhythm. The calculation of the second parameter requires a specific technique, in order to satisfy the demand for straight signal processing with minimum iterations and small memory size. This technique includes fast and resource efficient estimation of a histogram matrix, which accumulates dynamically the amplitude-temporal distribution of the successive QRS pattern waveforms. The pilot version of the method is developed in Matlab and it is tested with internationally recognized ECG databases. The assessment of the online single lead QRS detector showed sensitivity and positive predictivity of above 99%. The classification rules for detection of pathological ventricular beats were defined empirically by statistical analysis. The attained specificity and sensitivity are about 99.5% and 95.7% for all databases and about 99.81% and 98.87% for the noise free dataset. The method is applicable in low computational cost systems for long-term ECG monitoring, such as intelligent holters, automatic event/alarm recorders or personal devices with intermittent wireless data transfer to a central terminal.