Edge Computing and Dynamic Vision Sensing for Low Delay Access to Visual Medical Information

1 Abstract — A new met h o d is prop osed t o de crea se t he trans missi on delay of vi sual and non - vi sual m edical r ecord s by using edg e co mputi ng and Dynam ic V isio n S ens ing ( D VS) technolog ies . The simulation results show that th e pr o posed scheme can decre ase the t rans missi on dela y by 89. 15% to 93.23% . T he maxi mum numb er of pati ents w ho can be s erved by edge d evices is analy sed. Keyword s: Medical Recor ds ; Edge Compu ting; DV S; Sensors; T ransmission D el ay I. I NTRODUCTION Electronic Medical R ecords (EMR) integrate patients’ clinical data a nd constitute a main sou rce of reference for their care [ 1 ]. It i s ver y impo rta nt for the ho spital or GP to access patient’s EMR in order to provide timel y medical care [2]. An important trend in med ical informatics is th e adoption of electronic patien t record systems that f acilitate access to clinical infor mation and work to ward preventing the loss or misplaceme nt of infor m ation [ 3, 4 ]. In the past decades, m any teleco mmun ication technologie s have been proposed an d used to process patients’ el ectroni c medical records. With the advan ced telecommunication techno lo gies, t he pa tient s’ medical records are no longer paper - based and can be w e ll organized an d shared among the hospitals, whic h is essential i n delivering a goo d qualit y o f medical care [5]. Clou d compu ting is one of the i mportant technolo gies in the healthcare sect or . In [ 6 ], cloud comp uting is proposed to collect patie nts ’ data . T he pu rpose of the proposed des ign is to reduce the possibilit y of typing mistakes in t he pro cess of medical data collecti on a nd t o provide a l wa ys - on, real - tim e data collecti on as oppos ed to the manual ha ndling of patien t information . In [7], clou d com puting is proposed to share pat i en t me dical records. T he aim in [7] is to provide flexible access for patient s and different medical pr ofessi onals . Optimal adaptive an d cross layer methods in transm ission of data over wireless netw or ks have been studied in the literature [8 - 11 ]. Cloud co mputin g has b een widely applied in v ario us area s and has p rove n to be an effecti ve method to collect and process all the data in the cloud. For the ter minal users ( i.e. mobile users, computer users etc. ) , d iffer ent type s of terminal devices can acc ess the cloud easily and share data via the cloud [12 ] . So by apply ing cloud c omputi ng in hea lthc are area, it ’ s a great convenience for patents, fa mily and medica l Ziy ang C hen, P h.D is w ith the I nformatics Depar tment , King’ s C ollege London , London, UK (corr espondi ng auth or to provid e e - mail: Ziyang .chen@kcl .ac.u k , tel ephon e numbe r: +44( 0)756284 5158 ). Taman na Shikh - Bahaei is with Bar ts and the L ondon, S chool o f Medi cine and D entis try , Quee n Mar y Uni ver sity of L ondon, L ondon, U K (e - mail: t.sh ikh - ba haei @smd1 4.qm ul.ac .uk ). Moha mmad S hikh - Baha ei, Dr. is w ith the I nformatics Depa rtmen t, King ’s C ollege Lond on, London , UK (e - mail: m.sbahaei@ kcl.ac.uk ). care workers who want to access EMR immediately . For the service provider ( i.e. medical care provid er) , the cloud is highly scalable. L a r ger scale services can be expanded easily i f m ore service is dem and ed [13 ] . Also , cloud compu ting can be easily im p lemented . The para digm of cloud c omputin g applied in healthcare is sho wn in Fig ure 1. T he medical data from various sensors is collected by medical care gateway ( i.e. sma rt pho ne o r othe r mobile devices). Medical care gateway transfer s medical data to clou d. Medical data is processed and analysed in the cloud. After medical care providers and family receive d requirem ents from cloud , instruction s ， sugges tions and confi gura tions would send back to p atients. Fig ure 1. The para digm o f cl oud co mputing Ho we v e r, the re’s a n i ncre ase in the a mount of da ta correla ting with the rapidl y in creasing number of IOT (internet of things) d evices . T he s peed of data t rans portation is beco ming the bottlenec k for the Clo ud bas ed computing paradigm [14 ] . For example, wearable s ensors a nd monito ri ng came ra s pro duce huge amount s of image o r vi deo data every second an d add a heavy traffic bur den to th e curr ent net wor k s . Cloud comp uting, e spec iall y, is no t efficient enough for those health devices that require very short r esponse t ime s, th us could cause intolerable network latency. On the o ther ha nd, the d ata which is produced by wearable sensors or oth er health devices , is usually private and confidentia l. Transferring and processing patient information th rough pu blic cloud w ould pose a breach of patient confide ntiality . Edge Co m puting and Dyna m ic Visi on Sensing for Lo w Delay Access to Visua l Medical Infor mation Z iyan g Chen , Tamanna S hikh - Bahaei, Mohammad Shikh-Bahaei 2 Here we p ro pose edge co mputing to cache