The growing complexity and volume of data in the Chinese critical care setting necessitates the need to predict mortalities with intelligent and scalable and explainable systems. Conventional approaches, like rule-based models and independent machine learning models, are largely ineffective at combining the multimodal characteristics of ICU data in Chinese hospitals, especially when only structured clinical variables or time-series vital data are considered. To resolve them, Medi CloudX presents a hybrid Deep Learning (DL) model based on TabNet when working with structured electronic health records (EHRs) and Informer when dealing with long-term time-series data on ICUs. This is a combination of the two which enables a higher accuracy of prediction by selecting interpretable features among structured data and extracting long-term dependencies in ICU signals. The Reptile Search Algorithm (RSA) search hyperparameter optimization improves the performance of models with minimum human intervention. MediCloudX on a dataset of Chinese ICU scored an accuracy of 98.0 %, sensitivity of 100, specificity of 96.0, and F1-score of 98.04, surpassing state-of-the-art models such as CatBoost (AUC = 0.889), and LSTM-augmented scoring systems (AUC ≈ 0.898). The cloud-native structure of MediCloudX guarantees scale elasticity, minimal inference latency, and safe data handling, which are suitable to real-time applications in the ICU in China. This smart and high-achieving system is explainable and efficient in resource utilization, and it has great prospects of implementation in intelligent hospitals.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号