An optimal asthma disease detection technique for voice signal using hybrid machine learning technique

Abstract

Current clinical practice in the treatment of chronic respiratory diseases does not have an effective method that allows patients and caregivers to constantly monitor the severity of respiratory symptoms. This is the “asthma breath” that occurs in the respiratory tract. In this article, we propose an optimal asthma disease detection technique for voice signal using hybrid machine learning (OADD‐HML) technique. In OADD‐HML technique, we used the improved weed optimization (IWO) algorithm for asthma detection and forecasting. We analyze whether the scattered signal pattern follows the frequency of normal and abnormal breathing sounds. Second, we illustrate an enhanced hunting search (EHS) algorithm for feature extraction and selection process. Then, deep Q neural network (DQNN) classifier used to identify the asthma, crackle, and normal speech. After performing the function recovery and classification using DQNN, the respiratory number classes will be identified. The test results show several classifications of specific OADD‐HML breathing sounds, respectively, using accuracy, sensitivity, and specificity. Comparison results of asthma and non‐asthma classification of respiratory sounds suggest that the particular method works better than usual existing state‐of‐art.

Respiratory Diseases Diagnosis Using Audio Analysis and Artificial Intelligence: A Systematic Review

“…In relation to disease identification, for COVID-19, several studies [ 13 , 71 , 72 , 73 , 74 , 77 , 78 , 87 ] have been conducted with accuracy rates ranging from 73% to 97% [ 73 ]. Similarly, studies have also focused on asthma [ 80 , 81 , 82 , 83 , 84 ], achieving accuracy scores ranging from 74% to 100% [ 80 ], and COPD [ 75 ], with an accuracy of 75% [ 76 ]. Notably, one study [ 76 ] suggests the potential to achieve a Matthews Correlation Coefficient of 1, indicating a correct classification rate of 100%.…”

“…Regarding the type of voice recording data, studies have utilized vowel sounds such as /a/, /e/, and /o/ [ 72 , 73 , 82 ], speech recordings including counting numbers and reading sentences [ 13 , 75 , 76 , 83 , 84 , 85 ], and in some cases, cough recordings [ 78 , 79 ]. Notably, several studies [ 71 , 74 , 77 , 80 , 81 , 86 , 87 ] have utilized multiple types of voice recording data.…”

“…In a specific study [ 79 ], a Data De-noising Auto Encoder (DDAE) was utilized to extract in-depth acoustic sound signal features. In several studies [ 13 , 71 , 72 , 73 , 77 , 78 , 79 , 80 , 84 , 85 ], researchers have employed various computational techniques to compute significant features for analyzing audio signals related to COVID-19 and other applications. These features include MFCCs, which capture the spectral characteristics of the sound, shimmer (measuring amplitude variability), jitter (measuring temporal variability), Spectral centroid (indicating the “brightness” of the audio signal), and others.…”

“…The authors of [ 79 ] studied five data augmentation methods, including time-stretching, pitch-shifting, compression-of-range, and the addition of background noise, to enhance the audio data. In another study [ 80 ], the authors exploited the Improved Weed Optimization (IWO) algorithm for Feature Extraction and the Enhanced Hunting Search algorithm for signal selection. Asthma prediction [ 84 ] involved producing the feature vector through the Ripplet-2 transform, which likely captured relevant information for the task.…”

“…Conversely, the studies [ 72 , 73 , 74 , 75 , 76 , 80 , 82 , 83 , 85 , 86 , 87 ] fall into the second group, employing models that are based on the statistical learning theory, like SVM and RF, for voice analysis. It is worth highlighting that SVM is the most commonly used ML model for respiratory disease classification due to its effectiveness on small and complex datasets.…”