and share medical records. Edge Computing en ables mobile subscribers to access IT and cloud com putin g services at a close pro ximit y withi n the range of Ra dio Acc ess N et work (R AN) [15 ] . In t he traditional clo ud computi ng, all the i nformation and data are transferred to the cloud net work, the n user s would receive the data through the corr espondin g ba se station. Comparing w ith the tra ditiona l cloud com puting, edgin g co mputin g has t hree main ad van t ages: • E dge co mputing co uld cache and process c omputing task at the edge of t he network witho ut transferri ng to the clou d netw ork. E dge compu ting can off load part or all of networ k traffi c from the clo ud network to the edge, which would drop the networ k l a te ncy and decr eas e the ban dwi dth c onsum ption [16 ] . • F or the i ssue of privacy, caching pat ients’ health da ta at t he ed ge is sa fer. P at i ent s can ha ve their o wn medical information at a close proximity and control it should the health da ta be transfe rred to c loud and service pro viders. • Ed ge computing s av es expens e and does not need more network infr astructure. F or a patient, m obile phone is the edge be tween body sensor s and the clo ud. Fo r a sma rt h o me , gateway or Wi - Fi is the ed ge between ho me sensors a nd the c loud. [14 ] The para digm o f edge co mpu ting i s sho wn in F igure 2. We consider the scenario wher e edge com puting o ffloa d s all the net wor k fro m cloud firstly. R a w m edical data from various sensors i s sent to edge devices. Before being sen t to cloud, medic al da ta is cached and processed at the edge. Medical care providers and family wo uld se nd ins truc tion s, sugge stio ns and c onfig ura tions b ack aft er re ceive d requirem ents from cloud. In this scenario , cloud is used to transfer processed data only. Fig ur e 2. The para digm of e dge com puting For the s cenario that edg e com puting off load part of the net wor k fr o m cloud, part of raw medical data is processed o n the edge. The rest is s ent to cl oud. Medical care pr oviders and fami ly send respons es back after receive requirements from cloud. I n this scenario, cloud is used to transfer processed data and process raw medical data both . E dge c o mputi ng can increase the edge re sponsibilit y and allows computatio n and servi ces to be hosted at the edge [17 ]. Pr eviousl y Zi yang [1 8 ] pro posed a new medical files allocation algor ithm by u sing femto cachi ng to transfer and cache EMR s . The results show a decrease in transmission delay b y 44% to 91% in the n ewly propos ed scena rio. Co mp a r ed to fe mt oc aching, edge comput ing ha s greater security of data, does not requi re infrastructure and before the data is sent t o the core network it is analysed. Also, i n paper [18 ], EMR is d ivided into three levels according to the type and size of th e files ( text/word , digital image an d v ideo ) . Ho we ve r , t he size of video (e.g. p a tient m o nitoring), accounts for a bi g part of the EMR (around 70 %), which causes massi ve d elay whe n tra nsfer ring t he EMR . TABL E I . THE THREE L EVE LS OF EMR EMR The med ic al files The fi rst lev el Text/word The s econ d level Ima ge The t hird level Video /audio I n this p aper, EMR is still divided int o three lev els. Howev er, anot her tec hnology is pr opos ed to decreas e the video size by using a dynamic vision sensor camera [ 19 ] . T his camera no t only save s transm issio n time for th e patient , but also sa ve s more sp ace for other patients wh o need to cache their medical files. In par tic ular , for t he fir st time , we will be exploiting the recently devel oped d ynamic visio n sensor (DVS) tec hnology to achie ve s maller m edical visual data files in our proposed tele - health system. DVS is dif ferent to t he conve ntional video camera ; i t can detect tempora l contrast of br ightness and contai ns an array of asynchronous a utonomous self - signal ing pixel s w hich re spond quick ly t o rela tive ch anges accord i ng t o the intensity of light [20 ]. T he c o nventional video camera s are fram e based, w hich mean s the ca mera c apture s a certain num ber of still im ages in one sec ond and rapi dly play s it b ack , giving the v iewer the illu sion of m ot ion [21 ]. Conve ntional video cameras have tw o main di s advantages : • The first being the limite d number of fram es which will m iss most of the move ment if th e objects m ove too fas t [22]. For example using a conventiona l video camera with 30 fram es per se co nd to r ecord th e punche s of a bo xer ; the cam era cannot av o id th e problem that all the image sensors have the same timing sour ce. This weakness le ads to inad equa te dat a when analy s ing nuance s of the boxer ’s arm m ot ion. Higher frames and sophis ticated processing may im prove this case, how ever, li mited bandwi dth, pow er and computing r esources again lim it this possibi lity. • T he se co nd disadvantage is the huge storage capacit y requ ire d due to t he l arge qua nti ty of dat a pr oduc ed by 