A contrastive learning approach for\xa0ICU false arrhythmia\xa0alarm reduction

Generally in hospitals, false arrhythmia alarm rates are very high in intensive care units (ICUs) patient monitors independent of their brands and prices. These falsely issued alarms disrupt patients’ rest, drain hospital resources, and desensitize the hospital staff to potential emergency situations, which is named as false alarm fatigue. It has been estimated that 43% of life threatening electrocardiogram (ECG) alarms issued by bedside monitors are false, with some categories of alarms being as high as 90%. Ventricular-fibrillation/flutter is the most commonly identified arrhythmia in cardiac arrest patients and it usually ends in death within minutes unless an urgent treatment is not applied. Therefore, in our study, we consider the alarms triggered by ventricular-flutter/fibrillation condition. This type of alarm is usually triggered by ECG and pulsatile waveforms recorded by monitoring equipments, which have standard alarm triggering criteria such as instantaneous thresholds on the predictor values. Most of the ventricular-fibrillation/flutter 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 implement four different algorithms that use information from ECG, PPG and ABP waveforms, and compare the results. Our best result is 100%/98.1% in terms of sensitivity/specificity based on the Cinc2015 Challenge training dataset.

Background: Intensive care units (ICUs) admit high-risk patients whose physiological parameters are continuously monitored to provide timely care. However, traditional ICU monitors are known to generate high rate of false alarms (FAs) resulting in alarm fatigue. In this work, we investigate the use of a hybrid convolutional neural network (CNN) architecture aiming to reduce the burden of FAs. Methods and Results: We retrospectively analyzed 953 independent life threatening alarms from the streaming data of 410 ICU patients. Specifically, we used four channels of electrocardiographic (ECG), arterial blood pressure (BP) and photoplethysmographic (PPG) signals to detect noise in each of the physiological signals, as well as the onset of an arrhythmia. To this end, we developed a hybrid CNN architecture that fuses the information from multiple physiological signals as well as the handcrafted features (Figure 1). Using 5-fold cross validation repeated for 5 times, we obtained an arrhythmia detection accuracy of 87.5% with a standard deviation of 0.5% (Table 1). In practice, the proposed hybrid-CNN algorithm would have suppressed 77.05% of the FA generated by an existing monitoring system, without prior knowledge of the underlying alarm, which is superior to any other published algorithm. Evaluation on the PhysioNet 2015 Challenge database achieved an overall classification accuracy of 93.9%, indicating its efficacy and generalizability. Further, our method remains adaptable to various arrhythmic conditions as well as multiple physiological signals. Conclusions: A hybrid CNN architecture can significantly reduce FA burden and hence improve the quality of care in ICUs

Clinical alarms are highly prevalent in intensive care units (ICUs), serving to alert health care providers when patients' physiological parameters exceed preset thresholds or when medical devices malfunction. However, clinical alarms also introduce new challenges that threaten patient safety. Therefore, enhancing the management of clinical alarms is of critical importance. ICU nurses are the first responders to clinical alarms, and their ability to manage alarms is closely linked to patient safety. Although many studies have investigated ICU nurse-led strategies for managing clinical alarms, a comprehensive overview of the methods and indicators for alarm management in these settings is still lacking. To identify and map management approaches, interventions and evaluation metrics for clinical alarms in ICUs. A systematic search was conducted in PubMed, The Cochrane Library, Web of Science, CINAHL, EMBASE, CNKI, SinoMed and Wan Fang Data databases to include studies related to the management of clinical alarms in ICUs. The included studies (n = 23), guided by various approaches and frameworks, implemented different strategies for managing clinical alarms in ICUs. The primary management measures can be summarized as follows: tailoring alarm parameters, providing education and training on alarm management, standardizing the operation of medical devices, developing and applying assistive tools and technologies and establishing standardized alarm response procedures and protocols. Evaluation metrics addressed three levels: the alarm system, health care professionals and patients. The management of clinical alarms in ICUs remains an area of ongoing exploration, with a primary focus on the management of monitor alarms. Although some progress has been made, more comprehensive and systematic studies are still needed. This review summarizes and consolidates studies on the management of clinical alarms in ICUs, highlighting the importance of multidisciplinary collaboration and the implementation of intelligent alarm management systems. Future research should focus on strengthening interdepartmental collaboration, conducting more specialized studies based on the differences between various ICU specialties, enhancing ICU nurses' alarm management capabilities and further ensuring patient safety.

