Reasearch On Sequential Medical Health Data Mining

Intubation in the intensive care unit(ICU)is a common life-saving intervention that helps prevent patient suffocation.Intubation is a high-risk process,and the complications caused by it may have obvious side effects on the lungs,other organs and organic systems,and may be an important factor in the increase of critical illness morbidity and mortality in patients with mechanical ventilation.Because of the complex and diverse monitoring data in the ICU,predicting whether a patient needs to be intubated is a challenging task for the ICU clinical staff.In this paper,we propose an end-to-end deep learning framework that combines static and dynamic ICU data for intubation prediction.In this framework,Convolutional Neural Networks and Recurrent Neural Networks are used to model time series data and fuse the learned dynamic features with static data features.Our method was evaluated on a real clinical data set(MIMIC-?,PhysioNet)and compared with some traditional machine learning methods and the latest available methods.The experimental results show that our method AUC reaches 0.8397,and the performance exceeds that of other traditional methods and the most advanced methods.The main contributions of this article are as follows:(1)Firstly,this paper uses the traditional machine learning method to predict the intubation of ICU patients,and designs an effective data collection method to solve the problem of less data and data imbalance,and proposes an ensemble learning method to improve performance.(2)For time-series data in ICU,this paper considers the intubation prediction as a time-series classification problem.We discretize the time series,and for the lack of information that is universal in the sequence data,we introduce two representations to describe the missing patterns and use CNN and GRU to classify the time series,which greatly improves the prediction performance while avoiding artificial design features.(3)This paper designs a deep ensemble network of CNN and GRU for time-series classification,and finally proposes an end-to-end deep learning ensemble framework,which combines ICU time-series data with static data to predict intubation.It is significantly better than traditional machine learning methods and the current state of the art deep ensemble framework.