International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 09 Issue: 04 | Apr 2022
p-ISSN: 2395-0072
www.irjet.net
ICU MORTALITY PREDICTION Sitara Khadka1, Cavin Patel2, Harshada Musali3, Prof. Swati H. Shinde4 123Students,
Department of Electronics and Telecommunication Engineering, K. J. Somaiya Institute of Engineering and Information Technology, Mumbai, Maharashtra, India. 4Assistant Professor, Department of Electronics and Telecommunication Engineering, K. J. Somaiya Institute of Engineering and Information Technology, Mumbai, Maharashtra, India. ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Patients admitted to an Intensive Care Unit (ICU) have
evidence-based care. Keeping patients data stored safely, easily, in less space and for an uncertain amount of time and also being in digital format, it decreases the number of records off track.[4] However, internal data constraints such as sparsity, irregularity, heterogeneity, and noise make modeling and evaluating EHR data more difficult. For simulation, an available to the public clinical dataset called Medical Information Mart for Intensive Care MIMIC-II was used, which covered broad, diverse, and granulated data that was accepted and is used for improved clinical prediction or as a data source for constant validation among most published machine learning literature [2-3]. For the ICU different types of severity, scoring has been developed which are the acute physiology and chronic health evaluation system (APACHE II), the simplified acute physiology score (SAPS II), the Sequential Organ Failure Assessment (SOFA) score. The challenge faced in severity scoring the accuracy rate of mortality prediction is less accurate and so to attain precise mortality prediction we are using intelligent tools ie; deep learning. This paper introduces deep learning (DL) based models for the Mortality Prediction (MP) in ICU patients. The models rely on classification techniques based on Recurrent neural network (RNN) and Long Short-Term Memory (LSTM). This paper aims to build a deep learning model that can predict mortality of a patient within the ICU based on EHR database and the model will utilize the unstructured data to derive relevant clinical events, vital signs data and lab events to predict mortality. The remaining paper is categorized as follows: We include the literature evaluation on relevant investigations in section 2. The MIMIC-II dataset’s basic data are provided in Section 3. We discussed the preprocessing method we used to extract the features and the RNN models we presented in Section 4.
life-threatening health problems or are in poor health and require considerable care and surveillance in order to recover quickly. An early ICU mortality prediction is critical for identifying patients who are at a higher danger and for making better therapeutic decisions for the patient. Using Deep Learning-based approaches, an early Mortality Prediction can be used to support this analysis. The Medical Information Mart for Intensive Care (MIMIC-II), which is freely available, is being used for the evaluation. The F1 score, area under the receiver operating characteristic curve (AUC), and precision predictions show the model’s ability. Two models, the recurrent neural network-long short term memory (RNN-LSTM) and the convolution neural network (CNN) are employed and evaluated in order to produce the best mortality prediction utilizing the SAPS-I score and related parameters.
Key Words: Morality prediction, Deep learning, SAP-I, MIMICII, Recurrent neural network, Convolution neural network.
I. INTRODUCTION Intensive care units (ICUs) are the department in-hospital which is used to treat people who are terminally ill, had major surgery, accidents, trauma or serious infection. The patient who is admitted in ICU is closely monitored and provided medical aid from skilled physicians, nurses, respiratory therapist, physical therapist, community worker and also provides medications in order to ensure normal. Therefore, quick and dependable prediction implementations in support of delicate medical conditions would be useful for aiding. The greatest task in critical care research is Mortality Prediction (MP). The use of mortality prediction can not only identify the high threat and can also make sure that the available ICU bed is given to the patient who is more in need of it. During the ICU stay, different biological parameters are checked and analyzed each day. In scoring systems these parameters are measured and examined to gauge the ferocity of the patients [6]. Since the technology is advancing rapidly and can be comprehended with healthcare. In ICUs, accurate mortality prediction is critical for assessing the severity of illness and determining the value of novel treatments and interventions that may help to improve clinical outcomes [2]. To track a patient’s clinical progress, Electronic Health Record (EHR) is used to facilitate enhanced health care decisions and contribute
© 2022, IRJET
|
Impact Factor value: 7.529
II.RELATED WORK Intensive care units are departments that deal with patients who have decrepit health and are dealing with multiple diseases simultaneously. The medical data collected during their ICU stay is feasible for predicting the mortality which can help doctors to make optimal decisions for further
|
ISO 9001:2008 Certified Journal
|
Page 3529