DKPNet41: Directed knight pattern network-based cough sound classification model for automatic disease diagnosis.

Area of Science:

• Computational intelligence
• Biomedical signal processing
• Machine learning for healthcare

Background:

• Cough sound analysis is a growing area for machine learning applications in disease detection.
• Existing research often focuses on specific diseases like Covid-19, limiting broader diagnostic utility.
• A comprehensive dataset and a robust model are needed for accurate multiclass cough-based disease identification.

Purpose of the Study:

• To develop and validate a novel machine learning model, DKPNet41, for the automatic multiclass classification of respiratory and cardiac conditions based on cough sounds.
• To create and utilize a large, diverse dataset of cough sounds from 642 subjects, encompassing Covid-19, heart failure, acute asthma, and healthy individuals.
• To establish an efficient and accurate method for cough-based disease diagnosis.

Main Methods:

• A novel feature generation technique using a directed knight pattern (DKP) was employed.
• Signal decomposition was performed using four pooling methods, followed by feature selection with iterative neighborhood analysis (INCA).
• The k-nearest neighbor (kNN) classifier with ten-fold cross-validation was utilized for classification.

Main Results:

• The DKPNet41 model achieved a high accuracy of 99.39% in multiclass classification.
• The model effectively processed 41 feature vectors, selecting the ten most informative ones for classification.
• Iterative neighborhood analysis (INCA) successfully identified the most discriminative features for accurate diagnosis.

Conclusions:

• The DKPNet41 model demonstrates high performance in automatically classifying cough sounds for disease diagnosis.
• The study highlights the potential of DKPNet41 for efficient and accurate multiclass disease detection using cough audio.
• This research contributes a valuable tool for non-invasive disease screening and diagnosis.