3 these cameras . The real - tim e tracki ng system requires larg e com putat iona l eff ort an d is con sequ ent ly don e on high - pe rform ance computer platf orms [23 ]. For exam ple in using a conve ntional video camera to monitor and record the sleep p attern of a n arcole psy pa tie nt at ho me , o nly the dat a relating to the pati ents’ mov e men ts ar e needed. Howe ver, huge unval ued and redunda nt data is include d whe n the subjects are no t moving , t he furniture a nd other unchanging p arts of the roo m . T his sig nificantly increases process ing tim e and caus es a was te of stor age sp ace. In [24 ], the a uthor u se s DVS camera to track the real - tim e move ment o f ve hicle s. T he resul ts s how it c an re duce the computatio n burden signific antly compared to traditio nal traffic surveillance systems. Compare d with t he co nventi on a l camera, the DVS camera h as the fastest response time in rec ognisi ng fast m otion. As ment ioned bef ore, the conventional camera is f rame - based. Ho w ever, the DVS camera is event - based. The rate at which f rames are cap tured is enti rel y depe ndent up on t he ra te of li ght inte nsit y chan ges, whi ch not only provides more accurate analysis for the f as t motions, but also decreases the use of storage c apacity, bandw idth an d power . A DVS camera typically requ ires orders of m agnitude low er storage capacity tha n that of a co nvent ional fr ame - based camera [21 ]. T he workin g pat ter n of DVS is sho wn i n the follo wing chart. TABL E II . THE WOR KING PATTERN OF DVS Moti on of obj ects DVS cam era workin g pattern Fast movi ng Taking sam ples with high er rat es Slow movin g Taking sam ples w ith lower rates No ch anges No sa mples b eing t aken In this p aper : • E dge computing is propo sed to proce ss and cache EMR. Com paring tradi tion al cl oud c omputi ng, Edge computi ng could provide more effici ent a nd safer servi ce s. P atient can control their own medical data , made by w earable sens ors , on the edge . • In parti cular, for the first time, the recently develope d dynamic vis ion sensor (DVS) technology is e xploit ed to achiev e smaller m ed ical visual data fi les in our p roposed tele - heal th sys tem. • A medical fi le s alloca tion algorithm is propo sed to cac he and alloca te EM R on the ed ge. Pat ient and clinical peo ple can access medical r ecords in a timely manner, a nd transmission d elay is reduced when the medical data is being shared and allocated among the hospi ta l. • A scenario that patients share thei r storage capacity of edge d evices is consider ed. More pati ents could get effici ent an d effecti ve m edical care se rvices. II. T HE NEW SYSTEM TO ALLOCATE AND CACHE EM R A ne w sy st e m is proposed to decrease the medical v ide o size and allocate a patient ’s EMR b y us ing edge co m puting and DVS ca mera . A. Ne w System fo r recording an d transmi tting medical visual and non - v isual data In prop osed sy stem we assume that th e registered hospital has the complete EMR of th e p atient s. Patients use wearable sensor devices to moni tor their h ealth conditi on and recor d their location. The wearable sensor h as t wo main functions. The first one is the detectio n o f e merge ncy si tuat ions (e.g. falls) . In an emerge nc y situat ion, a n elect ronic impul se pro duce d by patient’s body may be above or below the critical value. The senso r would the n send an e merge ncy mes sage t o the ho spit al and clinical profession als. T he patien ts can receive immediate medical services [15]. The caregiv ers and family of the pati ent ar e noti fied when t he e mergenc y sit uatio n happ ens . Also, ot her notif ications would be sent in ma ny cases, f or example, if the patient requires assistance in taking th eir medicine. Clinical personnel can remotely monitor patient’s status and be alerted in cas e a medical decis ion has to be made [26] . The se cond main funct ion o f the se nsor is t he recording of the patient’s location. Using this approach, a warni ng ab out the fall and the loca tion of the subject unde rgoing moni toring is t ransmitt ed to a careg iver or famil y member via S MS, email and Twitter messages, etc. A lso, this app roa ch can r ecor d ho w long the pati ent stays i n thi s location. In thi s monito ring s yste m , we insta ll DVS ca mera s and wearable sensors in the home area (e.g. tracking the sleeping patt ern for the p atie nt who is s ufferi ng fr om a sle ep disorder). Also, the DVS camera and wearable sensors are located in their workplace (e.g. for tracking the die t of the diabetic patient ) . In other places, patients use wearable sensors to record current health situations . As mention ed before, v ideo rec ords acc ount f or m ost of the space of EMR s. The size of video files, such as those for reco rd ing falls or m onitoring Alzheimer patients, can be hundr eds of Giga bite s , e specially if a hig h - resolutio n camer a is depl oyed. In ou r daily lives, C IF resolu tion (352x240 ) is t ypi ca ll y u se d b y mid - level sta nd - alone