In ancient times, doctors followed unorganized practices to diagnose cardiac arrest conditions in healthcare patients. The medical procedures were not organized and accurate, even though cardiac patients were analysed using a stethoscope. However, the latest technological advancements, such as contactless remote patient monitoring, AIoMT (Artificial Intelligence of Medical Things), advanced big data, and cloud-based analytics and alerts, have created a paradigm shift in healthcare and provided 24x7 connectivity. In this study, we have also proposed to design a customized dataset, Voice-2000, of the human voice and heart sound signals. The customized dataset contains more than 2000 heart sounds and human voice samples from cardiac and non-cardiac patients. The average duration of the recorded heart sounds and human voice samples is between 9 and 13 seconds. In this study, we have analyzed and evaluated various human voice and heart-sound-based acoustic events using Librosa machine-learning libraries. Furthermore, the study has proposed a contactless VCardiac Framework to classify and detect cardiac anomalies from the recorded heartbeat sound signal. Eventually, the customized Contactless VCardiac Framework’s performance is evaluated with other machine and deep learning algorithms such as LSTM, RNN, Bi-LSTM, SVM, KNN, etc.
The objectives of the proposed study are:
• Creation of a Customized Cardiac Dataset termed as "Cardiac-2000". During this phase, a customized dataset will be created with 2000 human-voice based heartbeat acoustic event samples to balance the normal and abnormal heartbeat acoustic events. (Partially Completed: The customized creation of the dataset on-going, already we have collected 500 samples and performed various experiments on them. In future, we will complete this task and publish the customized "Cardiac-2000" dataset for fellow researchers.)
• Design and development of a VCardiac Framework for the early stages of heart diseases and forecasting cardiac arrest conditions in advance. (Fully Completed: Design and development of a VCardiac Framework for the early stages of heart diseases and forecasting cardiac arrest conditions is fully completed. The puedocode of the VCardiac Framework is available at
https://github.com/sharnilpandya84/VCardiac(s’ouvre dans une nouvelle fenêtre)).
• Design and develop a VCardiac Risk Prediction Framework to classify and predict cardiac risk conditions of various age groups. (Fully Completed: (Fully Completed: Design and development of a VCardiac Framework forclassifying age and gender-wise risks is fully completed. The puedocode of the VCardiac Framework is available at
https://github.com/sharnilpandya84/VCardiac(s’ouvre dans une nouvelle fenêtre))
• Design and develop a texture-based methodology to convert human voice-based normal and abnormal heartbeat acoustic events into noise-robust events. (Fully Completed: The design and development of the texture based methodology to denoise normal and abnormal heartbeat sound event is completed and its available at
https://github.com/sharnilpandya84/VCardiac(s’ouvre dans une nouvelle fenêtre))
• Testing of VCardiac Framework under a variety of SignalToNoise Ratio conditions. (Fully Completed: a separate Test-bed will be implemented at the health department of Linnaeus University to test the proposed VCardiac Framework under a variety of noisy conditions such as SignaltoNoiseRatio0, SignaltoNoiseRatio3, SignaltoNoiseRatio6, SignaltoNoiseRatio9, SignaltoNoiseRatio12, SignaltoNoiseRatio15, and SignaltoNoiseRatio18 to achieve better accuracy, effectiveness and throughput. The detailed description of the same is available in the explanation of the work section)
