Video Processing for Remote Respiration Monitoring

Abstract

Monitoring of vital signs is a key tool in medical diagnostics to asses the onset and the evolution of several diseases. Among fundamental vital parameters, such as the hearth rate, blood pressure and body temperature, the respiratory rate plays an important role. For this reason, respiration needs to be carefully monitored in order to detect potential signs or events indicating possible changes of health conditions. Monitoring of the respiration is generally carried out in hospital and clinical environments by the use of expensive devices with several sensors connected to the patient's body. A new research trend, in order to reduce healthcare service costs and make monitoring of vital signs more comfortable, is the development of low-cost systems which may allow remote and contactless monitoring; in such a context, an appealing method is to rely on video processing-based solutions. In this dissertation, novel techniques for the visualization and analysis of respiration by remote video monitoring, based on the study of breathing-related movements, are proposed. Due to the modest extent of movements related to respiration in both infants and adults, specific algorithms in order to efficiently detect breathing are needed. For this reason, motion-related variations in video signals are exploited with various algorithms to identify respiration of the monitored patient and simultaneously estimate the respiratory rate over time. In particular, three algorithms are proposed and analyzed. In one solution, video signals are first processed for motion magnification and then analyzed for the monitoring of respiratory movements. In a second solution, the subtle motion amplification is integrated with the video-processing algorithm for the analysis of movements, with the aim to simplify the processing and improve the performance. In a third solution, a direct Maximum Likelihood-based (ML) video processing algorithm for periodicity analysis is investigated. An important feature of the proposed algorithms is the possibility to detect temporary absence of breathing movements, which may be the indicator of serious diseases, like apneas. Such events can be also caused by life-threatening diseases, that need timely treatments; moreover, in newborns, there may be a risk of Sudden Infant Death Syndrome (SIDS). A significant problem which makes the design and optimization of video processing-based monitoring systems quite critical, is the lack of large video databases, associated with clinical data, obtained from real patients with respiratory disorders. Hence, a Continuous-Time Markov Chain (CTMC) statistical model of breathing patterns, including the possibility to describe respiratory pauses and apnea events, is proposed. The model, tuned and driven by real data extracted from monitored patients, is able to describe realistic breathing patterns. Then, two suitable simulators, software- and hardware-based, have been developed, demonstrating that the proposed CTMC-based statistical model can be strategic to devise simulators useful to test and design novel and effective video processing-based monitoring systems. Finally, performance evaluation of the proposed video processing-based algorithms is performed, by the use of the previously proposed simulators (hardware- and software-based), as well as real cases.