DVR recorders w hen recording re al time video [2 7 ]. It is also typically used b y higher end s yste ms for re mote I nterne t vie wing. M ost monitoring an d surveillance cameras use C IF resolution. The bit rat e of a CIF camera is 512kbps. So in the proposed s yst e m, we use 352x 240 CIF with DV S to record the sleep beha vior p atter n for a whol e night. I NILABS, a company sp ecialized in DVS technology, [28 ] studi ed sleep beha vior pattern b y using a 128x128 DVS ca m era. T he DVS o nly outp uts the sub jec ts’ move me nts. A whole nig ht of s leep c an be rec orded in 100 MB of storage and play ed back in les s than a minute. Activity levels can be automatically extracted and any part of th e recording can be v iewed at one mil lisecond re solution . The bit rat e for 128x12 8 DVS cam era is s uggested t o be 256kbps . So the size of the v ideo made by 352x240 C IF cam era with DVS would be around 200M B. However, the size of recorded visual data using convention al CIF re solution cameras is around 2.35GB (512x12x 3600/8/1 024) for a whole n ight. We study a scenario with limited edge cach ing capacity, and for quantitative an alysis we assu m e that video records with conventio nal CIF resol ution camera require a st orage 4 capacity of 200GB. So the si ze of video records made by CIF resolu tion c amera with D VS would be 16.6 6 GB. There are 6 parts (The registered hospital, home area, work place, family’s h ome, friend’s home and other places ) i n the proposed scenario , a s shown in F igure 3 . The registered hospital has the entire EM R of t he patie nt and would tra nsfer medical records anytime if other hospitals requi red it . E dge device A ( E A) and edge device B ( E B ), are located near patient’s home and workplace respectively . I n t he home area and w ork , DVS cameras and wearable sens ors are used to monito r t he pati ent ’s heal th st atus . Medical data is processed o n the EA an d then transf erred to the macro base station . Als o, if any thin g happens, the edge devices would se nd emergency messages to the neares t hospital. The nearest hospital would chec k the situation of pa tients and access the EA to get the health infor mation. In the work place, DVS cameras and wearable sensors are used as well, which has sa me functio n as the one i n the home area. T he EA and E B can process the medical data an d cache the most effective medical records on the edge (Medical R ecords Allocation Algor ithm, mentio ned in ne xt part). T he regi stered hospital has t he comple te EM R o f the patient a nd upd ates EM R accordingly . T he data produced by D VS ca mera s a nd wearable sensor s are stored in edge caches and transmitted to the registered hospital durin g low data traffic times, e.g. a t night. Also edge caches can req uest the r est medical files from the re gistered hospita l to provide more effective m edical care . Fig ure 3. The propo sed s yste m in hom e are a and w orkplace The Fi gure 4 , sho ws t he who le of the sc enar io with 6 part s. Edge com puting C (EC), e dge com puting D (ED) and edge computing E (EE), are respectively located near each family’s home, friend’s home and other places. In those three part s , only wearable sensors are used to monitor the patient’s healt h s tatus due to a shorter visiting time. E ach edge still has the function to allocate, transfer and cache medical records. Fig ur e 4 . The pro posed sys tem in ho me are a, wo rkplace, fam ily’ s home, frie nd’s home and ot her pl ace s Here we consider a ty pical cas e where a patien t spe nds, on aver age, 1 0 hour s at ho me , 8 hours at wo rk , 3 hour s at t he famil y ho me, 2 ho urs at t heir friend ’s ho me and 1 ho ur i n other locatio ns. W e combine the DVS camera and edge comp uting technologies to decrease t he processi ng time and delay, and i mprove the trans mission efficie ncy of medical files. In the next part, a medical file s allocation algor ithm is propose d to alloca te the most important patient medical files to cache in edge dev ices. Also, we considered a situation in whic h the pat ients c an s hare t he same edge device . T he r esult s sho w how many mor e patie nts t he proposed sy stem c an serve. Results show a greater number of patients can be served th is way . B. Med ical Files Alloca tion Algorithm on the Edg e Due to the limiting stor a g e capacity , each ed ge device is proposed to select the most i mportant a nd suitable med ical files for the h ospital near home, w ork place and other places [24] . We assume the capacities of edge devices are var y . In syste m , the storage capacities of EA, EB, E C , ED and E E are 100GB, 500GB, 15 0G B , 50GB an d 10GB, respecti vely . The ideal situatio n would be such that physician s can check pati ent’s healt h co nditi on without need ing to re quest it fro m their regi s tered hospital . An effective medical file s cachin g algori thm is propos ed by deploy ing the Knapsa ck mo d el with penalty mini mization [18 ] for optim izing the proce ss o f fil e placement. Each Medical fil e is corr elated to a series o f penalty parameters . In proposed sys tem, the sta ying ti me o f patients in eac h location is considered as a factor in optimizing t he caching process. Al so which medical file ha s the best value for clinical professionals according to the disea se condition a nd the file size is consi dered . The opti mization problem is to minimize the o verall penalty in order to determine the EMR with the lowest pena lty and hi ghest p rio rity t o b e stor ed in t he edge d evices 5 [30 ] . T h e s ta yi n g ti me of patient in different area is considered . The p atients w ho sta y longe r in o ne are a sho uld be given higher priorities . Fo r ex a mpl e , if the patients spend most of time at h ome, the edge device of home area should cache more medical f iles comparing other places wh ere patients stay in with less ti me. T he e vents of higher priori ty should be a ssociated with lower penalties [31 ] . T he optimisatio n problem is mode led by Equ ation (1) : Minimize: ff fM Px α ∈ = ∑ Subject to : f x = 0 or 1 (1) W her e P : Mini mum va lue o f penalty . M : The different medical types from complete EMR .{ text, i ma ge , DVS vide o} . f α : T he penalty coefficient of staying time. f x : T he decision about allocatin g medical file f in edge device or not . The penalty coefficient f α is present ed as i n TA B L E III. TABL E III . THE STAYI NG TIME PENAL TY PARAMET ERS Acc ording to th e conditi on of disea se , the medica l files’ value should be con sidered in allocatio n a lgorith m . Fo r example, X - ray images are more im portant th an video files in term s of diagn osing th e pat hogeni c co ndition of patient who is suff ering fracture s [ 32 ]. In t his situation, X - ray im ages have higher priority and low er penalties tha n other type o f medical files. Therefore, the optimizatio n equation (1) is updated by adding penalty paramet er of medical files ’ val ue : Minimize: ff f f fM fM Px y αλ ∈∈ = + ∑∑ Subject to : , ff xy = 0 or 1 (2) W here: f λ : The P enalty coefficien t of the medica l files’ value according to the diseas e . : T he decision about allocating file f in edge devices or not. In proposed sy stem , t he image file is considered as the m ost i mp orta nt file for clinical people to diagno se the patie nt’s disea se . T he vide o file is considered as the leas t important file . Therefore, the penalty coefficient f λ can be presented as in TAB LE IV . TABL E IV . TH E PEN ALTY COEFFICIENT OF M EDI CAL FILE S’ VALUE File Type Penalty Co efficient f λ Ima ge s 1 Text/word 2 DVS Vid eo 3 The size of medical file s is considered as one o f the penalty factors , which is related to tr ansmission time. The more medical files cached in edge devices, the more transmission delay can be decreased. The ideal situation is the edge device s ha ve the whol e EM R and no ne ed to reque st medical files fro m the registere d hospital. DVS camera is used in sys tem to decrease the size of vi deo. However, the capacity storage of each edge device is still limited . C aching more medical files on edge devices could save more transmission ti me . Therefore, a highe r prio rity a nd l o wer penalty should apply when ed ge devices have m ore medical files. Th e opti misation Equati on (2) is u pdated by adding penalty parameters of the tran smission del ay and limited s torage constraint : Minimize: n f f f f ff fM fM fM Px y z αλβ ∈ ∈∈ = ++ ∑ ∑∑ Subject to : fn fn n fM yS ϕ ∈ < ∑ ,, f ff x yz = 0 or 1 (3) Where: f β : The penalty coefficie nt of medical files a fter a ddi ng DVS camera. f z : The decision about allocating medical file m in edge devic es or not by adding DVS . fn ϕ : The size of th e medical records that allocate in ed ge devices n. n S : T he maximu m storage capacity of each edge device . As m entioned bef ore , the patient’ s medical record s are defin ed int o text/word, images and DVS v ideo. Therefore, there are seven co mbinat ions that edge devices would cache . The penalty coefficient f β can be presented as in TABL E V. TA B LE V . PENALTY PARAMETE RS OF TRANSMISSION DE LAY The Medical Files Cache d On Edge Penalty Co efficient Text/w ord, images, DVS video (10 6.66G ) 2 Image s, DVS vid eo (10 3.66G) 4 Text/word , images ( 90G) 6 Im ages ( 87G) 8 Text/word , DVS vid eo (19.66G) 10 DVS Vid eo (16.66G) 12 Te xt/ wor d ( 3G) 14 fD β Staying T ime (hours) 1 2 3 4 … 21 22 23 24 Penal ty Coeff ici ent; 24 23 22 21 … 4 3 2 1 6 Accor ding to the storage capacity of each edge devi ce , and the penalty coeffi cients ( TABL E III , T AB LE IV , TABL E V ) , the optimizatio n mod el in Equa tion (3) is implem ented by usin g LpSol ve. T he o ptimiz ed re sults ar e sho wn in TABL E VI. TAB LE VI . THE OPTIMIZED MED ICAL FILES ALL OC ATION Edg e devic es Medic al files alloca tion Edg e devic es A Text/word s , I mage (90G ) Edg e devic es B Tex t/word, I mages, DVS Video(1 06.66 G) Edg e devic es C Text/w ord, Im ages, DVS Vide o(10 6.6 6G) Edg e devic es D Te xt/ wor ds (3 G) Edg e devic es E Te xt/ wor ds (3 G) III. M ODELING AND S IMULATION A NALYS IS The medical record