Detección y clasificación de valvulopatías mediante procesado de señal y técnicas machine learning a partir del análisis de sonidos cardíacos.
Fecha
2024-10-16
Autores
Título de la revista
ISSN de la revista
Título del volumen
Editor
Resumen
[es] La auscultación cardíaca es un método no invasivo ampliamente utilizado para detectar enfermedades cardiovasculares mediante la identificación de sonidos cardíacos anómalos. Sin embargo, requiere una gran experiencia médica, ya que los cardiólogos experimentados reconocen alrededor del 80% de las valvulopatías, mientras que los médicos de atención primaria solo detectan entre el 20-40%. Dado que los
algoritmos de aprendizaje profundo han superado a los métodos tradicionales en el procesamiento de
señales biomédicas, este proyecto analiza varias arquitecturas de deep learning para comparar su
rendimiento en la detección y clasificación de sonidos cardíacos a partir de representaciones tiempofrecuencia
bidimensionales (2D) o secuencias temporales unidimensionales (1D). El objetivo final es
implementar estas arquitecturas en dispositivos móviles, evaluando su viabilidad en función de los recursos de hardware, y desarrollar una aplicación para smartphones que alerte sobre la presencia de sonidos cardíacos anómalos asociados a valvulopatías.
Cardiac auscultation is a non-invasive method widely used to detect cardiovascular diseases by identifying abnormal heart sounds. However, it requires significant medical expertise, as experienced cardiologists can recognize around 80% of valvulopathies, while primary care physicians detect only 20-40%. Given that deep learning algorithms have outperformed traditional methods in biomedical signal processing, this project analyzes several deep learning architectures to compare their performance in detecting and classifying heart sounds from two-dimensional time-frequency representations (2D) or one-dimensional temporal sequences (1D). The ultimate goal is to implement these architectures on mobile devices, evaluating their feasibility based on hardware resources, and to develop a smartphone application that alerts users to the presence of abnormal heart sounds associated with valvulopathies.
Cardiac auscultation is a non-invasive method widely used to detect cardiovascular diseases by identifying abnormal heart sounds. However, it requires significant medical expertise, as experienced cardiologists can recognize around 80% of valvulopathies, while primary care physicians detect only 20-40%. Given that deep learning algorithms have outperformed traditional methods in biomedical signal processing, this project analyzes several deep learning architectures to compare their performance in detecting and classifying heart sounds from two-dimensional time-frequency representations (2D) or one-dimensional temporal sequences (1D). The ultimate goal is to implement these architectures on mobile devices, evaluating their feasibility based on hardware resources, and to develop a smartphone application that alerts users to the presence of abnormal heart sounds associated with valvulopathies.
Descripción
Palabras clave
Sistemas de Telecomunicación e Imagen y Sonido