Early Prediction of Sepsis using Heart Rate Signals in Wearable Devices with Genetic Optimized LSTM Algorithm
Alireza Rafiei, Farshid Hajati, Alireza Rezaee, Amirhossien Panahi, Shahadat Uddin · 2025
Read Early Prediction of Sepsis using Heart Rate Signals in Wearable Devices with Genetic Optimized LSTM Algorithm by Alireza Rafiei, Farshid Hajati, Alireza Rezaee, Amirhossien Panahi, Shahadat Uddin (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-24253.txt.
Introduction
Alireza Rafiei and four colleagues aim wearables at early sepsis warning. Sepsis is a runaway immune response to infection. It brings high mortality and heavy healthcare costs. Early intervention depends on early prediction. Most existing models target intensive care patients. Detection outside hospital wards remains a gap.
The team designs four machine learning models for wearable devices. Heart rate signals are the only input. A genetic algorithm refines each model's architecture. The search balances accuracy against computing cost and memory. Those constraints decide whether a wearable can run the model at all. Initial models predict sepsis one hour ahead. Transfer learning extends the window to four hours. Results suggest wearables could flag sepsis beyond wards and intensive care. That prospect motivates the whole design.
Catalog record
| CREATOR | Alireza Rafiei, Farshid Hajati, Alireza Rezaee, Amirhossien Panahi, Shahadat Uddin |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-24253
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Early Prediction of Sepsis using Heart Rate Signals in Wearable Devices with Genetic Optimized LSTM Algorithm Alireza Rafiei, Farshid Hajati, Alireza Rezaee, Amirhossien Panahi, Shahadat Uddin Abstract Sepsis, a critical condition resulting from a dysregulated immune response to infection, significantly impacts mortality, morbidity, and healthcare costs. Effective and timely prediction of sepsis is essential for early intervention, particularly outside the intensive care unit (ICU) where monitoring may not be as rigorous. This study advances the application of machine learning in sepsis prediction by optimizing four established algorithms—LGB, MLP, LSTM, and a novel LSTM-FCN hybrid—for use with wearable devices that monitor heart rate. These models have been specifically enhanced to address the challenges of real-time data processing in non-ICU settings, focusing on improving computational efficiency and reducing memory demands. The architecture of each model was meticulously refined using a genetic algorithm to optimize performance metrics crucial for deployment on wearable technology. Initially developed for a one-hour prediction window, these models were later expanded to four hours using transfer learning techniques. The promising results demonstrate significant potential for wearable devices to support early sepsis detection in diverse clinical environments, highlighting our novel approach to enhancing the practical deployment of machine learning models in healthcare settings. K eywords Sepsis; Machine learning; Wearable device; Early prediction; and Genetic optimizer 1 INTRODUCTION Wearable technology consists of smart electronic devices worn near the surface of animate skin [ 16 ] . They can detect, process, analyze, and transmit information to other devices and data centers. In the last few decades, wearable devices, such as smartwatches and smart bands, have been ubiquitous in modern societies because of their practical applications, especially in health-oriented fields. Such equipment can measure some of the human’s vital signs and ambient features by dint of their high-tech sensors, connecting to other devices and the internet, and analyze information for instant biofeedback to the wearer on an ad hoc basis. As such, healthcare professionals use wearable devices with human activity and sleep pattern recognition ability to acquire behavioral and physiological data to diagnose, manage, and treat acute and chronic diseases [ 12 , 42 ] . Studies have shown that remote healthcare monitoring and early warning can potentially improve the quality of health services and reduce patient inconvenience [ 6 , 14 ] . Practical implementation of Internet of Things (IoT)-based telemedicine is burgeoning to prevent disease, predict difficulties at early stages, and generally promote global health [ 5 , 50 ] . Leveraging the power of Artificial Intelligence (AI) in telehealth not only aids physicians in making data-driven and real-time decisions as a critical factor in providing better health outcomes, but it also can produce primarily warn to individuals. With the accelerated increase of wearable gadget technologies and unprecedented algorithms, individuals