Taxonomy on Existing Techniques of Reducing False Alarms in Sensor-Based Healthcare Monitoring - Literature review Example

Taxonomy on existing techniques of reducing false alarms in sensor-based healthcare monitoring systems Number: Lecturer: Taxonomy on existingtechniques of reducing false alarms in sensor-based healthcare monitoring systems There are a number of studies that have been done with respect to false alarm reduction in sensor-based systems. The studies are categorically divided into two: technical approach and sensor configuration. Sensor controls are further sub-divided into classification, clustering, correlation, fuzzy logic and statistical and has been researched by Qassim (2014), Seki (2009), and Yin (2011). Sensor configuration studies have been conducted by researchers such as Charlon (2013) describing sensor localization, Christos (2011) dealing with personal adjustments, and Solet and Barach’s investigating the use of parameters in reducing false alarms. According to Qassim, Patel and Modzhin (2014), fuzzy logic risk analysis is implemented to reduce false alarm instances and maintain sufficient level of security against serious attacks in intrusion and detection systems. The primary purpose of an intrusion and detection system is to identify attackers trying to infiltrate a network and expose vulnerable resources. By using a Fuzzy Logic-Risk Analysis (FLRA) model, the Qassim et al. (2014) aimed to reduce instances of false positives. By calculating the significance and severity of each suspected attack, the system establishes whether an activity can be classified as attempted attack or normal behavior miss judged by the detection system. The model comprises four layers: integrated interface, knowledge manager, autonomic manager, and resource manager. The topmost layer represents the integration point of the system administrator and the intrusion detection system. At this level, strategies and policies are defined and implemented. The section responsible for false positive reduction is the Intrusion Detection Manager. This second layer referred as the Fuzzy Logic-Risk Analysis Intrusion Detection Manager is responsible comprises of four modules: monitor module, analyzer, planner module and control module. All the modules act systematically to detect any attempts of intrusion, eliminate false positives and perform necessary changes to the protected element. Fuzzy Logic-Risk Analysis model works in two mechanisms: identifying the risk and assessing the risk. Once the risk has been identified it is assessed using weighted averages and consequently categorized. Threats are categorized based on their ability to cause harm to asset elements. Once the risk has been identified, residual and exposed risks are analyzed using Fuzzy logic from which different countermeasures are applied to mitigate it. FLRA model works by taking the inputs and outputting applicable ranges which are termed as Critical or Low with intermediaries in between. The output of the inputs is counter measured by applying the three fuzzy sets of “Avoidance”, “Transference” and “Acceptance” ( Tabar, Keshavarz, & Aghajan, 2006). Another fuzzy logic model was proposed by Seki (2009). The fuzzy inference-based model is used to detect daily behavior pattern and different it from non-daily behavior pattern among elderly people. The model used omni-directional vision sensor to detect normalcy from abnormally. The learning is achieved by image feature extraction that is compared against pattern appearance frequency to detect unusual behavior based on probability distribution inference method. However, the model faced problems because it relied on vision sensor and were not useful when light was not available. Yin, Yang and Junfeng (2008) presented a novel two-phased approach for establishing abnormal activity on wireless sensor networks in human body. The model uses a Support Vector Machine approach to filter out activities that are highly likely to be normal. The model also derives abnormal activity using the general normal model through a kernel nonlinear regression (KNLR) to minimize instances of false positives in the unsupervised manner. The model detects abnormal activity by comparing the profiles of normal activity via those of abnormal activity. Any deviation from the normal is flagged as a potential alarm and appropriate action is undertaken. This model uses two-phased pipelined