Acoustic Noise Analysis and Management in Intensive Care Units

Abstract

Intensive Care Units (ICUs) are environments where patients and staff are exposed to excessive acoustic noise that significantly exceeds recommended guidelines, posing a substantial risk to physical and mental well-being. This thesis tackles ICU noise with a comprehensive approach. The aims are to perform extensive acoustical measurements in typical ICUs, intelligently characterise the acoustic environments where measurements are conducted, and finally, develop effective solution for reducing undesired noise. To accomplish these aims, a cost-effective measurement system is first designed and deployed in three ICUs for data collection. An Acoustic Event Classification (AEC) system, supported by the YAMNet model, is then proposed to characterise noise sources. Finally, Active Noise Control (ANC) technology is introduced as a strategy to mitigate noise around patients’ headrests using innovative adaptive algorithms.
Deployed in three ICUs with continuous multi-day operation, the developed system delivered accurate measurements. In the monitored ICUs, the analysis of noise levels mostly ranged from 49–60 dB, 53–65 dB, and 47–56 dB in the 1st, 2nd, and 3rd ICUs, respectively, with extremes from 41 dB to nearly 82 dB–well above recommended limits.
The YAMNet-based AEC system offered insights into ICU noise sources by classifying various sound events, achieving average precision scores of 96.76%, 96.16%, and 96.84% across the three investigated ICUs, respectively. The intelligent analysis led to the establishment of three efficient acoustic datasets. Despite removing speech for privacy reasons, it was identified as a major contributor in all ICUs, constituting 58.28%, 52.90%, and 68.46% of the cumulative noise duration.
Finally, the simulation results demonstrated that the proposed ANC algorithms can efficiently attenuate various ICU noise alarms, achieving an average reduction of 22– 23 dB at the patient’s ears. Comparative evaluations confirmed that the proposed algorithms provide a robust balance between convergence speed, stability, and noise attenuation, which is the essential objective of developing any ANC system. Accordingly, such system development for implementation in real ICU conditions is well-grounded, enhancing the relevance and potential impact of research outcomes.

Date of Award	14 Jul 2025
Original language	English
Awarding Institution	University of Portsmouth
Supervisor	Hongjie Ma (Supervisor), Edward Smart (Supervisor) & Maria Machimbarrena (Supervisor)