کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Mortality Rate Prediction in Intensive Care Units using Gated Recurrent Unit Networks

نویسندگان :

Pooya Rahbari Shad¹ Mohammad Bagher Khodabakhshi² Mohammad Reza Rezaeian³ Reza Hashemi⁴

1- hamedan university of technology 2- hamedan university of technology 3- hamedan university of technology 4- AJA University of Medical Sciences

کلمات کلیدی :

Intensive Care Unit (ICU)،Mortality Prediction،Gated Recurrent Unit (GRU)،Deep Learning،MIMIC-III،Sequence Analysis

چکیده :

Early prediction of in-hospital mortality in Intensive Care Units (ICUs) is paramount for clinical decision-making and improving patient outcomes. Despite the abundance of data in Electronic Health Records (EHRs), predictive modeling remains challenging due to high-dimensional noise, rare clinical events, and the risk of data leakage. In this study, we propose a robust deep learning framework based on Gated Recurrent Unit (GRU) networks to predict mortality using the MIMIC-III database. Our primary contribution is an "Intelligent Vocabulary Refinement" (IVR) strategy, which filters infrequent medical codes and eliminates administrative artifacts that often lead to artificial performance inflation. The proposed architecture leverages an Embedding layer for feature representation followed by a GRU layer to capture the complex temporal dependencies within clinical event sequences. Experimental results demonstrate that our model achieves a state-of-the-art Accuracy of 96.42% and an Area under the ROC Curve (AUC-ROC) of 96.67%. Notably, the model maintained a high Recall of 90.76%, ensuring superior sensitivity in identifying high-risk patients. These findings suggest that the integration of sequence-aware architectures with meticulous data preprocessing can significantly enhance the reliability of clinical early-warning systems.