Seminar Details

Seminar Title
::
Design and development of a deep neural network for classification and prediction of cardiac arrhythmias
Seminar Type
::
Progress Seminar
Department
::
Biotechnology and Medical Engineering
Speaker Type
::
Student
Speaker Name
::
BUDARAJU DHANANJAY ( RollNo : 518BM1004)
Date  &  Time
::
11 Jun 2021  11.00 AM
Venue
::
Online Mode (Through MS-Team) Code:7oxla31
Contact
::
Prof. Sivaraman J.
Abstract
::

In recent times, it has become quite challenging to clinically and manually diagnose cardiac arrhythmias. Many cardiac arrhythmias display a lot of similar clinical characteristics, thus leading to misdiagnosis. The aim of the work is to design and develop a neural network for the classification and prediction of cardiac arrhythmias. A machine learning model has been developed to classify Sinus Rhythm (SR), Sinus Tachycardia (ST), and Atrial Tachycardia (AT) cardiac signals. The clinical ECG parameters were considered as the input features and the Prediction Value Change (PVC) algorithm imbibed in the CatBoost (CB) machine learning model was used to rank the features. The clinical ECG features consisted of amplitude and durational aspects of the ECG signal parameters. Apart from the CB model, other machine learning models considered were Extra Trees (ET) and Ridge Classifier (RC). All the three models considered were evaluated based on accuracy, sensitivity, precision, and F1 score. The CB model displayed 99 %, 99.17 %, 99.25 %, and 99 % of accuracy, sensitivity, precision and F1 score respectively. The other machine learning models considered ET and RC displayed a 99 % accuracy, sensitivity, precision, and F1 score. The computational time required to classify SR, ST, and AT cardiac signals for CB, ET, and RC are 0.0078s, 0.2401s, and 1.2493s respectively. The models ranked P-wave (µV), PRI (ms), and PPI (ms) as the most important features to be considered for classifying SR, ST, and AT cardiac signals. On comparing the performance of the developed CB, ET, and RC models, the CB model performed better. The computational time is minimal in the CB model compared with ET and RC due to the use of symmetric trees-based inference system. The over-fitting problems encountered in the CB model are minimal due to the boosting algorithm present in the CB classifier.