MEASUREMENT CHANNELS’ INFORMATION RELATIONSHIPS IN BODY SURFACE POTENTIAL MAPPING RELACJE INFORMACYJNE KANAŁÓW POMIAROWYCH W MAPOWANIU POTENCJAŁU POWIERZCHNI CIAŁA

Body Surface Potential Mappings (BSPM) measurements allow heart rate monitoring. Based on measurements obtained from BSPM, defects and heart rhythm disorders can be detected. Due to the multitude of data obtained from BSPM measurements, the crucial aspect in the potential test signals analysis on the human body is the automation of the disturbance detection process. This automation process involves the use of advanced algorithms and machine learning techniques to analyze the recorded data and identify patterns indicative of heart rhythm disorders. In the article, the information dependencies in the measurement channels were analyzed and presented. PCA analysis was used to determine informational relationships between the channels. The study was carried out on the data from simulations made for the BSPM measuring vest on a simulation system created for this purpose and based on the ECG signal generator. As a result of the conducted research indicated informational relationships between the channels for signals with disturbances: Atrial fibrillation, Brachycardia, Normal signal, PVC, Tachycardia, Ventricular Fibrillation, and Ventricular Tachycardia.


Introduction
ECG examination is a primary diagnostic procedure for detecting heart rhythm disorders.It is based on the electrical potential analysis at ten measurement points (leads/electrode).It is possible to identify distortions in the ECG signal characteristic of different abnormalities using diagnostic charts obtained from measurements.Some defects require the connection of a heart rate monitor for a more extended period.Automation of the analysis process and automatic detection allow for efficient analysis of large amounts of recorded data during extended measurements (Przysucha, 2020;Kłosowski, 2020).
It is difficult to continuously detect abnormal heart rhythms automatically.Measurement must be performed to minimise uncertainty related to the poor contact between the electrode and the body.We also need to be able to correct signals affected by static electricity generated by the patients' clothes.
To improve the measurements, an electrical potential is measured on the entire chest (Robinson and Curzen, 2009).Such a method is called Body Surface Potential Mapping (BSPM) and has been done with 62 or 102 electrodes in the past (Rodrigo, 2017).This research has been carried out since the 1960s (Lux, 2010, Polak-Jonkisz, 2009).However, only in connection with machine learning and the rising computing power of machines can research on the automatic classification of disturbances be conducted (Polak-Jonkisz, 2009;Hoekema, 2004;Hänninen, 2003;SippensGroenewegen, 2000;Rymarczyk, 2020).
The possibilities of BSPM are much more extensive compared to classical ECG tests (Korhonen, 2009;Simelius, 2003).In BSPM, obtaining the potential from electrodes placed on the entire chest is possible.The most critical are the channels placed at the heart level in front and the back of the chest.They allow for the data acquisition of heart activity from the back walls of the ventricles.Data from these regions is valuable for detecting abnormalities such as coronary artery seaside or past myocardial infarctions (Zarychta, 2007).The BSPM was also helpful in diagnosing acute coronary artery occlusion (Daly, 2013).
The problem of investigating the information dependence of BSPM measurement signals has been raised in many papers.However, the studies were usually limited to a particular disease entity.For instance, in (Weber, 2011), the authors focused on determining a representative set of latent variables (Principal Components) for modifications caused by changes in tissue conductivity.Latent variables were defined for blood, skeletal muscle, lung and adipose tissue changes.In (Mark Potse, 2017), a study was carried out on the analysis of PCA for Ventricular Fibrillation disorder for measurements where sampling took place every half second.Three latent variables were selected to represent a set of measurement signals.The work (Bonizzi, 2009) focused on PCA analysis for Atrial Fibrillation disorder.The results of PCA analysis were used to classify cases with and without the disorder.In a paper (Donnelly, 2006), signal measurements from patients with old inferior myocardial infarction and healthy persons were examined.First, PCA was used to reduce the dimensions of the measurement space.Then, based on the obtained latent variables, case classification was performed.There are also many papers about PCA in ECG, for example (Castells, 2007).
The main aim of the research is to verify and indicate the advantages of PCA analysis in detecting heart diseases, especially in the context of the classification of ECG signals.The paper addresses the issue of determining and comparing latent variables for a large class of illnesses by identifying locations on the patient's body connected with latent variables obtained through PCA analysis.This information may be beneficial in the event of examinations utilizing, for example, 12-channel ECGs, in which the areas responsible for recognising the disease entity are positioned in a different location than the electrodes used in a conventional ECG.Moreover, this manuscript discusses the issue of information reliance between channels and identifies the importance of analysis of BSPM data during measurements to detect invalid measurements.Understanding the mechanics of electrical potential diffusion in the patient's body requires understanding information linkages.
This research provides a starting point for clinical studies related to the testing of the BSPM measurement vest.Knowing the information dependencies created as a result of performing the PCA analysis on simulation data, there is a possibility of verifying measurements from patients and detecting defects of the measurement system or noise possibly occurring in the measurement path.The information relations in PCA channels performed on simulation data and, thus, on model data will enable the validation of the operation of the measuring system.In case of faulty signal registration from a particular electrode or signal disturbances, the information relations between the channels resulting from the conducted PCA analysis will be disturbed.The research was conducted using data simulated from the certificated ECG simulator.The simulation system was created as a part of the test mechanism for the designed BSPM vest with 102 electrodes.The conducted tests showed that the reduction of PCA dimensions increases classification efficiency through the increased ability of the model to reduce noise and increase classification speed.A not insignificant advantage of the complexity reduction achieved by using PCA is the simplification of the BSPM system design (both hardware and software), which, in the case of commercialisation of the tested solution, positively affects the price competitiveness of the device.
This manuscript is an expansion of the study (Wójcik, 2022).In particular, the described research was performed using an upgraded phantom.In the survey (Wójcik, 2022), the measuring hardware was directly connected to resistor mesh, while in the work presented in this paper, we use a much more sophisticated fandom where with a mannequin torso with a copper imitation of human skin and textile electrodes Improved hardware was also used.The differences include a new filtering system and communication protocols inside the measuring board.
The work presented in this paper is part of a project whose primary goal is to develop a mobile tomographic system for 3D imaging of the heart and lungs using BSPM and electrical impedance tomography (EIT).
Our system is divided into two significant parts: the mobile measuring device and the intelligent cloud framework (Rymarczyk, 2019).The main module is an integral section of the mobile measuring device.Its functionality includes acquiring and preliminary processing of measurement data, calibrating active electrodes, sending the data to the web-server system, and powering the entire system.The data acquisition device (DAD) consists of intelligent clothing monitoring vital signs.It performs ECG measurements through the body surface potential mapping (BSPM) and electrical impendence tomography (EIT) and thus monitors the lungs.This wearable system is designed in such a way that it does not constrain the movement of the patient while at the same time taking reliable data.At the moment, the garment is equipped with 102 electrodes.The 32 are arranged in two rows (16 electrodes in each row) and B. PRZYSUCHA, M. KOWALSKI, M. OLESZEK, M. SUTRYK distributed uniformly in a plane around the chest perimeter.These electrodes are used for the EIT measurements.The rest of the electrodes are being used for BSPM.The second part of the DAD is the electronics connected to the electrodes responsible for sensing, filtering, and amplifying biosignals.The sensing analogue front-end unit amplifies biosignals to the analogue-to-digital converters (ADCs) level, and preliminary filtering is required.The digital processing part is a chip (SoC) system consisting of a field-programmable gate array (FPGA), microcontroller, and wireless communication modules, such as BT/BLE and Wi-Fi.
The system also contains algorithms for image reconstruction for a human chest model with lungs and heart.Electrical impedance tomography was used for this purpose.It is a non-invasive imaging method involving examining an unknown object using electric currents and appropriate measurements of voltage drops at its edge.In medical clinical trials and practice, imaging has become an essential part of diagnosing and studying the anatomy and function of the human body.The fusion of complementary methods will significantly improve diagnostic capabilities.The measurement modules implemented in the portable device, namely electrical impedance tomography and body surface potential mapping, will track lung ventilation and cardiac activity in real-time.
The system will be a world novelty in terms of functionality and the development and application of technological solutions.The main novelty of this paper is the construction of the new hardware with a unique filtering and measuring setup capable of measuring not only the BSPM signal but also 3D electrical impedance tomography (EIT).The solution is based on a custom application of the FPGA.

