A Deep Learning Approach for Automated Prediction of Cardiac Arrest from Vital Sign Data of Intensive Care Unit Patients

Cardiac arrest, abbreviated as CA, is a leading cause of death worldwide, emphasizing the significance of early detection for prompt treatment and improved patient outcomes. Early detection is critical as it allows for timely intervention that saves lives. However, detecting CA can be challenging due to its non-specific symptoms, and time, which is the essence. By doing so, we can implement preventive measures before such incidents occur. In this study, we propose a prediction CA in a patient method based on employing a deep learning algorithm is convolutional neural networks (CNN) method to learn the underlying patterns in the data and to generate an accurate prediction of CA in vital sign data. To evaluate the performance of our prediction methods, we utilized publicly accessible database known as the Medical Information Mart for Intensive Care III (MIMIC III) v1.4. Our proposed model has been developed to analyze 51,577 ICU stays’ worth of data, focusing on eight essential vital sign features: body temperature, heart rate, respiration rate, systolic blood pressure, diastolic blood pressure, mean blood pressure, oxygen saturation, and glucose. Through this evaluation, we demonstrate the effectiveness of our prediction techniques in enhancing the accuracy of predictive models and compared it with another state-of-the-art prediction method is Bidirectional Long Short-Term Memory (Bi-LSTM). The test results demonstrate that the accuracy of the proposed deep learning model based on CNN is 98%, whereas Bi-LSTM is 82%. Our results demonstrate that the CNN method outperforms other methods in terms of accuracy for predicting CA.