Enactment of Progressive AI Architectures for Real-Time Heart Rate Monitoring and Anomaly Detection in Critical Care Environments

Neha joshi

Department of Electronics, Willingdon College, Sangli

Abstract - Continuous cardiac monitoring is crucial in high-stakes environments like Neonatal Intensive Care Units (NICUs). However, the current threshold-based systems often cause too many false alarms, which results in to enervation among clinical staff as well as produces disturbance for the patient admitted in special care units. There is a critical prerequisite for smart, automated solutions that can articulate apart signal artefacts from serious conditions like early-onset sepsis or cardiac distress. This research portrays a pioneering computational structure intended to exceed the restrictions of traditional cardiac monitoring through a high-fidelity, end-to-end analytical pipeline. By coalescing raw physiological signals from various clinical repositories, the system employs a rigorous pre-processing treatment using the Pan-Tompkins algorithm and digital filtration to separate accurate cardiac biomarkers from noise-laden environments. At its technical core, the study explores an innovative hybrid CNN-Transformer architecture that synergizes spatial convolutional feature extraction with multi-head self-attention mechanisms to decode complex, non-linear dependencies in heart rate variability. Experiential validation demonstrates that this cohesive methodology achieves a superlative 98.2% accuracy and an AUC of 0.99, significantly concealing conventional machine learning benchmarks like Random Forest. With a streamlined 12ms inference latency and a robust capacity to eliminate critical false negatives in arrhythmia detection, the framework offers a scalable, real-time solution for medical IoT applications. These results underscore the necessity of hybrid deep learning designs in bridging the gap between raw sensor data and actionable, life-saving clinical intelligence.Key Words:  Neonatal Intensive Care Units (NICUs), Pan-Tompkins algorithm, CNN-Transformer architecture, Heart rate variability, AUC (Area Under Curve), IoT, Deep learning