Web application to predict lung diseases from auscultation signals

Respiratory diseases are one of the leading causes of death worldwide, such as COPD, pneumonia, and RTIs in recent years. Despite that a large number of scientific works have been mechanisms of prevention, diagnosis and treatment, many social sectors do not benefit from this research. For this reason, new research works based on computer tools are required, since in this way the scope can be greater. In this sense, a computer tool capable of detecting lung diseases from auscultation signals through the use of neural networks is proposed. To achieve this objective, a package of public auscultation signals was processed using adaptive filters and EEMD signal decomposition; then, the resulting signals were used to generate three types of different datasets (statistical vectors, spectrograms, and MFCC images) that are used for the training of three classifiers destined to predict between five classes: COPD, Pneumonia, RTI, BRON and Healthy. Once the classifiers have been trained, they are capable of generating predictions, which are grouped into an Ensemble Classifier to make a final prediction. The classifier models individually obtained a significant performance since the precision varies from 88% to 93%. However, the Ensemble Classifier achieved an accuracy of 93.4% and specificity of 96.2%, showing that this classifier is a more reliable model. Finally, the classifier algorithm developed was implemented into a web application to be used from anywhere connected to the internet.

• Egile nagusia: Suquilanda Pesantez, Jefferson Daniel (author)
• Formatua: bachelorThesis
• Hizkuntza: eng
• Argitaratua: 2022
• Gaiak: Auscultación; Sonidos pulmonares; Filtro adaptativo; Espectrograma; Redes neuronales; Ensamble de clasificadores; Auscultation; Lung sounds; Adaptive filter; Spectrogram; Neural networks; Ensemble classifier
• Sarrera elektronikoa: http://repositorio.yachaytech.edu.ec/handle/123456789/501