Research Methodology
There are many systems, methods and algorithms for data measurement and analysis (Rymarczyk, 2021;Rymarczyk, 2019;Koulountzios, 2019;Korzeniewska, 2021;Korzeniewska 2021;Banasiak, 2014;Dusek and Mikulka, 2021;Kryszyn, 2019).The BSPM vest was constructed in such a way as to be able to measure the BSPM and EIT (Rymarczyk, 2019).For the EIT, we concluded that to reconstruct a 3D image, we need to include at least two rows of electrodes.These two rows are in the middle part of the vest, and the top two are located below the armpits (see Figure 1).The rest of the electrodes are located in such a way as to have a dense grid of electrodes around the patient's body.Also, the location of electrodes overlaps the points at which classical 12-lead ECG is performed; thus, the vest can measure standard ECG.The vest consists of 102 electrodes, all of which can be used for BSPM.In addition, the innovative design of the vest (shown in Figure 2) and the use of elastic textile-silicone electrodes (shown in Figure 3) allow for sound pressure and contact with the body and, consequently, better data reading.The data for the analysis comes from the measurement signals simulation for the designed vest.The simulation system consisted of the following: The ProSim 4 ECG simulator generates an ECG-like signal for all standard and abnormal heart rhythms; the phantom is used to convert the 10-lead ECG simulator signal to 102 BSPM distributions.This was achieved by designing a net of resistors that mimicked the human body.Multiplexer systems are connected in series using FFC tape.A controller circuit board amplifies and converts signals from analogue to digital signals.

MEASUREMENT CHANNELS' INFORMATION RELATIONSHIPS IN BODY SURFACE POTENTIAL MAPPING
Due to the specifics of the measured signal (low signal-to-noise ratio, low signal amplitude), it was necessary to design and develop a custom filtering system to compare the measurements obtained to the state-of-the-art Holters.The simulator can generate a standard ECG signal, slow heart rate down to 30 BPM, increase heart rate up to 300 BPM, and various disorders such as atrial fibrillation, premature ventricular contraction, ventricular tachycardia, ventricular fibrillation, transvenous pacer pulse, second-degree AV block, third-degree AV block.The signal from the simulator is close to the natural heart rhythm -with a similar electrical amplitude level and type of noise.This device is used in hospitals to test and validate the correct operation of ECG monitors and other equipment used to monitor heart rhythm (Randazzo, 2022;Sadrawi, 2017).
A phantom was created to test the designed solution, converting signals from a 10-channel ECG simulator into a 128-channel potential distribution map compatible with the previously designed multiplexer blocks.Each measurement channel has an independent forming and pre-filtering system, thanks to which the signal fed to the analogue keys shows a much higher signalto-noise ratio than the directly measured signal.Additionally, input blocks ensured the maximum input impedance of the measurement system.The schematic of the measurement step is shown in Figure 4. Figure 4 shows the schematic diagram corresponding to the experimental setup used during the data collection.On the Left, the FLUKE ProSim4 ECG signal simulator is visible.Fluke ProSim4 is used in hospitals for periodic validation of ECG devices.It is designed to test patient monitors at the point of use.The easy and secure connection ensures the simulator can test 12-lead ECG, NIBP, IBP and respiration in seconds.Built-in test sequences, intuitive operation and one-touch screen access to most functions make the ProSim 4 an indispensable diagnostic tool for evaluating the performance of monitoring devices.The ten leads from the simulator are connected to the phantom equipped with conduction copper pads.Inside the phantom is a network of resistors that mimic the human body's continuous resistance.
The simulated potentials are measured via removable elastic textile electrodes mounted on the vest.The signals are measured, filtered, and transmitted via USB or Wi-Fi to the computer for further analysis.The vest contains eight boards that can control the measurement of 128 channels.In the vest, only 102 are used.Each board is connected in series with the other.Finally, the last board is connected to the main EIT/BSPM board.The mainboard has the FPGA that controls electrical impedance (EIT) and BSPM.It is also equipped with Analog Digital Converters (ADC) and multiplexers.The collected data is then sent to the computer over a USB or Wi-Fi ESP32 module.The mainboard also has analog and software filters.The FPGA also plays a role in signal filtering.One of the board's goals is to deal with noise and baseline wander cancellation.
The system has been designed using analogue multiplexers coupled with filters and amplifiers that form the signal to values valid for measurements.The signal is formed after leaving the multiplexer block, and it is filtered again using an additional stopband filter cutting off 50Hz frequency and a combiner system using the resultant Wilson signal as a reference.It is passed on to ADC converters, where the last process of signal filtration takes place with the FFT filter.The measurement process is carried out cyclically on all 128 measured channels, thus obtaining an adequate sampling of the signal at the level of 1ksps.
The system was tested with the developed potential reconstruction algorithm, where the potential collected on the heart was simulated on the body.The inverse of that problem was also calculated where the potential on the heart is reconstructed from the measurement on the body, as shown in Figure 5.After leaving the multiplexer block, the signal is filtered with a band-stop filter cutting off 50Hz frequency and a combiner system using the resultant Wilson signal as a reference.Finally, such signal is passed from analogue to digital converters, where the last process of signal filtration takes place using the FFT filter.The measuring process occurs cyclically on all the available 128 measured channels, obtaining an adequate signal sampling at 1ksps.It is worth noting that although the system can measure 128 channels, only 102 were used for BSPM.The measured data is then transmitted via Ethernet to an IoT hub located on the company server cluster for analysis.
Signal simulations were performed with the following heart disorders: AFib_Fine -Atrial fibrillation; Bradycardia -heart rate below 60 times per minute; Normal -Normal sinus rhythm; PVC -Additional ventricular stimulation; Tachycardia -heart rate above 100 times per minute; VFib_Fine -Ventricular Fibrillation -suddenly attempt to contract at rates of up to 500 bpm; TachFine -Ventricular Tachycardia -broad complex tachycardia originating in the ventricles.Fragments of signals with disturbances are presented in 9. Principal component analysis (PCA) was used to find information relationships between channels (Jolliffe, 2002).Principal Component Analysis is a method of orthogonal linear transformation to a new system of variables.It consists in transforming the data matrix into a new coordinate system.Consider an X-data matrix with dimensions .In the X matrix, each column is a p-dimensional data vector from each sensor.Normalization is performed

