Can Machine Learning Algorithms Detect Acute Respiratory Disease Based on Cough Sounds?

Machine Learning

A machine-learning algorithm for detecting and classifying acute respiratory illness showed good predictive ability based on cough sounds in pediatric patients, according to a new study published in . International Journal of Medical Informatics.

Cough is a common symptom of acute respiratory illness and one of the most common symptoms in primary care worldwide. However, assessment of cough sounds is limited by subjective interpretation in clinical settings, which can lead to misdiagnosis and the need for emergency hospitalization.

“Despite the potential importance of objective cough sound assessment in clinical decision-making in acute respiratory disease, the synthesis of evidence on this topic is still incomplete,” the researchers wrote. “Therefore, we conducted a systematic review to determine the ability of machine learning methods to predict acute respiratory illness in a pediatric population using cough sounds.”

In this study, researchers examined the objective use of artificial intelligence (AI) as a potential aid in clinical respiratory disease diagnosis.

In this study, we reviewed 6 papers that were cited from the Scopus, Medline, and Embase databases on 25 January 2023. These papers included cough sound features and AI algorithms in diagnosing pediatric patients under the age of 18. In addition to cough sound signatures, studies based on non-cough sound signatures, such as demographic and clinical data, are also included. Furthermore, the quality assessment of these studies is based on the checklist for medical AI assessment (Chamai).

As a result, the analysis showed variability in inputting the algorithm, including different cough sound signatures and combinations of sounds and clinical features. Moreover, machine learning algorithms were different from traditional algorithms.

Furthermore, accuracy in detecting bronchiolitis, croup, pertussis, and pneumonia across five papers ranged from 82% to 96%, whereas a significant drop in accuracy was found in the detection of bronchiolitis and pneumonia in the sixth paper. I was. The researchers believe this loss of accuracy demonstrates how subjective clinical decision-making is for diagnosing these two respiratory diseases.

The researchers also found that the cough sound features used to detect croup were more accurate than the combined cough and clinical features. It is believed that this may be due to the croup patient’s barking cough from a distance.

However, due to the limited number of studies, researchers need to do more research to better understand how AI in medicine can be used to detect acute respiratory illness in children. I think.

Despite its limitations, the researchers believe the results of this review are a good starting point, noting that “the promising diagnostic accuracy in most of the reviewed studies supports its potential as a tool for evaluating respiratory disease.” shows,” the researchers wrote. “The knowledge gained from this systematic review can be used for future research planning and will also benefit regulators, technology manufacturers, engineers, data scientists and clinicians.”


Sharan RV, Rahimi-Ardabili H. Detection of acute respiratory illness in the pediatric population using cough sound features and machine learning: a systematic review. International Journal of Medical Informatics. Published online on April 18, 2023: 105093. doi:10.1016/j.ijmedinf.2023.105093

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *