Transcription

UNIVERSITÀ POLITECNICA DELLE MARCHEFACOLTÀ DI INGEGNERIACorso di Laurea Magistrale inBIOMEDICAL ENGINEERINGEvaluation of a model to detect vital signs of asubject trapped in hard-to-reach environmentusing a laser doppler vibrometry techniqueSupervisor:Thesis Written By:Prof. Lorenzo ScaliseCalvin AbongaAcademic Year 2021/2022

ACKNOWLEDGEMENTFirst and foremost, I want to thank God the almighty for seeing me through this project. Secondly,I want to express my deepest appreciation to my supervisor Professor Lorenzo Scalise for hisconstant motivation, inspiration and, guidance throughout the journey of this study. In a specialway, I want to pour out my heartfelt gratitude and appreciation to my co-supervisor EngineerLuca Antognoli for his time, guidance, and tireless assistance while carrying out the experimentsand analyzing the data for this study. This study would not have been possible without myclassmates and friends who voluntarily accepted to take part in it and others who helped in dataanalysis, thank you very much and may God largely reward you all. I also want to extend mygratitude to the UNIVPM fraternity for availing me the opportunity to expand my skill setsthrough Flor scholarship and the DIISM for accommodating me in their laser laboratoriesthroughout this study. Lastly but not the last, I want to thank my parents and family memberswho constantly supported me with words of encourage and prayers which saw me through thisjourney.1 Page

ABSTRACTVital signs detection and monitoring is very key for monitoring the state of life for a patient in aclinical setting or subject trapped in a hard-to-reach environment like war zone, radiation leakedareas etc. In the clinical setting, the contact method is widely used whereas in the hard-to-reachenvironment, the most feasible method is contactless. In this study, cardiorespiratory signal wasacquired contactlessly using the Laser doppler vibrometer (LDV). The experiment recruited 17subjects and 213 data sets of 60 second long were obtained. The LDV signals were preprocessedby filtering out noise at 40Hz, heart rate between (1 to 5 Hz), and Respiratory rate at (0.1 to0.5Hz). Features extracted from the signal included, power spectral density (PSD), root meansquare (RMS), peak to peak intervals. Using the PSD, the behavior of the signal with variousvariables of distances, angles, anatomical positions, skin color and cloth tightness were analyzedas shown in Figure 11-16. Data was then divided into two set. One set was for data obtainedhorizontally from the chest position at a standard distance of 0.5m and angle of 0 degrees. Theother set had data collected from all the protocol variables used in this study. Data from randomenvironmental objects and a resuscitation baby mannequin were as well acquired to simulate ahard-to-reach environment and a dead person. The two data sets were used to create twodatabases in .cvs format. 70% of both databases was used in training the model and 30% formodel validation using a decision tree algorithm with 8 nodes and 42 random. Model 1 produceda classification accuracy of 89.0% and model 2 presented an accuracy of 92.0% when classifyingdata from the random objects and baby mannequin with data from the human subjects. Thesecond model had the best performance when compared to the first model due to the presenceof a large data set.2 Page

Table of ContentsContentsACKNOWLEDGEMENT . 1ABSTRACT. 2Table of Contents . 3List of Figures . 5List of Tables . 7Nomenclature . 8CHAPTER 1 . 91.INTRODUCTION . 91.1.1.1.1.Electrocardiography. . 91.1.2.Photoplethysmography. 111.1.3.Ballistocardiography. 121.1.4.Capnography. 121.2.Contactless based method of cardiorespiratory vital sign monitoring and detection . 131.2.1.Thermal Imaging . 131.2.2.Camera imaging . 131.2.3.Electromagnetic radar-based method . 141.2.4.Laser radar-based methods . 161.3.2.Contact based method of cardiorespiratory vital sign monitoring and detection . 9Aims of the study . 17MATERIALS AND METHODS . 182.1.Materials . 182.2.Study population . 182.3.Experimental setup . 182.4.Data acquisition. 192.5.Data Processing and Analysis. 202.6.Model development . 20CHAPTER 3 . 223. RESULTS. 223.1. Signal Acquisition and representation . 223.2. Variability of the LDV signal power spectral density with the various . 243 Page