process which comprises of a one-class SVM utilized for filtering out normal activity and minimizing false negatives and a collection of secondary classifiers which detect abnormal activity. A direct approach is used to learn decision boundary around normal and abnormal data. Training a one-class SVM requires the transformation of training traces into a fixed length feature vector. The reasoning behind this is that since traces are generated by concealed mechanisms linked with the users underlying activities, a generative model is better for modeling such data (Tabar, Keshavarz, & Aghajan, 2006) In training the one-class SMV classifier, the model uses Gaussian Radial Basis Function kernel because it accommodates the complexity of the nonlinear and noisy features of the sensor readings. Once the training is accomplished, the author states that the results of the one-class SVM are inputted into the system to output abnormal activities using iterative adaptation procedure. Additionally, a Maximum Likelihood Linear Regression is applied and seeks to establish a set of transformations that can reduce the likelihood of getting mismatches between initial model and adaptation data. The model was applied to a simulation process and results were favorable (Alemdar &Ersoy, 2010). According to Charlon, Bourennane, Betthar, and Campo (2013), a motion sensor network system can be used on different areas of care unit to follow activities of patients in their living environments. In the study, an electronic patch of sensors is worn by patients to identify and detect falls. The system utilizes a network of anchor points to locate tagged subjects within the care unit. The movement of data in care unit is analyzed using algorithms that detect abnormal situations and alert the care provider. Patient suffering from Alzheimers were recruited to participate in the research. The architecture of the system comprised of a local and remote agents. Data concerning the user’s movement was relayed to the local agent via IR motion sensors. The sensing system comprises of an accelerometer, microcontroller and wireless transreceiver. Care providers use a web application to locate user’s movements and alarms. This system was experimented using two patients for a period of one month. Results were encouraging. Out of ten falls, seven was successfully detected with only a single false alarm. Christos, Papadopoulos, and Rosso, (2011) presented CHRONIOUS system, a wearable platform that is applicable for the management of patients with chronic diseases. CHRONIOUS system comprises of a set of sensors and services which integrates together. The system uses data from vital signs, dietary habits, drug intakes, environmental and biochemical parameters and activities of the patient. Since it is a wearable device, it requires no output from the user especially the patient. Abnormal data is captured and reported to healthcare professionals for decision making and analysis purposes. The system architecture works on two layers: a first layer integrated with the Personal Digital Assistance devices and a second layer integrated with the Central Clinical Decision Support System. According to the author, the system presents some risks which must be addressed if false positives are to be eliminated. The performance of the system is based on hardware, software, human and environmental interaction which must be near perfect (Oman, Yaning, Ahmed, & Raouf, 2014). Solet and Barach (2012) presented a non-technical approach to eliminating false positives in healthcare patients with intractable problems. Solet and Barach (2012) developed a phenomenon referred as alarm fatigue. Alarm fatigue has limited potential to identify false and categorize alarm signals leading to delayed and failed alarm responses as well as deliberate alarm de-activation. According to the author, alarm fatigue is caused by false alarm rates which are as high as 83-85%. The author posits that though the large numbers are clinically irrelevant, they contribute to a great deal to staff desensitization. For false alarms to be eliminated, changes in patient’s conditions must be documented, transmitted and communicated sufficiently. The process should be done in a quick manner and in an environment where alarms should be discerned, attributed and acted on. The author presents a joint study between American College of Clinical Engineers and Joint Commission