CoronaHack-Respiratory-Sound-Dataset

Citation Author(s):
Asha Latha
Thandu
K L Deemed to be University, Vaddeswaram, Vijayawada
Pradeepini
Gera
K L Deemed to be University, Vaddeswaram, Vijayawada
Submitted by:
Asha Latha Thandu
Last updated:
Mon, 05/06/2024 - 06:05
DOI:
10.21227/z6eq-hw49
License:
119 Views
Categories:
Keywords:
0
0 ratings - Please login to submit your rating.

Abstract 

The COronaVIrus Disease of 2019 (COVID19) pandemic poses a significant global challenge, with millions

affected and millions of lives lost. This study introduces a privacy conscious approach for early detection of COVID19,

employing breathing sounds and chest X-ray images. Leveraging Blockchain and optimized neural networks, proposed

method ensures data security and accuracy. The chest X-ray images undergo preprocessing, segmentation and feature

extraction using advanced techniques. Simultaneously, breathing sounds are processed through tri-gaussian filters and mel

frequency cepstral coefficient features. The fusion of audio and image features are achieved with a progressive split

deformable field fusion module. The proposed Dual Sampling dilated Pre-activation residual Attention convolution Neural

Network (DSPANN) enhances classification accuracy while reducing computational complexity through augmented snake

optimization. Furthermore, a privacy-centric blockchain-based encrypted crypto hash federated algorithm is implemented for

secure global model training. This comprehensive approach not only addresses COVID-19 detection challenges but also

prioritizes data privacy in healthcare applications. The proposed framework exhibited recognition accuracy rates of 98%,

specificity of 97.02%, and sensitivity of 98%.

Instructions: 

In this research, Python serves as the programming

environment for image processing experiments. The

experiments run on a personal computer with a robust

configuration, including a 3.40 GHz, 16GB RAM and Intel

Core i7-6700 CPU, providing ample computational power.

To ensure a thorough evaluation, the prepared databases are

systematically divided into a training set (80% of the

database) and a separate test set (20%). This partitioning

strategy enables rigorous testing and validation of

methodologies.

The proposed framework utilizes chest X-ray and audio

sample data from the COUGHVID dataset for accurate early

diagnosis of COVID-19. The dataset includes 2,800 expertlabeled

coughs with diverse participant information. The

chest X-ray dataset comprises 13,808 images, split into 80%

training and 20% testing sets. A random 10% of the training

set is used for model validation. This comprehensive

approach ensures effective training, testing and validation of

the machine learning models for COVID-19 prediction.