3.3. Validation of the LDV signal with the pulse signal . 273.4. Result of the developed models . 29CHAPTER 4 . 354.DISCUSSIONS . 35CHAPTER 5 . 385.CONCLUSIONS . 385.1.Future works. 38CHAPTER 6 . 39REFERENCES . 39APPENDIX . 424 Page

List of FiguresFigure 1: (a) 12 lead ECG lead system (Image obtained from [2]). (b) Ambulatory ECG system (Imageobtained from [2]). (c) ECG signal (Image obtained from [2]) . 10Figure 2: PPG measurement set ups and the DC and AC component of the signal generated (Image wasobtained from [32]). 11Figure 3: (a) Capnography device setup (Image obtained from https://aneskey.com/capnographymonitoring/). (b) Signal of capnography (Image obtained from [34]) . 12Figure 4: Illustrations of cardiorespiratory signal acquisitions using COMS camera (Image obtained from[34]). . 14Figure 5: Image illustrating the working principle of a laser doppler vibrometer. Imaged was obtainedfrom [27] . 16Figure 6: Wavelet decomposition levels. Red color is a representation of frequency range for respiratorysignal and green is for heart rate . 22Figure 7: Representation of Raw LDV and pulse reader signal acquired from the chest of 1 subject . 22Figure 8: Filtered Respiratory activity across the different anatomical positions . 23Figure 9: Filtered Heart activity across the different anatomical positions . 23Figure 10: A comparison of the peak numbers and their positions extracted from both the LDV andfinger pulse sensor reader . 24Figure 11: Variations of different distances with LDV’S power spectral density . 24Figure 12:Variability of LDV’s power spectral density distribution with angles. . 25Figure 13: Variability of LDV’s power spectral density distribution with different anatomical positions. . 25Figure 14: Variability of LDV’s power spectral density with cloth tightness . 26Figure 15: Variability of LDV’s power spectral density with skin color . 26Figure 16: Distribution of heart beats by LDV plotted against the heart beats from Finger pulse reader . 28Figure 17: The mean percentile (50%), high percentage (90%) and low percentile (10%) accuracies of theclassified human subject data varying with distance. (a). Error plot of data classified using the 1st model.(b) Error plot of data classified using the 2nd model . 29Figure 18: Error difference between the accuracy of model 1 and model 2 with distance . 30Figure 19: The mean percentile (50%), high percentage (90%) and low percentile (10%) accuracies of theclassified human subject data varying angles. (a). Error plot of data classified using the 1st model. (b)Error plot of data classified using the 2nd model. . 30Figure 20: Error difference between the accuracy of model 1 and model 2 with Angles . 31Figure 21The mean percentile (50%), high percentage (90%) and low percentile (10%) accuracies of theclassified human subject data varying with anatomical positions. (a). Error plot of data classified usingthe 1st model. (b) Error plot of data classified using the 2nd model. 31Figure 22: Error difference between the accuracy of model 1 and model 2 with Anatomical positions . 32Figure 23: The mean percentile (50%), high percentage (90%) and low percentile (10%) of the classifiedsubject data varying with skin color. (a). Data classified using the 1st model. (b) Data classified using the2nd model. 32Figure 24: Error difference between the accuracy of model 1 and model 2 with skin color . 33Figure 25: The mean percentile (50%), high percentage (90%) and low percentile (10%) of the classifiedsubject data varying with skin color. (a). Data classified using the 1st model. (b) Data classified using the2nd model. 335 Page

Figure 26: Error difference between the accuracy of model 1 and model 2 with cloth fitness . 34Figure 27: Experimental setup for signal acquisition from the chest anatomical position while varyingdistances. . 42Figure 28: Experimental setup for signal acquisition from the chest anatomical position while varyingangles of acceptance. 42Figure 29: Anatomical Positions that were evaluated with LDV device for cardiorespiratory vital signssignal presence . 43Figure 30: An image of the LDV device of type PDV-100 and its outline specifications . 44Figure 31: Ultra-Wide Band (UWB) radar systems . 45Figure 32: Standard breathing Application. Image from apps . 45Figure 33: Set up of the LDV while acquiring signal from the baby Mannequin . 49Figure 34: Principle of CW radar monitoring of the chest movement: phase shift Ɵ(t) caused onthe reflected wave by the chest displacement x(t). . 506 Page