Ventricular tachycardia (VT) is a life-threatening arrhythmia that requires both relatively rapid and accurate detection in intensive care units (ICUs). Continuous monitoring systems play a crucial role in detecting them. However, previous studies have reported that nearly 9 out of 10 arrhythmia alarms in ICUs tend to be false positives, which usually transpire to a well-documented phenomenon called "alarm fatigue" that leads to desensitization, delayed responses, and increased cognitive burden on healthcare providers. We developed a deep learning based, one-dimensional convolutional neural network (1D-CNN) to classify VT alarms using multiple raw waveform inputs, including two electrocardiogram (ECG) leads, photoplethysmogram (PPG) and arterial blood pressure (ABP) signals. The model was trained using the publicly available VTaC Arrhythmia Benchmark Dataset. We used the 10-second waveform segments that preceded each VT alarm, pre-processed and then used them to train the machine learning model to correctly classify the VT alarm. On the test set, the model achieved an area under the receiver operating characteristic curve of 0.901, overall accuracy of 83.22%, F1-score of 73.3%, sensitivity of 77.53%, specificity of 85.63%, and positive predictive value of 69.57%. The model successfully detected over three-quarters of them while significantly reducing false positive rates for the detection of VT. This study demonstrates that a deep learning based 1D-CNN model using short segments of raw waveform data can achieve robust performance in distinguishing true and false VT alarms.

Development of an algorithm to discriminate between valid and false alarms in a loose-parts monitoring system

This study assessed the feasibility of using multimodal data, namely ECG, ABP and PLETH for reducing the incidence of false alarms in the Intensive Care Unit (ICU) for the PhysioNet/Computing in Cardiology 2015 Challenge. Our approach relies on the annotation of heartbeats using all available channels for each alarm recording. In addition, we also combine ECG and ABP/PLETH channels to create additional signals for analysis. The heartbeat annotations are performed using the gqrs and wabp routines in the WFDB toolbox as well as our in-house algorithm. For ventricular tachycardia alarms, the morphology of the ECG signals for a specified window centered on the annotated heartbeats are also analyzed. Subsequently, the intervals between heartbeats are computed for each channel and for the combined signals. A majority voting scheme with an alarm-specified threshold optimized to the training dataset is used to determine if the triggered alarm is a true or false alarm.

Interdisciplinary collaboration in critical care alarm research: A bibliometric analysis

False arrhythmia alarms pose a major threat to the quality of care in today’s ICU. Thus, the PhysioNet/Computing in Cardiology Challenge 2015 aimed at reducing false alarms by exploiting multimodal cardiac signals recorded by a patient monitor. False alarms for asystole, extreme bradycardia, extreme tachycardia, ventricular flutter/fibrillation as well as ventricular tachycardia were to be reduced using two electrocardiogram channels, up to two cardiac signals of mechanical origin as well as a respiratory signal.In this paper, an approach combining multimodal rhythmicity estimation and machine learning is presented. Using standard short-time autocorrelation and robust beat-to-beat interval estimation, the signal’s self-similarity is analyzed. In particular, beat intervals as well as quality measures are derived which are further quantified using basic mathematical operations (min, mean, max, etc). Moreover, methods from the realm of image processing, 2D Fourier transformation combined with principal component analysis, are employed for dimensionality reduction. Several machine learning approaches are evaluated including linear discriminant analysis and random forest.Using an alarm-independent reduction strategy, an overall false alarm reduction with a score of 65.52 in terms of the real-time scoring system of the challenge is achieved on a hidden dataset. Employing an alarm-specific strategy, an overall real-time score of 78.20 at a true positive rate of 95% and a true negative rate of 78% is achieved. While the results for some categories still need improvement, false alarms for extreme tachycardia are suppressed with 100% sensitivity and specificity.

High false alarm rates in Intensive Care Unit (ICU) is a common problem that leads to alarm desensitization -- a phenomenon called alarm fatigue. Alarm fatigue can cause longer response time or missing of important alarms. In this work, we propose a methodology to identify false alarms generated by ICU bedside monitors. The novelty in our approach lies in the extraction of 216 relevant features to capture the characteristics of all alarms, from both arterial blood pressure (ABP) and electrocardiogram (ECG) signals. Our multivariate approach mitigates the imprecision caused by existing heartbeat/peak detection algorithms. Unlike existing methods on ICU false alarm detection, our approach does not require separate techniques for different types of alarms. The experimental results show that our approach can achieve high accuracy on false alarm detection, and can be generalized for different types of alarms.

