Scholarlink

In recent times, the integration of emerging technologies such as the Internet of Things and artificial intelligence have considerably transformed healthcare. As technology advances, the healthcare sector faces growing challenges due to ever-increasing demand and traditional invasive procedures. In this context, this article presents a model as an effective alternative to conventional oximetry especially useful in places where medical equipment is limited. The model utilizes deep learning to estimate blood oxygen levels through Long Short-Term Memory (LSTM) networks, which consist of two main layers. These layers include key indicators such as heart rate (HR), respiratory impedance (RESP), and pulse, witch were sampled 53 individuals for these indicators. The first is an LSTM layer with 100 cells and 41,600 parameters with a hyperbolic tangent activation function, designed to capture complex temporal patterns. The second linearly activated, dense layer with 101 parameters converts these representations into a clear and useful output. The results provide a mean absolute error of 1.68 and a root mean square error of 2.45, which shows that the model is quite accurate and suitable for applications that require continuous monitoring. In this work, a study was done to see how the main body functions are related. It found that heart rate and pulse are strongly connected, but not as much with respiratory rate and oxygen saturation. Also, used a Random Forest algorithm to determine the relationship between these factors and SpO2 values. The analysis revealed that heart rate and pulse are highly correlated, while the correlation with respiratory impedance and oxygen saturation is low. Finally, when implementing the model with TensorFlow Keras, we observed that it generalizes well and shows no signs of overfitting, which is a positive indication of its reliability. The results of this study not only demonstrate how emerging technology can improve the accuracy of health data, but they also open the door to future developments in predicting vital parameters in various clinical settings.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Prediction of Blood Oxygen Saturation by Physiological Variables Using Machine Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Ronald H. Rovira,
  • Óscar W. Gómez,
  • Manuel Montaño,
  • Marcia M. Bayas,
  • Junior Figueroa,
  • Carlos Efrain Andrade

摘要

In recent times, the integration of emerging technologies such as the Internet of Things and artificial intelligence have considerably transformed healthcare. As technology advances, the healthcare sector faces growing challenges due to ever-increasing demand and traditional invasive procedures. In this context, this article presents a model as an effective alternative to conventional oximetry especially useful in places where medical equipment is limited. The model utilizes deep learning to estimate blood oxygen levels through Long Short-Term Memory (LSTM) networks, which consist of two main layers. These layers include key indicators such as heart rate (HR), respiratory impedance (RESP), and pulse, witch were sampled 53 individuals for these indicators. The first is an LSTM layer with 100 cells and 41,600 parameters with a hyperbolic tangent activation function, designed to capture complex temporal patterns. The second linearly activated, dense layer with 101 parameters converts these representations into a clear and useful output. The results provide a mean absolute error of 1.68 and a root mean square error of 2.45, which shows that the model is quite accurate and suitable for applications that require continuous monitoring. In this work, a study was done to see how the main body functions are related. It found that heart rate and pulse are strongly connected, but not as much with respiratory rate and oxygen saturation. Also, used a Random Forest algorithm to determine the relationship between these factors and SpO2 values. The analysis revealed that heart rate and pulse are highly correlated, while the correlation with respiratory impedance and oxygen saturation is low. Finally, when implementing the model with TensorFlow Keras, we observed that it generalizes well and shows no signs of overfitting, which is a positive indication of its reliability. The results of this study not only demonstrate how emerging technology can improve the accuracy of health data, but they also open the door to future developments in predicting vital parameters in various clinical settings.