transm ission dela y [33 ] in the pr opos ed system can be presented as f oll o ws : H TH W TW FM TFM FD TFD O TO D P D P D P D P D PD = ++ + + 12 12 1 2 12 12 1 2 ( )( ) ( ) WW H H FM FM H W FM NN NN N N PPP RR RR R R = ++ ++ + 12 12 1 2 12 ( )( ) OO FD FD FD O NN NN PP R R RR + ++ + (4) W her e : The trans m issio n delay . , , , , are respectively the possibilit ies that patient sp ends ti me at home, work, famil y’ s h o me , fri end ’s ho me , and other places . , , , , are the trans missi on d ela y when medical files are requested from each edge device of home, workplace, family ’ s home, fr iend ’ s home and other places respectively. , , , , are respectively the medical records which are allocate d i n edge device A, B, C, D and E in bits . , , , , are the rest of m edical re cords in bit fro m registered hospital . : T he transmissi on rate of edge devices . : T he trans missio n rate of macro cellular network . Acc ording t o the Poiss on distribution [ 34 ] , t he average transmission delay o f medical records w ith caching can be presented as: 12 12 12 12 !! () () W H K K W H WW K HHK e e NN NN KK D K RR K RR λ λ λ λ − − = ++ + ∑∑ 1 2 12 1 2 12 !! ( ) () FM FD KK FM FD K FM FM K FD FD ee N N NN KK K R R K RR λλ λλ −− + ++ + ∑∑ 12 12 ! () O K O OO K e NN K K RR λ λ − ++ ∑ (5) W here H λ : The length o f s tayin g t i me b y p ossibility in ho me area . K : T he number of occurrences . W λ : The lengt h of sta ying t ime b y poss ibilit y in workp lace . FM λ : The lengt h of sta ying ti me by possibilit y in fami ly ’ s ho me . FD λ : The lengt h of sta ying ti me by possibilit y in frie nds ’ ho me . O λ : T he l ength o f stayi ng t i me in other places . The other si mulation paramete rs of the studied sce nario: • T he proposed scenario is covered by a sing le 3GPP LTE R 8 cell. The edge devices use a simplifie d 802. 11n prot ocol. • T he edge devi ce is im plemented n ear patient s within 100 met er s. • The value s of H λ , W λ , FM λ , FD λ , O λ are 0.41 67, 0.3333, , 0.125, 0.0833 , 0.0417 resp ectively . • The va lues of , , , , ar e s hown in TAB LE VI , w hich are 90G, 10 6.66 G, 106,66 G , 3G, 3 G respectively • T he v al ue s of , , , , are th e re st of EMR, which are 16.66 G, 0G, 0G , 103.66,G 103.66G respectively . The simulation results are analy zed b y MAT LAB an d presented in the follo wing section. IV. SIMULATION RESU LT S The si mulatio n res ults of tr ansmi ssion d ela y are compare d w ith our prev ious res earch [18 ]. In our pr eviou s research, only femtocaching w as used in system. We combined edge caching with DVS camera technology in this work to decrease the size of medica l and improve transmission delay. We s till considered two scenario : the best situation and the worst situatio n. The best situatio n corresponds to the case that the edge caches have the enough medical files. The nearest h ospital can provide ef fective medi c al services and does not need to access th e whole of EMR which cac hed in register ed hospital. T he transmission dela y by usi ng p revio us s yste m (femt ocac hing o nl y) is 16 .59 minu tes. How ever the t ransmissi on delay by applying edge comp uting a nd DV S ca mera tec hnology is 9.872 mi nutes, which is a 40 .5 % improvem e nt. The w orst situation corresponds to the case th at the medical files from edge caches are not enough for the nearest hospital. The nearest hospital in each location need more complete medical data from the registered hospital to check the patients’ health cond ition. Comparin g with the best situation, t he transmission delay would incr ease. The transmission de lay by usin g previous s ystem (femtocac hing only ) is 139.652 m inutes. How e ver t he transmi ssi on de lay by apply ing edge c omputing and DVS camera techn ology is 7 26.855 m i nute s, which i s an 80.77 % i mp r ov e me nt . T h e resul ts ar e sho wn in Fi gure 4 . Figure 4. The figures to show how many delays we ca n save compar ed to the previous system We also compared the trans mission dela y of “ Wi thout edge comput ing and DVS” a nd “ With ed ge co mputi ng and DVS ” . T he resu lts ar e sho wn i n Fig ur e 5 . For the best situation, the tran smission del ay without using the prop osed syste m is 1 45.7 3 minute s. H o w e ver the t ransmis sion delay with DV S and edge c omput ing i s 9.8 72 minute s, which is a 93.23% im provement. For t he worst s ituation, the trans missi on de lay witho ut usi ng t he pr opo sed syst em i s 247.467 m inutes. Howev er the transm ission delay with DVS and edg e com puting is 26.855 m inutes , which i s a 89.15% improv ement. So the propo sed sy stem can improv e the transm ission delay by 89.15% to 93.23%. Figure 5. The figures to show how many delays we ca n save compar ed to the system withou t edge c omputin g and DVS In the proposed system, each pati ent ( whi ch call “Host” ) has their own edge device. We have considered the scenarios in which pa tient s who li ve nea rby (w e cal l “Guest”) can s hare the whole, p art a nd none of the sto rage capacity of edge devices. T he maximu m number