Setting a New Gold Standard

Canada’s RADARSAT-2 (R-2) low-resolution ScanSAR Detection of Vessels, Wide swath, Far incidence angle (DVWF) mode was implemented for vessel detection over wide swaths. Between 2013 and 2020, DVWF imagery was used operationally, by the Department of National Defence (DND), and exploited using a tool that detects clusters of bright pixels with vessel-like signature in the SAR imagery. The detected objects were then associated with automatic identification system (AIS) messages. Not all vessels transmit AIS messages, and not all detected vessels that transmit AIS are associated; therefore, analysts observed non-associated objects in an attempt to validate detected vessels. The new work described here evaluates the use of a convolutional neural network (CNN) that detects and rejects false alarms to assist analysts with this validation. A CNN is first calibrated with 1,963 DVWF images containing 16,490 detected and AIS associated vessels (referred to as SAR detected and AIS associated (SAR-AIS) vessels) and 5,654 known false alarms mostly off the coasts of Canada. The CNN’s usability is then evaluated, globally, with 94,562 DVWF images containing 209,746 SAR-AIS vessels, 203,559 validated vessels, and 261,499 unknown objects, detected operationally and non-operationally. During calibration, the CNN classified SAR-AIS vessels and false alarms with 91.6% and 96.3% accuracy, respectively. When evaluating usability, the CNN correctly classified 93.7% of the 209,746 SAR-AIS vessels. False alarms were determined for 5.4% of the 203,559 validated vessels, and 58.1% of the 261,499 unknown objects. These results suggest that a CNN designed to detect and reject false alarms could reduce the number of objects requiring validation by approximately 30%.

In this paper, we propose a new approach to deal with the high rate of false alarms generated by the health monitoring system (HMS). It consists of an intelligent alarm algorithm based on a new version of the support vector machines (SVM). Actually, medical staff face many issues when using the current HMS in intensive care unit (ICU). This latter generates a large number of alarms due to the exceed of the thresholds of the measured physiological parameters. Care-givers should set the thresholds each time there is a new monitored patient or when a new physiological parameter is added. However, in real situations, the default thresholds are often used for all monitored patients which causes a high level of false (and irrelevant) alarms. In order to overcome the ICU issues and improve the current system, we propose the incremental real-time SVM (IRTSVM) for health monitoring system. This new system can deal with data changing over time especially when the state of a patient is not stable. It also deals with incremental aspect by adding new monitoring parameters. Empirical study shows the efficiency of our new system when using real-world databases by providing important results. The new system guarantees the reduction of the rate of false alarms. Besides, it keeps a high level of sensitivity and detects relevant alarms. As a result, it provides doctors by all their needs which makes their decisions more accurate.

With the widespread application of automatic fire alarm systems in various types of buildings, the problem of fire false alarms has gradually become prominent, which not only causes resource waste, but also may reduce users’ trust in the alarm system, thereby affecting the efficiency of emergency response in actual fires. According to data from a certain fire cloud platform, 99.85% of the suspected fires predicted by its system are false alarms. Although existing models can recognize most fire accidents, the accuracy of fire alarm recognition is only 0.15%, due to loose judgment logic, which still requires a large amount of manpower to verify alarms. This article analyzes a large amount of false alarm data and explores the main causes of false alarms, including environmental interference, equipment failure, and improper human operation. By using a fire dynamics simulator (FDS) to establish fire simulation models under different data settings, horizontal and vertical multi-scene fire simulation data are obtained. The study combines simulation and platform data to form a fire and false alarm dataset using a one-dimensional convolutional neural network (1D-CNN) and deep neural network (DNN) deep learning techniques to learn the deductive rules of the fire scene, establish a two-stage judgment model, and gradually, accurately, judge the results. By quantifying the precision, recall, and F1 score of the model, a deep learning model designed to accurately identify genuine fire alarms while filtering out false ones is proposed that can significantly reduce the false alarm rate. The results indicate that the model can identify 1705 false alarms out of 2255 highly suspected true alarms identified by existing systems in multiple practical scenarios and eliminate 75.61% of false positive alarms. On the premise of ensuring an authenticity recognition rate greater than 98%, the accuracy of fire alarm recognition increased from 0.15% to 28.85%, which will significantly reduce the workload of staff verifying alerts, and has good practical value.

For the purpose of applying information technology to the creation of ink style painting, the algorithm of ink painting rendering based on the deep learning framework and convolutional neural network model is designed and improved. Firstly, the ink style rendering program is written in Python. Secondly, VGG under Caffe architecture and Illustration 2Vec models are transplanted to TensorFlow architecture, and the image is rendered in ink style based on deep learning framework and convolutional neural network model. Finally, based on Node.js, the server-side program for image ink style rendering is built. Among them, Express is adopted as the Web-side framework, and the front-end page effect is completed. The results show that the ink rendering logic program is applicable, and the expected purpose is achieved.