MEASUREMENT CHANNELS' INFORMATION RELATIONSHIPS IN BODY SURFACE POTENTIAL MAPPING
on the columns to eliminate the differences between the possible differences among the values in the columns.This transformation is an l-dimensional set of p-dimensional weight vectors w (k) = (w 1 , …w p ) (k) , which converts the x (i) lines matrix X into new variables

Research Methodology
Statistical analyses and data visualisation were performed in an R environment using complot, factor extra, and ggplot2 libraries.A corrupt matrix from R was used to show the correlation between the channels in the signals and individual disorders.This matrix is created to visualise the correlation strength from the correlation matrix calculated for the channels.If the correlation is strongly positive, the given point showing the correlation between individual channels is marked as the navy-blue intensity.The correlation strength is shown as the red colour intensity if the correlation is negative.The correlation matrix structure is similar for AFib_Fine, Bradycardia, Normal, PVC, Tachycardia, and VFib_Fine disorders, as can be observed in Figures 6 and 7 b) 's correlation diagrams.Figure 10 c) shows the correlation matrix for the VTach_Fine disorder.The structure of the matrix differs significantly from that of other disorders.

MEASUREMENT CHANNELS' INFORMATION RELATIONSHIPS IN BODY SURFACE POTENTIAL MAPPING
In the case of VTach_Fine disturbance, due to the similar percentages of PC1 and PC2 variable inputs in the system of variables, significant factor loadings in PC1 and PC2 are changed.Table 5 shows the significance of the channels for the VTach_Fine disorder in terms of variables.

Discussion
We analysed differences in the significance of the factor loadings for individual disturbances compared to the standard signal.For AFib_Fine, Bradycardia, Normal, and Tachycardia VFib_Fine disorders, the structure of significant channels in the PC1 and PC2 variables was similar.
For the PC1 variable, channel 79 was an additional important channel in the tachycardia disorder.In addition, channels 41,44,45,56,58, and 60 (Table 3) from the front left lower and upper left chest appeared insignificant in the PVC disturbance compared to the standard signal.For the PC1 variable for Tachycardia disorder, the channels lying on the left side and symmetrically placed channel 75 on the right side turned irrelevant compared to the signal without the disorder.
For the VFib Fine variable, these are channels 56, 60, and 65.The channel structure in the PC2 variable is more diverse than in the PC1 variable except for the VTach_Fine disorder, which, due to the most similar values of the percentage of explained variance of Variables PC1 and PC2, where the roles of the factors change places.
The main channels responsible for disturbance are those associated with a classical ECG measurement system with 12 channels labelled in Figure 8-11.Most of these channels are not significant in the PC1 variable but essential in PC2.This means that the PC1 variable will contain the information core of the similarities between the channels.The PC2 variable, on the other hand, will include information about specific disturbances in the BSPM signal.The main difference in the significance of the PC2 variable is that the channels on the left side and left back are essential for the Tachycardia disorder.Additional important channels are under the left armpit for AFib_Fine, PVC, and Brachycardia disorders.
For the variable VTach _Fine, probably by the most similar values of the percentage explanation of the channel system, the factor loadings in the variables PC1 and PC2 are exchanged with each other to the signal without disturbance.There is also a more differentiated division into channels with positive and negative factor loadings, which results in a more significant differentiation of the signs of loadings for both PC1 and PC2 than other disturbances.For the VTach Fine disorder for the PC1 variable, the important channels are those in the upper right and lower left front of the chest and the right side and right back.
The main factor containing information about disturbances in the conducted analysis is the PC2 factor, which determines the channels responsible for distinguishing disorders.The exception is the VTach_Fib disorder, where due to similar percentage values of explained variance of PC1 and PC2 variables, the PC2 variable is a variable with a similar structure to the PC1 variables from the remaining disorders.Therefore, for a VTach_Fine disorder, the PC1 variable will be responsible for channel differentiation for this disorder.
The differences between the Vtach Fine signal and the others are most likely due to the specifics of the signal itself.The VtachFine signal shows that the factors that enter PC1 and PC2 are swapped.Figure 7c shows a higher correlation between the channels (the vectors responsible for the channels are more concentrated), so these visitors will have higher factor charges and carry more information about PC1.Hence, there is also a higher percentage of explained variance in the PC1 variable compared to the other disturbances.