of pati ent s who can be served is analyzed. In or der to fin d the maximum number of patients that can be served, w e have t o guar antee that the “Ho st” has extra space in th eir edge devi ce after his/her complete EM R is cached. Also, we considered the si tuation i n whi ch th e “Guest” only cach e s the smallest size of medical files (text/w ord). In t he proposed s cenario , the storage capacity of EA, E D and EE is 100G, 50G and 10G respecti vely. T he t hree edge devices don’t have enough space to cache a complete EMR, thus can’t be shared with other patient s . The storage capacity of EB and EC i s 500G an d 150G. The n umber of patients that can be serve d when edge devices are shared is 132 ((5 00 - 106.66)/ 3+1) an d 15 ((150 - 106.66)/3+ 1). So in the propose d syst em, if the edge devices can be shared, the maximum numbe r of patients can be serv ed is 147 . Al so, t he maxi mum number of patients that can be served improved wh ile the storage capacity of shared edge devices increased. T he r esults a re s hown i n Fi gure 6 . The blue l ine s hows t hat only host patient could be serviced if edge devices cannot be share d with o ther s. Ho wev er, th e number of patient s would increase significantly when edge devices can be shared. Fig ure 6. The nu mber o f patie nts c an be s erv ed whe n storage capacity of sha red ed ge devi ces inc reased V. CONCLUSION In t his research , we introduced a new net work ele men t on the edge to cache, allocate and transfer patients’ medical records. DVS techn ology is proposed to decrease the size of medical video s to drop mor e transmiss ion delay . We also compared the proposed s yst em with ou r previou s work. Th e simula tio n res ults s how t he tr ansmi ssio n del ay ha s a improv ement by 40. 5% to 80.77%. We als o compar ed the situ ation wi thout us ing edge com puting and D VS. T he simula tio n res ults sho w t ha t the proposed s ystem can drop the transm ission delay by 89.15% to 93. 23%. Als o we analys ed the situation that edge devices can be shared with other patients, the result shows the maximum number th at can be serv ed is 147. The m ax imum number is increa sed when the storage capacity of edge devices is improved. In thi s pape r, there are still some issues that can be improved. For exam p le, patients who share t he ir edge devices if ed ge devices have extra capacity st orage. Ho w ever, the re is the issue of sec urit y as well as the issue of sacrificing one’s battery lif e and havin g to deal wi th a longer processing time if one is to sh are the edge device . A clear reward a lgorithm is needed . 9.872 26.855 16.59 139.652 0 50 100 150 The transm ission del ay for the b est situation The transm ission del ay for the worst situ ation T he Total Transmission Delay In Minitus Edge c omputin g and DV S Previ ous system （ Femtoca ching only ） 9.872 26.855 145.73 247.467 0 100 200 300 The transm ission del ay for the b est situation The transm ission del ay for the worst situ ation T he Total Transmission Delay In Minitu s Edge c omputin g and DV S Without edge computing and DVS 8 REFERENCES [1 ] M. D. Rodrígue z, J. Fav e la, E. A . Martínez, M. A . Muñoz . (2004) . Locati on - Awar e Acc es s t o Ho spita l Informati on and Servi ces . IEEE Tr ansactio ns o n info rma tion te ch nology in bi omedicine, Vol. 8 , No. 4 . [2 ] Medical Pract ice Eff ic iencies & Cost Sa vings , Healt h IT.gov, www.health it.gov/ provid ers - profess iona ls/m edica l - practi ce - effic ienc ies -c ost - savin gs. [3] A. W . Te mpleton, S. J. Dw ye r III , J. A. J ohnson, W. H . Ander son, K. S . Hensl ey, K. R. L ee, S. J. Rosent hal, D. F. Preston, and S. Batnit zky. (1985). Im pleme ntatio n of an o n - line a nd long term di gita l mana gement system. Radi ogr aphics vol. 5, no. 1 121 - 138 . [ 4] C. Li u , A. L ong, Y. L i , K . Tsai, H . Kuo. ( 2001). Sharing pa tient care re cor ds ove r the W orld W ide Web. N ational Librar y o f Med ici ne, US , 61(2 - 3):189 - 205 . [5 ] F. S arhan. (2009). Telemedic in e in h ealth care 1: exp lorin g its u ses, bene fits an d disa dvanta ges . Nurs ing T imes.ne t . [ 6] C. O. Rol im, F. L . Koch, C. B. W estphall , J. We rner, A . Fracalos si, G. S. Salvad or . (2010 ). A Cloud Compu ting Soluti on for Pati ent's Data Collection in Healt h C are I nstitutio ns . Sec ond Int ernat iona l Con ferenc e on eHea lth , Telem ed ic ine, and Soc ial M edicin e . [7] M. L i, S. Y u, Y. Z heng, K . Re n, W . Lou. (2012) . Scalab le and Secure Sharin g of Persona l Health Record s in Cloud Computing Using Attr ibute - Ba sed E ncry ption . IE E E T ransact ions o n parall el and distr ibute d systems, Vo l. 24, No . 