Pulse oximetry provides a simple, continuous, noninvasive assessment of arterial blood oxygen saturation. Introduced by Nicolai [1] in 1932, oximetry did not gain widespread acceptance until the 1970s, when it was first reported that arterial hemoglobin oxygen saturation could be measured by analysis of pulsatile blood flow [2]. The pulse oximeter operates by measuring the attenuation of red (660 nm) and infrared light (940 nm) as it passes through a pulsatile tissue bed. Pulsatile blood flow is essential to allow differentiation of arterial from other absorbers of light such as venous and capillary blood components, as well as overlying tissue and skin [2]. Conceptually, the value of pulse oximetry is in identification and correction of hypoxemic events, which, if untreated, could result in serious complications such as brain injury, cardiac arrest, or death [3]. In a review of anesthetic-related, closed malpractice claims, Tinker et al. [4] stated that 31.5% of cases with adverse outcomes could have been prevented through additional monitoring. Eichhorn [5] in a review of 1,001,000 ASA physical status I and II patients, found 11 major intraoperative accidents that could have been attributed solely to anesthesia; seven were deemed preventable. Unrecognized hypoventilation was the most common intraoperative event associated with severe patient injury. Other extensive clinical series document that pulse oximetry increases the identification of hypoxemic events [6,7]. Although pulse oximetry is now well established, the technology has not been greatly improved since 1988 [8], and a number of limitations persist. Kestin et al. [9] reported that 75% of all auditory alarms during routine anesthesia did not originate from changes in physiological variables and only 3% represented patient risk. Freund et al. [10] reported 11,046 cases, with a 1.12% failure rate, defined as the inability to obtain any pulse oximetry reading for a cumulative period of 30 minutes or more after all mechanical problems had been eliminated. A higher rate of failure was present in elderly patients, those with ASA physical status III, IV, and V, and during long or emergency operations. An algorithm developed by Timcenko and Reich [11] predicted a more frequent rate of pulse oximetry failure depending on patient age, ASA physical status, body temperature (<35 degrees C or >37 degrees C), and type of operation (cardiopulmonary bypass, vascular, and orthopedic operations), while Moller et al. [6], by analysis of data from 20,802 patients, identified a total failure rate of 2.5%, increasing to 7.2% in ASA physical status IV patients. A comprehensive review by Severinghaus and Kelleher [8] determined that failure of pulse oximetry could often be explained in terms of low signal-to-noise ratio due to either inadequate signal (e.g., low perfusion, vasoconstrictor use, improper probe placement) or excessive noise (e.g., motion, ambient light, electrocautery, venous pressure waves). Thus, technological evolution recently has been focused on improving function by optimizing the signal-to-noise ratio [12-21] and decreasing the frequency of false alarms [22-24]. Pan and James [23] found that incorporation of a 60-second delay could reduce the false alarm frequency by 88%. However, during a true hypoxemic event, such a delay would reduce the time available for intervention. Egbert et al. [24] developed a neural network pattern classifier for detecting artifact in pulse oximetric signals. Using factor analysis to differentiate pulse oximetry waveforms, they correctly classified 91.4% of the corrupted waveforms and 72.7% of the clean waveforms used to train the network, suggesting that further improvements in signal analysis could reduce artifact and improve performance. Artifact due to motion also can be reduced by coupling an electrocardiograph to the oximeter signal [25]. In this issue, Dumas et al. [12] report the use of another technological advance, Masimo signal extraction technology (SET Trademark), to differentiate the "true" signal from artifact during low perfusion and motion. Recent studies have demonstrated that this technique may have a significantly lower failure rate and a lower false positive alarm rate [12,18,19,21] than conventional oximetric monitoring. Dumas et al. evaluated a prototype Masimo SET Trademark pulse oximeter in 50 ASA physical status I-IV patients in the postanesthesia care unit and found a lower failure rate and a lower frequency and duration of false alarms when compared with conventional pulse oximetry. A similar prototype also alarmed less frequently than a conventional oximeter during inflation of a blood pressure cuff [17]. At present, Masimo SET Trademark awaits Food and Drug Administration approval, although the manufacturers anticipate that the technique will be available for general marketing in late 1996. As anesthesiologists, should we anticipate that this technological advance will improve pulse oximetry sufficiently to advance clinical practice? The answer may vary, depending upon the circumstances in which oximetry is used. In daily practice, the two most annoying limitations of current technology are signal failure in patients with poor peripheral perfusion and signal artifact in patients who have distal extremity movement. Masimo SET Trademark does little to improve the first problem; however, movement artifact complicates pulse oximetry in the postanesthesia care unit and intensive care unit and during emergence from general anesthesia. Anecdotally, the frequency of false alarms is sufficiently high that many personnel are desensitized; they regard a desaturation on pulse oximetry as artifactual until proven otherwise. Perhaps many of us have witnessed delays in treatment because "we didn't trust the monitor!" The Masimo SET Trademark should increase confidence in the validity of alarms, as it increased the incidence of true positive alarms from 12% to 41%. Perhaps our practice will be a bit improved. How much additional cost is justified? Perhaps a bit. However, any technological improvement in pulse oximetry can represent only a small incremental improvement on an enormous advance--and the enormous advance has been shown to make no apparent difference in clinical outcome [6,7].