List of TablesTable 1: The various anatomical sites and positions investigated for physiological vital signs using theLDV technology while varying the distances, angles, cloth sizes and skin tone for the chest region on aHuman subject . 19Table 2: The FITZPATRICK SCALE used in the classification of skin color/tone. 19Table 3: Features extracted from the filtered signal to calculate heart rate and respiratory rates perminute. PSD HR, PSD RR Power spectral density of Heart rate and respiratory rate, RMS HR,RMS RR Root Mean Square value for heart rate and respiratory rate, P2P HR and P2P RR Peak to Peakinterval for heart rate and respiratory rate, Disp. HR and Disp. HR displacement of the chest due toheart rate and respiratory rate. . 27Table 4: The accuracy and precision of the 1st model trained with the data collected only from the chestanatomical positions of subjects using the decision tree classification algorithm . 29Table 5: The accuracy and precision of the 2nd model trained with the data collected from all thesubjects using the decision tree classification algorithm . 297 Page

NomenclatureThis is a description of several acronyms used within the textsLDV Laser Doppler VibrometerMRI Magnetic Resonance ImagingECG ElectrocardiographyPPG PhotoplethysmographyLED Light Emitting DiodesiPPG Imaging photoplethysmographyUWB Ultra-Wide BandAD Instrument Analog to Digital InstrumentMATLAB Matrix LaboratoryDIISM Department of Industrial Engineering and Mathematical ScienceUNIVPM Universita Poltechnica Delle MarcheRMS Root Mean SquarePSD Power Spectral Density8 Page

CHAPTER 11. INTRODUCTIONDetection and continuous monitoring of human physiological vital signs, like heartrate andbreath rate, plays a very crucial role in predicting the state of a human being health andwellbeing. Heart rate is the number of beats produced by the heart under one minute whereasbreath rate is the frequency of respiration cycle accomplished under one minute by an individual.The normal heart rate and breath rate of an adult is 60-100 beats per minute and 12-20 breathsper minute [2,3,4]. These vital signs are both detected and monitored either with contact or noncontact methods [2]. It is however quite challenging to detect and monitor these signals in hardto-reach environments like war zones, radiation leaked sites, MRI rooms, infectious sites etc.using the contact methods because of gun shots fear, biochemical hazards exposure, MRI hazardsand biological hazards exposure respectively. According to [1], about 100,000 soldiers arereported dead at battle fronts annually around the globe. These numbers could be reducedexploiting the potential contactless methods offer especially Laser doppler vibrometrytechnique.1.1.Contact based method of cardiorespiratory vital sign monitoring anddetectionThis method of cardio-pulmonary vital signs monitoring involves the attachment of sensors orcontact electrodes directly on the subject’s body to acquire the heart rate and breath rate signals.These methods have their short comings when they are to be applied no subjects in hard-toreach environments as it will be discussed shortly. The following contact-based methods arediscussed; electrocardiography, photoplethysmography, ballistocardiography, and Capnography.1.1.1. Electrocardiography.Electrocardiography is the process of generating heart beats from the electrical activity of theheart while placing electrodes on defined anatomical landmarks of the body [5]. AnElectrocardiograph (ECG) is the signal generated and it is usually meant to monitor heart rate(Figure 1c), however, breathing rate can as well be extracted from it [16, 17], using the sinusarrythmia process [8]. The 12 lead ECG which is the gold standard for heart rate monitoring(Figure 1a) and the portable ambulatory ECG (Figure 1b) are the most common systems of ECGs.9 Page

