Abstract:
Sepsis is a fatal condition that is triggered by a dysregulated immune response as the body reacts to
an infection, and timely detection is one of the largest challenges in medical care today. Delays in
diagnosis and therapy account significantly for mortality, so there is great value in continuous
monitoring as well as prediction platforms. The present work introduces Sepside: An Edge-AI IoT
System for Early Sepsis Warning and Time-to-Deterioration Prediction using low-cost biomedical
sensors, in-embedded processing, and machine learning approaches so as to present real-time
clinical information. The system integrates an ESP32-based microcontroller with ECG, PPG, and temperature sensors, with online streaming of vital data that is available locally for analysis. A pipelined architecture was implemented with the following phases: preprocessing, feature extraction, and model training, followed by comparisons of Random Forest, Gradient Boosting, and XGBoost classifiers, with
regression-based models, in order to predict time-to-deterioration (TTD). The results indicated that
the XGBoost consistently performed better with an average AUROC of 0.965 and average AUPRC
of 0.402, respectively, for classification, as well as an average MAE of 1.87 with R2 of 0.845, for
regression of TTD. These findings are evidence of the potential of deploying efficient edge-based
AI models for the purpose of early warning and prognosis. Besides technical breakthroughs, the product abides by international software, hardware, and communication standards, as well as ethical, environmental, and societal expectations. Weakened due to small-scale prototyping and synthetic dataset usage, the study sets the path forward for verification with real patient datasets as well as clinical applications. By utilizing the convergence of IoT, biomedical engineering, and artificial intelligence, Sepside is an environmentally friendly, economically viable, and globally significant solution to sepsis management.