rein their well-being, which assists them and physicians in diagnostics and analysis of ailing in primary steps. Sepsis is introduced as a potentially life-threatening and highly complex condition that leads to multi-organ dysfunction and exhibits numerous morbidity and mortality rates [ 73 ] . Sepsis occurs when the response of the host body’s immune and inflammatory system injures tissues and internal organs instead of fighting infections [ 70 ] . Due to early associated symptoms with sepsis, which may be familiar with numerous other clinical conditions, accurate diagnosis of sepsis in the early stages is an uphill struggle for even skilled physicians [ 29 ] . According to the World Health Organization’s (WHO) first global report on sepsis, 49 million individuals were estimated to be affected by sepsis, which causes approximately 11 million potentially avoidable worldwide deaths each year [ 47 ] . Centers for Disease Control and Prevention (CDC) reported that sepsis develops outside the hospital for almost 80 percent of patients [ 19 ] . Overall, the epidemiology of sepsis in out-of-hospital patients is complex and varies widely across different populations and healthcare settings. A study conducted in Korea found an incidence rate of 28.7 cases per 1000 people each year [ 33 ] . According to research carried out in the United States, sepsis accounts for a significant portion of emergency department visits and hospitalizations among adult patients [ 64 ] . The study also revealed that patients who received prehospital medical care had lower odds of dying in the hospital. Sepsis survivors encounter severe long-term problems in the form of growth post-discharge mortality, cognitive and physical impairment, and mental disorders [ 27 ] . This tremendous mortality rate is often related to late diagnosis, inadequate infection prevention, and improper clinical management, which can influence vulnerable populations, such as pregnant and older adults [ 47 ] . Sepsis care is expected to be most consequential in the earliest phase of treatment, so the aim is to diagnose sepsis on admission and receive initial care in the emergency department. A one-hour delay in treating sepsis has a consequence of a 4% to 8% increment in mortality probability. Hence, early sepsis-onset prediction and required antibiotic intervention are the main steps for ameliorating sepsis outcomes [ 35 , 68 ] . To meet this challenge, physicians introduce various diagnosis criteria and scoring systems [ 28 , 45 , 75 ] . The majority of criteria for diagnosing and treating sepsis have been developed for patients in hospital and intensive care unit (ICU) settings. However, it is important to note that up to 50% of sepsis patients initially come into contact with healthcare providers in the pre-hospital setting [ 68 ] . The newest sepsis definition, Sepsis-3, was introduced in 2016, and proposed the quick Sequential Organ Failure Assessment (qSOFA) method for rapidly identifying adult patients suspected of infection in out-of-hospital and emergency departments, as well as the SOFA scoring system, which is used chiefly for tracking patients’ status in the ICUs [ 73 ] . The qSOFA utilizes three district criteria, assigning one point for systolic blood pressure (SBP \leq 100 mmHg), high respiratory rate ( \geq 22 breaths per min), or altered mentation (Glasgow coma scale ¡ 15). Although [ 73 ] promulgated that qSOFA is less robust than SOFA score due to no requirement of laboratory tests, it can be evaluated rapidly and repeatedly. The abundance of wearable devices capable of accurately measuring heart rates could be synergetic for transmission of early sepsis onset prediction obligation out of wards. In this paper, we employed four distinct machine learning (ML) approaches for real-time sepsis onset prediction using just low-frequent heart rate signals to install as an auxiliary algorithm on wearable devices. For this reason, an LSTM optimized network with a genetic algorithm (GA), which is trained using the ICU’s Electronic Health Records (EHRs), has shown superior performance. The genetics algorithm is used for the optimization of architectural parameters based on three diverse factors: model performance, the complexity of calculations, and primary storage. More aspects of our research are presented in the succeeding sections: Section 2 investigates the state-of-the-art intelligence system for sepsis prediction; Section 3 deliberates the proposed approach to figure out an appropriate model to predict sepsis onset out of ICU and ward using available features; Section 4 presents the results which are discussed in details in Section 5. 2 RELATED WORKS ML and traditional methods demonstrate significant performance