1 131 - 143 . [8] A Olfa t, M Shikh - Bahae i . (2005). Opt imum powe r and r ate a dapta tio n with i mperf ect ch annel estima tion f or MQ AM i n rayl eigh f lat fa ding c hannel , IE EE 62n d Vehi cular T echnol ogy Confe rence , 200 5. VTC - 2005 - Fall. 2005 : 2468 - 24 71 . Vehi c ular Techno logy Confer ence . VTC - 2005 - Fall. 2005 IEEE 6 2nd 4, 2468 . [9] A Kobravi , M Shikh - Bah aei . (2007 ). Cross - layer a daptive ARQ and modu lation trad eoffs , I EEE Conference on P ersonal, Indoor a nd Mobi le Radi o Communic ations ( P IM R C ), 2007. [10] K Nehra, A Shadmand, M Shikh - Bahaei . (2010). Cross - layer design for int erferen ce - limited s p ectr um shar ing sy stem s , Glo bal Te lecommu nic ations C onference (GLOBECOM 2010 ), IEEE, 1 - 5. [11] F Z arringh alam, B Se yfe , M Shikh - Bahaei, G Charbit, H A ghvami, Join t ly optimized rate and outer loop power con trol with sin gle - and mult i - user d etect ion , IEEE Transactio ns on W ireless Communica tions Volum e 8 (1), 1 86 - 195 , 2 009. [12 ] S. M a rsto n a, Z. Li a, S. B andyopadh yay a, J. Zhang a, A. Ghalsasi b . (2011) . C loud comput ing — the busi ness p ersp ectiv e . Els evier, D ecis ion Supp ort System, Vol. 5 1, No 1 176 - 18 9 . [13 ] Q. Z hang , L. Cheng , R. B ou ta ba . (2010). Clou d computing: state - of - the - art and re search chal lenge s. Jour nal of Inte r net S ervic es and Applicat ions, V ol.1, I ssue 1, 7 – 18 . [ 14 ] W. S hi, J. C ao, Q. Zhan g, Y. L i an d L . Xu . (2015). E dge computin g: Vis i on an d Challenges ”. [15 ] A. A hmed, E. Ahmed. (201 6). A Survey on Mobile Ed ge Computin g ”, 10th IEEE I nt ernat ional con ference on I ntel ligent Systems and Control . [16 ] M . Sa ty anar ayanan, P . Si moens , Y. Xiao, P. Pilla i, Z. Chen , K. Ha, W. Hu, B. Amos . (2015). E dge Analy t ics i n the I nternet of T h ings . IEEE Perva siv e Comput ing, Vol . 14, I ssue. 2, 24 - 31 . [17 ] M . T. Be ck, M. W erner , S. Fel d . (2014). Mobi le Edge compu ting: A Taxonom y ”, The Sixth I nternation al Co nfer ence on Adv ances i n Futur e Inter net . [18 ] Z. Chen, M. Sh ikh - Bah aei . (201 6). Locatio n - Aware Distribu ted F ile Allo cation w ith Fe mtocach ing and DVS for Low - Dela y Acces s to E lectr onic Medical Records , I EEE Engineeri ng in Med icine an d Biolo gy Socie ty, EMBC . [19] Z. Ch en, T. S hikh - Bahaei , P . Lu f f , M. Shikh - Baha ei. (20 17). Ed ge caching and Dy namic Vis ion Se nsing f or low delay access to v isual medical infor mation , IEEE Engi neering in Me d icine and B iol ogy Socie ty, EMB C. [20 ] J. W on, H. Ryu, T. D elbru ck, J. H . L ee, J. Hu. ( 2005). Proximit y Sensing Based on a Dynamic V is i on Sensor for Mob ile Devices. IEEE Tr ansactions on ind ustrial ele ctronic s, Vol. 62, No. 1, January . [21 ] C. P osc h, R. Be nosma n, R.E. Cumm ings . (201 5). Givi ng Mach ines Huma nlike - Viosn si milar to our own would let devic es captu re images more eff iciently ”, Spect rum. I EEE . Org, Nor th Amer ican, Dec . [22 ] E. Grenet, S. Gyger , P. H eim, F. He itger, F . Ka ess, P. Nussb aum, P . F. Ruedi . (2005). High Dy namic R ange Visio n Sens or fo r A uto mo tive Applicat ions. SP IE , vol. 5663, pp . 246 - 25 3. [23 ] P. L ichts tei ner, C. Po sch, T. D elbru ck. (2008) . A 128 128 120 d B 15 u s L atency Async hronous Te mporal Co ntrast V ision Se nsor . IEEE Journal o f solid - stat e circuit s, vol. 43, no. 2 . [ 24 ] M . Litzenb erger, C. Posch, D . Bauer, A. N. Belb ach ir1 , P. Schö n , B. Ko hn, H. Garn . (2006 ). IEEE E mbedd ed Visi on Syst em for Rea l - Time Obje ct T racking usi ng an A synchr onous T rans ient Vis ion Se nsor . 2006 IEEE 12th Digi tal Sig nal Pro cess ing W orksho p & 4th I EEE Sig nal Pr oces sing Educa tio n Wor kshop , pp. 1 73 - 178, 24 - 27 . [ 25 ] S . Aja mi , T. B a ghe ri - Tadi . (201 3). Barriers for Adop ting Elect ronic Heal th Records (EHRs) by Physicians. AC T A IN FOR M AT IC A M E D IC A , 21(2) : 129 - 13 4. [26 ] S. Pate l, H. P ark, P . Bonato , L . Chan, M. Ro dgers. A Rev ie w of Weara ble Sen sors and Systems wit h Appli cation in Re habili tation. Journal of neu roengi neeri ng and rehab ilit ation . [27 ] CIF CCTV Re solution , www .cctvcamer apros.com /CCTV - Resolut ion - s/356. htm#D1 - Resolu tion. [28 ] Case S tudy 4: Slee p disorde r resear ch, http: //ini labs. com/pro ducts /dy namic - v ision - sensors/dvs - case - studies/. [29 ] A. S hadmand, M. Shikh - Bahae i. (2010). Multi - u ser Ti me - Fre quency Dow nlink Sche duling and Reso urce Allo c at ion for LTE Cellu lar S ystems. IEEE C onfer ence on WCN C , 20 10 . [ 30 ] W. Leung, M . Shikh - Baha ei. (2015). A New Femto cac hing File Place me nt A lgor ithm f or T el emedici ne. P roce eding s, I EEE Eng ineer ing i n Med icine an d Biology Soci ety, EMBC 2 015 . [ 31 ] R. M. Freu nd . (2004 ). Penal ty an d Barrie r Metho ds for Constr aine d Optim izatio n. Mass achus et ts I nstitute of T echnol ogy . [32] E . R. Bogoch, V . Elliot - Gib so n , D . E. Beato n , S. Jama l , R. Josse, T. M. Mur ray . (2006) . Effec tive Initiation of Osteoporosi s Diagnosis and Tre atment f or Pat ients with a Fr agility Fr acture i n an Or thope dic Envi ronment . J B one Joint Surg Am . [ 33 ] Comput er N etworks: DELAYS IN TRANSM ISSION. ( 2013 ). [34 ] J. Kingm an, L. Rogers, P. Baxe ndale, P . Gree nwood, F . Kelly , J. Leg all, E. Par doux, D. W illiams. (2 002). Oxford stud ies in pro babili ty: Poiss on Proc esses ,” Oxford Universit y Pre ss.