Just like any type of ECG, these two common types require the placement of electrodes on thesubject’s body and are then connected to a processing unit either wirelessly or with cables whichare often done in clinical settings. The process of electrode placements requires an access to asubject by the third party and this is almost impossible due to the dangers it poses or timeconsuming when one is trapped in a hard-to-reach environment. In other cases, like for burnspatients, elderly, and neonates whose skins are very delicate, the electrodes might causeulcerations, discomfort, strangulation, and irritations among others [9].ECG signalProcessing unit(a)(b)(c)Figure 1: (a) 12 lead ECG lead system (Image obtained from [2]). (b) Ambulatory ECG system (Imageobtained from [2]). (c) ECG signal (Image obtained from [2])10 P a g e

1.1.2. PhotoplethysmographyPhotoplethysmography (PPG) is an optical technique that measures and detects blood volumechanges in tissues’ microvascular bed due to each heartbeat (Figure 2). Photoplethysmograph isthe signal obtained from photoplethysmography [10]. For this method, usually one or more Lightemitting diodes (LED) sensors are directly placed on the anatomical position of interest like thehand wrist, fingers, ear lobes and toes. The wavelengths of the light emitted by the diodes arebetween 500 nm and 600nm which corresponds to yellow and green optical region [11]. Andmost photoplethysmography make use of the green light to acquire heart rate signals. The redand infra-red regions in most photoplethysmography device are used for monitoring bloodoxygen saturation [12].The photodetector which is meant to receive and detect the vital signs is positioned dependingon the acquisition mode. If the acquisition is in the reflectance mode, the photodetector is placedon the same side with the light emitting diode and for the transmission mode, the photodetectoris placed on the opposite side to the light emitting diode [13]. In addition, the technique can aswell be used to estimate respiratory rate because breathing causes changes in the amplitude andfrequency of the signal [13]. Despite the contribution of PPG in cardiorespiratory vital signsmonitoring, its limitations in hard-to-reach environment are quite similar to those presented bythe ECG methods, since all require the placement of sensors directly on the subject’s skin.Figure 2: PPG measurement set ups and the DC and AC component of the signal generated (Image wasobtained from [32]).11 P a g e

1.1.3. BallistocardiographyBallistocardiography is a method that relies on the movements of the chest or body due to thecontraction of the diaphragm and the heart allowing the flow of air into the lungs and blood tothe whole body respectively. These displacements account for up to 7.37cm chest expansioncircumferentially [14]. Most sensors used to measure these displacements are based on strainsensing, impedance pneumography, and movement sensing using gyroscopes, accelerometers,and magnetometers. This technology portrays limitations of strapping the sensors on thesubject’s body prior to monitoring and detection of cardiopulmonary vital signs in hard-to-reachenvironments. The fabrics holding the sensors could as well cause skin irritations and ulcerationsamong the elderlies and neonates as well.1.1.4. CapnographyCapnography is a system made up of carbon dioxide sensors, a gas sampling tube, and a signalprocessing unit (Figure 3) used to detect the amount of carbon dioxide that we inhale and exhaleat each breath [16]. Most of the systems use infra-red and fiber optic sensors. The setup of thesesystems can be side stream, in which the sensors and the main processing unit are placed awayfrom the subject and in contrast, mainstream in which the sensors are between the processingunit and endotracheal tube [15]. This method is unique in the sense that motion artifacts doesn’taffect it however, the chemicals in the sensors could react to other gases, its sensitivity variesbasing on temperature, and humidity of the environment as well [17]. In the hard-to-reachenvironment, this method is impractical because it requires setting up the test on the subject bythe third party.(a)(b)Figure 3: (a) Capnography device setup (Image obtained from https://aneskey.com/capnographymonitoring/). (b) Signal of capnography (Image obtained from [34])12 P a g e