Conclusion
The paper presents research on determining information dependencies in the BSPM measurement channels.First, based on the simulation system constructed for testing the innovative BSPM measuring waistcoat, heart rhythm signals were performed with disorders: Atrial fibrillation, Bradycardia, Normal signal, PVC, Tachycardia, Ventricular Fibrillation, Ventricular Tachycardia.Then, the connections between measurement channels were indicated using the PCA method, and latent variables, grouping individual measurement channels (electrode signals), were specified.Two new latent variables, PC1 and PC2, were selected for individual disturbances and standard signal, following the Kaiser criterion and the percentage of explained variance.Based on the number of electrodes in the waistcoat coinciding with the classical ECG electrodes, it was indicated that the variable PC2 is the variable based on which a particular disturbance is identified.The exception is the VTachFine disturbance for which variable PC1 is shown.The PCA analysis conducted in this paper has two tasks.The first is data reduction.Determining latent variables by PCA results reduced the data dimension from 102 variables in measurement data from the measuring electrodes to 2 variables, PC1 and PC2.In addition to data reduction, PCA also has the advantage of getting rid of measurement noise and irrelevant variables without removing variables but only parts of them.The second is mapping areas on the patient's body for significant measurement electrodes, which can be determined from significant factor loadings in the latent variables PC1 and PC2.Such areas can be helpful during the detection of disease areas, e.g., electrodes during ECG measurement can be applied in other regions than classically used in ECG.A disadvantage of the PCA method is its sensitivity to measurement signal disturbances.Therefore, a distortion-free, filtered, or pre-filtered signal is needed.The authors will extend this research by showing how the indicated areas on the patient's body correlate with regions in the heart, which will result in a very detailed diagnosis.
The work presented here is used in a cloud expert system that analyses and classifies EIT and BSPM signals.It is currently patented (patent no.437101).
The research will also be the basis for validating clinical studies conducted on patients, where additional problems may arise due to signal interference caused by electrode friction, improper adhesion of electrodes to the patient's body, or patient movement during the test.

Figure 1 .
Figure 1.The schematic view of the electrode location on the measuring vest

Figure 3 .
Figure 3.The elastic silicone-textile electrode was designed and manufactured by Netrix S.A.

Figure 4 .
Figure 4. Diagram showing the measurement setup

Figure 5 .
Figure 5.The electrical potential on the heart was reconstructed from the simulation of BSPM measurement

Figure 6 .
Figure 6.The BSPM vest measures the signals for a single channel.The signals presented are as follows: a) AFib_Fine, b) Brachycardia, c) Normal disturbance, d) PVC disturbance, e) Tachycardia, and f) VFib_Fine disturbance.

Figure 7 .
Figure 7. Corrplot diagram of the correlation matrix.Channel numbering starts at 1 for the upper-left corner of the matrix

Table 1 .
Percentages of explained variance for individual PC variables for data from individual disorders

Table 2 .
Significant channels for the standard signal are divided into PC1 and PC2

Table 3
shows differences in the significance of channels to the signal without disturbance for thePC1 and PC2 variables MEASUREMENT CHANNELS' INFORMATION RELATIONSHIPS IN BODY SURFACE POTENTIAL MAPPING

Table 5 .
Significant channels for the signal with VTach_Fine disturbance divided into PC1 and PC2 variables