Akademik Veri Yönetim Sistemi
Tezin Türü: Yüksek Lisans
Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Mühendislik Fakültesi, Elektrik ve Elektronik Mühendisliği Bölümü, Türkiye
Tezin Onay Tarihi: 2018
Öğrenci: ERDEM YANAR
Danışman: YEŞİM SERİNAĞAOĞLU DOĞRUSÖZ
Özet:Generally in hospitals, monitoring devices in the intensive care units (ICU) have high rates of false arrhythmia alarms independent of their brands and prices. These falsely issued alarms have financial and physiological effects such as redundant usage of hospital resources and hassling patients’ rest, reducing sensitivity of the hospital staff to potential emergency cases, which is named as “false alarm fatigue”. According to Deshmane et al. (2009), 43% of arrhythmia alarms in ICUs are false. Moreover, This rate reaches 90% in some of the arrhythmia types. In our study, we considered that the alarms are triggered by five life threatening conditions, which are asystole (ASY), bradycardia (EBR), tachycardia (ETC), ventricular tachycardia (VTA), ventricular flutter/fibrillation (VFB). These alarms are usually triggered by analysis of ECG and pulsatile waveforms recorded by patient monitoring equipments, which have standard alarm triggering criteria such as instantaneous thresholds on the predictor values. Most of the ICU false alarms are caused by single channel artifacts. In this study, we aim to fuse ECG features with information from other independent signals and get more robust alarm algorithms for ICUs. Pulsatile waveforms, which are highly correlated signals, can be used to corroborate the alarm category and to suppress significant number of false ECG alarms in ICUs. Photoplethysmogram (PPG), arterial blood pressure (ABP) or both PPG, and ABP can be used for this purpose. These waveforms are the least noisy pressure signals available in certain ICUs, and rarely contain ECG-related artifacts. We implemented four different algorithms that use information from ECG, PPG and ABP waveforms. We trained and tested these algorithms on Physionet Challenge 2015 database, which consists of 5 main arrhythmia types and total of 750 recordings. These algorithms have main analysis steps as: pre-processing (bandpass filters to remove baseline artifacts, scaling to normalize the amplitude of waveforms), beat detection, alarm decision (for the generic algorithm). Our results show that if we use only ECG data of the whole dataset, we can obtain 88.3% sensitivity and 77.4% specificity with negligible difference in results between two simultaneous ECG channels. When we use ECG with ABP and PPG combinations, our sensitivity was increased by 8% but specificity decreased by 4%. When we use ECG with PPG combinations, our sensitivity was increased by 6.7% but specificity decreased by 7.9%. These improved methods obtained in this work are around the tolerances accepted by expert physicians, and slightly outperform the results of EBR and VFB cases by the other known algorithms evaluated with the same database.