Our research advances critical care through artificial intelligence, edge computing, multimodal sensing, and clinical collaboration. Explore our peer-reviewed publications, and studies supporting innovation in healthcare.
Harpreet Singh, Ravneet Kaur, Satish Saluja, et al.
This study presents MORPHEUS, an edge-hosted large language model (LLM) pipeline for automated NICU discharge summary generation. Evaluated across 401 neonatal admissions using a clinician-defined rubric spanning 72 clinical fields, the system demonstrated higher accuracy, completeness, coherence, actionability, and factual consistency than clinician-authored summaries while preserving privacy through edge-based orchestration.
Ravneet Kaur, Monika Jain, Ryan M. McAdams, et al.
This work presents a knowledge-driven clinical decision support system that combines ontology modeling and semantic rule-based reasoning to deliver personalized nutrition recommendations for premature infants in the NICU. The Nutrition Recommendation Ontology (NRO) integrates neonatal nutrition guidelines with electronic medical record data to automatically recommend enteral and parenteral nutrition while identifying deviations from clinical guidelines. Validated using data from 601 NICU patients and reviewed by neonatologists, the system achieved 98% agreement with expert recommendations, demonstrating the potential of semantic AI for personalized neonatal care.
Ryan M. McAdams, Ravneet Kaur, Yao Sun, Harlieen Bindra, Su Jin Cho, Harpreet Singh
A comprehensive systematic review of AI and machine learning applications in neonatal intensive care, highlighting predictive models, clinical decision support, and the future of personalized neonatal medicine
This study introduces the High Definition Phenotype (HDP), a novel patient representation that combines fixed, intermittent, and continuous physiological data collected from multiple clinical sources. Using machine learning and deep learning models, the HDP framework significantly improved neonatal mortality prediction compared with widely used clinical scoring systems, enabling continuous bedside risk assessment throughout a patient’s NICU stay.
This study presents an AI-powered bedside clinical decision support system that predicts neonatal intensive care unit (NICU) length of stay using electronic medical records, nutrition adherence, medication deviations, antenatal and perinatal factors, and clinical diagnoses. Using data from 836 neonatal admissions for model development and 211 admissions for validation, the system demonstrated accurate patient-specific predictions while providing clinicians with real-time visualization of the factors influencing hospitalization duration.
Harpreet Singh, Ravneet Kaur, Abhilash Gangadharan, et al.
This paper presents Neo, an affordable bedside IoT device designed to integrate heterogeneous neonatal intensive care unit (NICU) medical devices into the iNICU clinical platform. Neo automatically acquires real-time physiological data from monitors, ventilators, infusion pumps, blood gas analyzers, and bedside cameras, enabling continuous monitoring, automated nursing documentation, remote visualization, and AI-driven clinical decision support. The system was validated using data collected from 92 preterm infants across two NICUs and demonstrated the feasibility of real-time physiological analytics for early detection of neonatal morbidity.
Abrar Majeedi, Ryan M. McAdams, Ravneet Kaur, Shubham Gupta, Harpreet Singh, Yin Li
This study presents a deep learning framework that automatically detects routine neonatal care activities—including diaper changing, tube feeding, and patting—from bedside video recordings and combines them with physiological signals such as heart rate and oxygen saturation. The system accurately quantifies the duration, frequency, and physiological impact of caregiving activities, enabling objective assessment of neonatal stress and laying the foundation for continuous AI-assisted stress monitoring in the NICU.
Harpreet Singh, Satoshi Kusuda, Ryan M. McAdams, et al.
This study developed one of the earliest AI-powered systems capable of automatically recognizing routine neonatal caregiving activities—including patting, diaper changing, and tube feeding—from bedside NICU videos while simultaneously synchronizing physiological signals such as heart rate and oxygen saturation. Using a combination of deep learning and synchronized multimodal monitoring, the framework demonstrated accurate automated documentation of neonatal care activities and revealed how routine handling affects neonatal physiology.
DOI: December 2020 (Published in Volume 8, Issue 1, 2021)
Harpreet Singh, Ravneet Kaur, Satish Saluja, et al.
This study presents an open-source neonatal intensive care unit (NICU) data dictionary consisting of 1,555 standardized clinical fields designed to harmonize patient documentation, device integration, clinical workflows, and quality indicators. Evaluated across two NICUs, the framework improved data quality, standardized clinical definitions, and enabled meaningful comparisons of practice variations and clinical outcomes while supporting interoperability and future AI-driven clinical decision support.
