Project: VCardiac - A Novel Framework for Contactless Detection and Forecasting of Early Stages of Heart Diseases and Cardiac Arrest Conditions

Project information

Projekt name
VCardiac: A Novel Framework for Contactless diagnosis and forecasting of Cardiovascular Diseases 
Project manager
Hemant Ghayvat and Sharnil Pandya 
Other project members
Annett Wolf och Tolkyn Abdikarimova
Participating organizations
Linnaeus University
Financier
European Commission 
Timetable
9th August 2022 – 30 June 2024
Subject
Computer science (Department of Computer Science and Media Technology, Faculty of Technology)
Research group
Linnaeus University Centre for Data Intensive Sciences and Applications (DISA)
Linnaeus Knowledge Environment:
Digital Transformations
Web site
https://cordis.europa.eu/project/id/101065536 

Short description

For performance evaluations of the proposed VCardiac Framework, the collected heartbeat acoustic samples will be classified using LSTM-CNN, RNN, LSTM, Bi-LSTM, CNN, K-means Clustering, and SVM methodologies. Furthermore, the proposed VCardiac Framework will also assist in classifying Age and Gender-wise risks using methodologies such as Kaplan-Meier and Cox-regression survival analysis. These methodologies will also assist in identifying the probability risk and the 10-year risk score prediction. In the end, the proposed VCardiac Framework will be tested for various Signal-to-noise ratio conditions to achieve better accuracy, effectiveness, and throughput. 

More about the project

In ancient times, doctors followed practices to diagnose cardiac arrest-related diseases by listening to cardiac patients' heartbeat sounds directly using medical equipment such as a "stethoscope." The followed procedures of cardiac arrest diagnoses were not very accurate and organized. Furthermore, even on ethical grounds, such practices were not scientifically proven. Especially in elderlies, the mortality rate due to cardiac arrest diseases has been very high. The older populations are maturing at a rapid pace. The U.N. reports have predicted that the count of the more ageing population will be multiplied by two by the year 2050.

In recent times, the increase in the number of cardiac arrest patients and the emergency departments have affected cardiac arrest diagnosis practices and have caused delays in the treatment procedures. Such delays have been the primary reason for the increase in the elderly mortality rates and adverse outcomes of cardiac arrest treatments[]3. However, the latest advancements in the technologies such as the AIoMT, AI-based remote healthcare monitoring, and big data and cloud-based notifications have created a paradigm shift in the domain of healthcare and provided 24 x 7 connectivity between smart devices, healthcare cloud, and medical experts. In general, the heartbeat signal of a human contains two sound components: lub and dub. These components are associated with closing valves of the heart. The medical practitioners diagnose the cardiac arrest patient based on these sound components using a medical diagnosis device such as a "stethoscope." The heartbeat sound components define an obvious pattern such as lub to dub and dub to lub in regular heart patients with the heart rate varies between 60 to 100. Therefore, these kinds of heartbeat sound components can be classified as normal. Another type of heartbeat sound is a murmur with a similar characteristic; however, it has a specific noise pattern such as rumbling, whooshing, and roaring. Similarly, adults and children have a unique heartbeat sound classification pattern, such as lub-lub to dub and lub-dub to lub. Predicting healthcare patients' cardiac arrest conditions is feasible because a sudden deterioration of vital signs can be noticed in health patients' for almost 24 hours. 

In this study, after conducting a riourous and detailed survey, we concluded that fellow researchers have carried out scattered research in the area of conditional cardiac monitoring and forecasting using audio signals, heterogeneous sensors, and machine learning and deep learning-based predictions. However, a complete Contactless Cardiac Classification and Risk Prediction Framework have not been proposed to the best of our knowledge. Therefore, in the undertaken study, a contactless VCardiac Framework has been proposed to classify various human voice-based acoustic heartbeat events such as normal, murmur, extrasystole, artifact, and other unlabeled heartbeat acoustic events using various machine learning and deep learning methodologies such as LSTN-CNN, RNN, LSTM, Bi-LSTM, CNN, K-means Clustering, and SVM. In addition, the proposed VCardiac Framework will also assist in classifying Age and Gender-wise risk predictions. Furthermore, the proposed VCardiac Framework is an economical alternative due to the low-cost Eko ECG and digital stethoscope, USB probe, 4-channel amplifier, and other equipment used in this experiential study. Moreover, Dr Hemant and I will also develop a testbed facility to test the proposed VCardiac Framework under various Signal to Noise ratio conditions such as SignaltoNoiseRatio0, SignaltoNoiseRatio3, SignaltoNoiseRatio6, SignaltoNoiseRatio9, SignaltoNoiseRatio12, SignaltoNoiseRatio15, and SignaltoNoiseRatio18 to achieve better accuracy, effectiveness and throughput. In the end, a complete prototype of the VCardiac Framework will be designed, tested, and deployed at the eHealth department of Linnaeus University.   

Cardiac arrest is when your heart suddenly stops pumping blood around your body. The symptoms arise suddenly and there is little time for tests. Sudden cardiac arrest is life-threatening. However, advances in technologies like contactless remote patient monitoring can help predict if (and when) someone will suffer a cardiac arrest. The Marie Skłodowska-Curie Actions project ContactlessFramework will develop a risk prediction framework, VCardiac, to diagnose and identify early stages of heart diseases and forecast cardiac conditions in advance, using the human voice. A customised dataset of 2 000 human voice samples will be designed to classify acoustic heartbeat events as normal, murmur, extra systole and artefact as well as other, unlabelled heartbeat acoustic events. 

In the undertaken study, a contactless VCardiac Framework has been proposed to classify various human voice-based acoustic heartbeat events such as normal, murmur, extrasystole, artifact, and other unlabeled heartbeat acoustic events using various machine learning and deep learning methodologies such as LSTN-CNN, RNN, LSTM, Bi-LSTM, CNN, K-means Clustering, and SVM. In addition, the proposed VCardiac Framework will also assist in classifying Age and Gender-wise risk predictions. Furthermore, the proposed VCardiac Framework is an economical alternative due to the low-cost Eko ECG and digital stethoscope, USB probe, 4-channel amplifier, and other equipment used in this experiential study. Moreover, Dr Hemant and I will also develop a testbed facility to test the proposed VCardiac Framework under various Signal to Noise ratio conditions such as SignaltoNoiseRatio0, SignaltoNoiseRatio3, SignaltoNoiseRatio6, SignaltoNoiseRatio9, SignaltoNoiseRatio12, SignaltoNoiseRatio15, and SignaltoNoiseRatio18 to achieve better accuracy, effectiveness and throughput. In the end, a complete prototype of the VCardiac Framework will be designed, tested, and deployed at the eHealth department of Linnaeus University. 

The project is part of the research at the Linnaeus University Centre for Data Intensive Sciences and Applications (DISA) and is included in the Knowledge Environment Linnaeus: Digital Transformations.