Aplicación móvil para la estimación de riesgo de accidente cerebrovasculares basado en aprendizaje automático.
Loading...
Date
2024
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
La presente memoria de título se enfoca en el desarrollo de una aplicación móvil para la estimación del riesgo de accidente cerebrovascular (ACV) utilizando modelos de Machine Learning (ML). La incidencia de ACV es una problemática significativa a nivel mundial, dada su alta tasa de mortalidad y las secuelas que provoca en los sobrevivientes. Por ello, una herramienta de evaluación temprana y accesible es fundamental para facilitar la prevención y el monitoreo de esta afección.
La aplicación, desarrollada en Android Studio y basada en el modelo de Red Neuronal Artificial (ANN), permite al usuario ingresar datos personales y recibir una estimación probabilística sobre su riesgo de sufrir un ACV. La elección del modelo ANN, junto con la integración de métodos como el undersampling para el balance de clases, demostró ser la opción más efectiva para alcanzar resultados confiables y específicos en el contexto del conjunto de datos disponible. A pesar del desbalance de clases y otras limitaciones de los datos, el modelo mostró una capacidad adecuada para detectar posibles casos de riesgo, priorizando el recall sobre la especificidad, aspecto clave en aplicaciones de salud donde es preferible minimizar los falsos negativos.
La aplicación fue evaluada mediante múltiples métricas (accuracy, precisión, recall, especificidad, F1-score y AUC-ROC), y los resultados demostraron un buen desempeño, sin sobreajuste, según lo indicado por los gráficos de pérdida de entrenamiento y validación. Esto confirma la consistencia del modelo y su adecuación para estimar el riesgo de ACV en distintos perfiles de usuario.
Entre las conclusiones de este proyecto, se destaca que el modelo de red neuronal artificial implementado en la aplicación cumple satisfactoriamente su propósito de estimar el riesgo de accidente cerebrovascular, proporcionando resultados confiables y específicos basados en los datos ingresados por el usuario. La herramienta es particularmente útil para la predicción temprana de riesgos de ACV, priorizando el recall en la identificación de casos de riesgo, lo cual es crucial en el ámbito de la salud.
The present project focuses on the development of a mobile application for estimating the risk of stroke using Machine Learning models. The incidence of strokes is a significant global issue due to its high mortality rate and the lasting impact it has on survivors. Therefore, an accessible and early assessment tool is essential for facilitating the prevention and monitoring of this condition. The application, developed in Android Studio and based on an Artificial Neural Network (ANN) model, allows users to input personal data and receive a probabilistic estimation of their stroke risk. The choice of the ANN model, along with the integration of methods such as undersampling for class balancing, proved to be the most effective option for achieving reliable and specific results given the available dataset. Despite the class imbalance and other data limitations, the model showed adequate capacity to detect potential risk cases, prioritizing recall over specificity, a crucial aspect in health applications where minimizing false negatives is preferable. The application was evaluated using multiple metrics (accuracy, precision, recall, specificity, F1-score, and AUC-ROC), and the results demonstrated strong performance without overfitting, as indicated by the training and validation loss graphs. This confirms the robustness of the model and its suitability for estimating stroke risk across different user profiles. Among the conclusions of this project, it is highlighted that the Artificial Neural Network model implemented in the application successfully fulfills its purpose of estimating the risk of stroke, providing reliable and specific results based on the data entered by the user. The tool is particularly useful for the early prediction of stroke risk, prioritizing recall in identifying high-risk cases, which is crucial in the healthcare field.
The present project focuses on the development of a mobile application for estimating the risk of stroke using Machine Learning models. The incidence of strokes is a significant global issue due to its high mortality rate and the lasting impact it has on survivors. Therefore, an accessible and early assessment tool is essential for facilitating the prevention and monitoring of this condition. The application, developed in Android Studio and based on an Artificial Neural Network (ANN) model, allows users to input personal data and receive a probabilistic estimation of their stroke risk. The choice of the ANN model, along with the integration of methods such as undersampling for class balancing, proved to be the most effective option for achieving reliable and specific results given the available dataset. Despite the class imbalance and other data limitations, the model showed adequate capacity to detect potential risk cases, prioritizing recall over specificity, a crucial aspect in health applications where minimizing false negatives is preferable. The application was evaluated using multiple metrics (accuracy, precision, recall, specificity, F1-score, and AUC-ROC), and the results demonstrated strong performance without overfitting, as indicated by the training and validation loss graphs. This confirms the robustness of the model and its suitability for estimating stroke risk across different user profiles. Among the conclusions of this project, it is highlighted that the Artificial Neural Network model implemented in the application successfully fulfills its purpose of estimating the risk of stroke, providing reliable and specific results based on the data entered by the user. The tool is particularly useful for the early prediction of stroke risk, prioritizing recall in identifying high-risk cases, which is crucial in the healthcare field.
Description
Tesis presentada para optar al título de Ingeniero Civil Biomédico.
Keywords
Aplicaciones móviles, Accidente cerebrovascular, Machine learning