1.2.Contactless based method of cardiorespiratory vital sign monitoring anddetectionThis method of cardiorespiratory vital signs monitoring depends on the subtle movements andother variability on specific anatomical landmarks across the body because of the mechanicalactions of the lungs and heart. To acquire these signals, techniques utilizing the doppler effect,thermal imaging and camera imaging are being exploited.1.2.1. Thermal ImagingThermal imaging relies on modulating radiations from different anatomical positions of the bodyto detect cardiorespiratory signals [2]. The best operating spectrum of this method is infraredand the major four bands utilized are: near infrared (0.75 to 3 µm), middle infrared (3-6µm), farinfrared (6-15 µm) and, extreme infrared (15-100 µm) [18, 19]. To predict cardiac activity, thethermal imaging technique detects the slight heat variability caused by the pulsating bloodflowing in the major superficial arteries in certain anatomical sites. And sensing the respiratorysignal requires detecting small temperature variability around the nostrils using the thermalcamera. This method of non-contact measurement is very advantageous in acquiring signals fromsome anatomical locations and works well under low illumination. The challenges posed bythermal imaging are mainly related to temperature variability and this becomes very impracticalin the hard-to-reach environments. In addition, the signals from the subjects are also affected bythe motion artifacts and the camera can only detect signal at short distance.1.2.2. Camera imagingThe camera imaging technique is divided into color-based method and motion-based method.The color-based method of camera imaging works using the principle of photoplethysmographyof light transmittance and reflectance due to due blood volume [20]. This technique is calledimaging photoplethysmography (iPPG) and can be used to detect both cardiac and respiratoryactivities (Figure 4). A dedicated video camera is used as a photodetector which could be havingthe light emitting source embedded with it in case of a reflectance principle or an independentlight source when using the transmittance principle. The color-based method is associated withvarious advantages such as reliability of results when subject is still or in motion, possibility ofperforming multiple signal acquisitions from various subjects in a short time. However, the13 P a g e

signals are obtained using this method is affected by skin color tone, and image focus. Signals canonly be acquired at short distances and depth.In motion-based method of imaging, a video camera is used to capture small amplitude motionsthat are caused by the mechanical actions of the heart and lungs [21]. The images and videos arethen processed to extract data that are used to predict respiratory rate and heart rate. The twomost important components in this method are the type of camera and a signal processingalgorithm or method used. This method is not affected by skin color tone or physical barrier likecloth during monitoring. However, the problems faced by this method includes motion artifactsand short-range operation distances.Figure 4: Illustrations of cardiorespiratory signal acquisitions using COMS camera (Image obtained from[34]).1.2.3. Electromagnetic radar-based methodElectromagnetic radar-based method of human physiological vital signs detection andmonitoring is divided into; continuous wave, frequency modulated (FM) and ultra-wide band(UWB) technique [2]. The continuous wave radar systems are the simplest and most commontypes of electromagnetic radar-based method that has a transceiver with both a transmitting andreceiving antenna. The transmitter antenna sends a continuous wave of signal to the humananatomical position of interest and the reflected signal is captured by the receiver. Throughdemodulation and processing signal, heart rates and respiratory rates are extracted from the14 P a g e

received signal [2]. The equations illustrating the time domain transmitted signal denoted by T(t),received signal denoted by R(t), and the instantaneous displacements denoted by x(t) areillustrated below [2].T(t) AtCos(ωt φ(t)),R(t) ArCos[ωt -4𝜋𝜆(d0 x(t)) φ(t -(1)2d0𝑐)],x(t) AbCos(ωbt φb) AhCos(ωht φh),(2)(3)where At and Ar, ω, λ, c, d0, Ab, Ah, φb, φh are the amplitudes of transmitted and received signals,angular frequency of transmitted signal, the carrier wavelength, the speed of light, the constantdistance between the antennas and the subject, amplitudes, and phase shifts of the chestdisplacement due to breathing and heartbeat respectively [2].The frequency modulated radar (FM) systems are classified into the frequency modulatedcontinuous wave (FMCW) radars and the stepped frequency continuous wave (SFCW) radars [2].For the FMCW radar systems, the frequency of the signal output with respect to time is linear andin SFCW radars, the output signal compromises of N frames that are linearly transmitted to thesurface of interest with an interval of Δf between each frame. The transceiver architecture of allfrequency modulated radar systems is like that of continuous wave radar systems and as well theiroperations. To decrease the high computational workload of these systems, the received signalsare directly converted with a repli

0.5Hz). Features extracted from the signal included, power spectral density (PSD), root mean square (RMS), peak to peak intervals. Using the PSD, the behavior of the signal with various variables of distances, angles, anatomical positions, skin color and cloth tightness were analyzed as shown in Figure 11-16. Data was then divided into two set.