patient safety goals and results indicated that false alarms in the critical care units reduce trust in alarms and prompt caregivers to eliminate them. In a similar study, John Hopkins Hospital conducted a pilot project that probed 16,934 alarm sounds over a 18-day period. Using a multi-pronged alarm management system, the hospital was able to reduce critical alarm false positives by 43% without patient harm (Chambrin, 2015). References Alemdar, H., Ersoy, C. (2010). Wireless sensor networks for healthcare: A survey. Computer Networks, 54(15), 2688-2710. Baumgartner, B.; Rodel, K.; Knoll, A., "A data mining approach to reduce the false alarm rate of patient monitors," Engineering in Medicine and Biology Society (EMBC), 2012 Annual International Conference of the IEEE , vol., no., pp.5935,5938, Aug. 28 2012-Sept. 1 2012 Charlon Y., B. W. (2013). Activity monitoring system for elderly in a context of smart home. Science Direct. Christos B., P. A. (2011). CHRONIOUS: A wearable platform for monitoring and management of patientswith chronic diseases. IEEE. Chambrin, Marie-Christine. “Alarms in the Intensive Care Unit: How Can the Number of False Alarms Be Reduced?” Critical Care 5.4 (2001): 184–188. PMC. Web. 2 Apr. 2015. Hilbe, Johannes, et al. "Development and alarm threshold evaluation of a side rail integrated sensor technology for the prevention of falls." International Journal Of Medical Informatics 79, (January 1, 2010): 173-180. ScienceDirect, EBSCOhost (accessed March 16, 2015). Hons., S. (2009). Fuzzy Inference Based Non-Daily Behavior Pattern Detection for Elderly People Monitoring System. 31st Annual International Conference of the IEEE EMBS. Minneapoolis Minnesota. J. Lester, T. Choudhury, N. Kern, G. Borriello, and B. Hannaford, “A Hybrid Discriminative/Generative Approach for Modeling Human Activities,” Proc. 19th Int’l Joint Conf. Artificial Intelligence (IJCAI ’05), pp. 766-772, July-Aug. 2005. J. Yin, X. Chai, and Q. Yang, “High-Level Goal Recognition in a Wireless LAN,” Proc. 19th Nat’l Conf. in Artificial Intelligence (AAAI’04), pp. 578-584, July 2004. Knoll, G. Nakhaeizadeh, and B. Tausend, “Cost-Sensitive Pruning of Decision Trees,” Proc. 18th European Conf. Machine Learning (ECML ’94), pp. 383-386, Apr. 1994. L. Liao, D. Fox, and H. Kautz, “Learning and Inferring Transportation Routines,” Proc. 19th Nat’l Conf. Artificial Intelligence (AAAI ’04), pp. 348-353, July 2004. Noury, N., Fleury, A., Rumeau, P., Bourke, A. K., Laighin, G. O., Rialle, V., & Lundy, J. E. (2007, August). Fall detection-principles and methods. In Engineering in Medicine and Biology Society, 2007. EMBS 2007 29th Annual International Conference of the IEEE (pp. 1663-1666). IEEE. Oman Salem, Yaning Liu, Ahmed Mehaoua, and Raouf Boutaba, “Online Anomaly Detection in Wireless Body Area Networks for Reliable Healthcare Monitoring”, IEEE Journal Of Biomedical And Health Informatics,VOL. 18,NO. 5, SEPTEMBER2014 Qassim Q., P. A.-Z. (2014). Strategy to Reuce False Alarms in Intrusiuon Detection and Prevention Systems. International Arab Journal of Information Technology, Vol. 11 No. Solet M., B. P. (2012). Managing alarm fatigue in cardiac care. Progress in Pediatric Cardiology, 85-90. Scanaill, C. N., Carew, S., Barralon, P., Noury, N., Lyons, D., & Lyons, G. M. (2006). A review of approaches to mobility telemonitoring of the elderly in their living environment Annals of Biomedical Engineering, 34(4), 547-563. Shih, E. I., Shoeb, A. H., & Guttag, J. V. (2009, June). Sensor selection for energy-efficient ambulatory medical monitoring. In Proceedings of the 7th international conference on Mobile systems, applications, and services (pp. 347-358). ACM. Tabar, A. M., Keshavarz, A., & Aghajan, H. (2006, October) Smart home care network using sensor fusion and distributed vision-based reasoning. In Proceedings of the 4th ACM international workshop on Video surveillance and sensor networks (pp. 145-154). ACM. Wilson, Preventing patient falls. AACN clinical issues, advanced practice in acute and critical care, Preventing Negative Outcomes of Acute Illness in Adults 9 (1) (1998) 100–108 Xiang T., S. Gong, “Video Behavior Profiling and Abnormality Detection without Manual Labeling,” Proc. IEEE Int’l Conf. Computer Vision (ICCV ’05), pp. 1238-1245, Oct. 2005. Yin J., Y. Q. (2008). Sensor-Based Abnormal Human-Activity Dectection. IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, 20(8). Read More