Date of Award

22-5-2024

Document Type

Thesis

School

School of Electrical & Electroncis Engineering

Programme

Ph.D.-Doctoral of Philosophy

First Advisor

Dr.N.R.Raajan

Keywords

2D CNN, Capacitance, Electrocardiogram, Electrode, Guarding, Gold Coated Copper, Optimal Lead Position, Physionet, Shielding, Transfer Learning

Abstract

In biosignal acquisition and physiological monitoring, developing advanced electrode technologies is crucial in enhancing signal quality, minimizing interference, and improving overall reliability. One such innovative approach is the utilization of capacitive-coupled electrodes, a cutting-edge solution that addresses some of the challenges associated with traditional electrodes in biosignal recording.

Capacitive-based sensors have become prominent in physiological signal measurement over the past decade. The electric field from the external affects both the human body and the electrode. The primary aim while designing the capacitive electrode is to maximize the signal-to-noise ratio. Hence, we introduce shielding and guarding techniques to improve the signal-to-noise ratio. In this work, we proposed a Capacitive Electrode (CE) with optimal Shielding and Guarding design (SGD). The copper was coated on the base material.

Then, gold was coated over the copper to prevent oxidation and improve conductivity. The base material used was FR4. Three different designs were proposed. The performance of each design was evaluated using the Finite Element Method (FEM). On comparing all the electrophysical parameters of the three designs, the design: ”The shielding was placed around the sensing plate, and the guarding was formed as a concentric around the sensing plate” was shown as the CE with optimal SGD, which enables higher performance.

The electrostatic potential of the optimal design is -1e-3 V, which is higher than the electrostatic potential of CE designs. The proposed electrode was applied for the ECG application. The optimal electrodes (OE) were placed in the lead II and V1 V2 lead positions. The comparison of the OE and other electrode at the lead II position (electrostatic potential is -1e-3 V) and chest lead v1 and v2 positions (electrostatic potential is -1e-4 V) is performed, and the results show that the lead II position is the OE lead position. The OE and electrode position to fetch the ECG signal were more accurate with less noise than other designs.

The proposed capacitive electrode with optimal Shielding and Guarding design can be used to fetch other bio-signals. When the proposed electrode was applied to fetch the ECG signal, we obtained the optimal signal-to-noise ratio of the ECG signal of order. The signal sensed in the capacitive sensors has to be classified for the abnormality. However, the raw data obtained from these sensors often requires advanced processing techniques for accurate analysis and classification of ECG patterns. Detecting cardiac disorders is difficult due to several contributing factors linked with patients and diagnostic materials. It required high computational complexity and a feasible method.

Many classification techniques have been proposed in the literature. This work focused on the novel methodology for classifying seven arrhythmic abnormalities of ECG signal. A novel Deep Convolution Neural Network method called RaNet has been developed. Then, transfer learning has been diligent in the RaNet for fast computation and classification. The training has been performed with a 12-lead ECG dataset from the physionet (unbalanced dataset).

In our method, the testing dataset need not be a twelve-lead dataset. We used ECG signals taken from single-lead, two-lead, trio-lead or six-lead. The efficiency of this method is analyzed by comparing its accuracy with that of other neural networks in the literature.

The network performance is evaluated by comparative analysis between different architectures in transfer learning. The proposed model RaNet with transfer learning reaches the accuracy of 98.44% with an F1 score of 98.98%.

The precision and recall values are 98.77% and 99.21%, respectively. The proposed method is also compared with other 2D image-based CNN methods, and it has been proved that the proposed method is faster and more efficient than other methods for classifying ECG abnormalities. The major advantage of the proposed method is that it reduces the computation time and has higher accuracy. The capacitive sensor monitoring system in this work is designed as one lead system.

Also, the detection and transmission of 12-lead ECG data for anomaly detection proves time-consuming. Hence, in this work, we developed a CNN algorithm to reconstruct the 12-lead ECG model into a 2-lead ECG model. Initially, a data set with 23 abnormalities is considered and the best 2-lead combination for each of the most commonly occurring anomalies from all the available combinations using the K-mean clustering algorithm. This is done by detecting any deviation from the ideal PQRST ECG.

Download

Share

COinS