Earl Hughes III

MIT Department: Electrical Engineering and Computer Science

Undergraduate Institution: Hampton University Main Campus

Faculty Mentor: Thomas Heldt

Research Supervisor: Andrea Fanelli, Minoru Matsushima

Website: LinkedIn



I call Lawrenceville, Georgia my home. Currently, I call Hampton University, where I majored in Electrical Engineering, my Home by the Sea. My passion for music inspired my research interest in Signal Processing, which I believe is a significant crossroad between my major and my passion. Outside the classroom, I enjoy playing music, video games, trading card games, and researching new ways of thinking.

2017 Poster Presentation

2017 Research Abstract

Reducing False Alarms in Intensive Care Units (ICUs) Using Automated Signal Quality Evaluation

Earl Hughes III, Department of Engineering, Hampton University

Andrea Fanelli, Institute for Medical Engineering and Science, Massachusetts Institute of Technology

Minoru Matsushima, Nihon Kohden Innovation Center Inc.

Thomas Heldt, Institute for Medical Engineering and Science, Massachusetts Institute of Technology

In Intensive Care Unit (ICU) environments, staff utilizes high-resolution physiologic waveform data, such as intracranial pressure (ICP) waveforms, to judge the physiologic state and proper treatment of a patient. Monitors collect and display this data, and use built-in alarms to alert staff of abnormalities. Often, the data contains low-quality segments caused by artifacts and noise, which can set off the alarms more often than needed. This fatigues staff and desensitizes them to the alarms, which jeopardizes the well-being of ICU patients under their care. Amelioration of this fatigue and desensitization can be achieved through attenuation of the total alarms the staff must manage. This can be done using an algorithm to identify and quantify the signal quality of physiological waveform signals, and to disregard the low-quality segments.

In this work, we adjudicated the results of such automated signal-quality assessment and developed an approach for comparing two sets of adjudications of ICP waveforms. Previously collected ICP waveforms from hospitals were divided amongst two researchers to manually adjudicate for the locations of clean and noisy segments, using a GUI designed for annotating the segments as noisy or clean. The waveforms were then exchanged for cross-adjudication. MATLAB algorithms were developed to compare these adjudications using the Cohen’s Kappa statistic to quantify the degree of consensus between them, with a value of 1 representing total consensus, and a value of 0 representing very low consensus. One algorithm was annotation-based, while the other was index-based using indices between the annotations. Qualitative results from manual adjudication showed satisfactory observational consensus, despite noticeable differences in adjudication styles. Quantitative results reflect this bittersweet conclusion, with an overall annotation-based Cohen Kappa of 0.0048 over 14,036 annotations, and an overall index-based Cohen Kappa of 0.5388 over 1,328,747 indices. Future work includes testing and development of a reliable automatic adjudication algorithm for ICU monitors.