| | SLO | ENG | Cookies and privacy

Bigger font | Smaller font

Show document

Title:ANALIZA SEKVENČNE DEKOMPOZICIJE SESTAVLJENIH SIGNALOV S POMOČJO KOMPENZACIJE KONVOLUCIJSKIH JEDER
Authors:Glaser, Vojko (Author)
Zazula, Damjan (Mentor) More about this mentor... New window
Holobar, Aleš (Co-mentor)
Files:.pdf MAG_Glaser_Vojko_2010.pdf (6,41 MB)
 
Language:Slovenian
Work type:Master's thesis (m2)
Organization:FERI - Faculty of Electrical Engineering and Computer Science
Abstract:Analiza bioelektričnih signalov, ki jih lahko izmerimo na človeškem telesu, je pomemben sestavni del diagnosticiranja v medicini. Klinična diagnoza za mnoge mišične in živčne bolezni se da postaviti dosti zanesljiveje, če lahko ugotovimo, kakšni so prispevki posameznih delov mišic v skupnem bioelektričnem signalu, imenovanem elektromiogram (EMG). V Laboratoriju za sistemsko programsko opremo so razvili dekompozicijski postopek za signale EMG. Temelji na inverzu korelacijske matrike in se imenuje kompenzacija konvolucijskih jeder (CKC). Metoda je zelo uspešna in je bila obširno klinično preizkušena. CKC deluje bločno, z daljšimi odseki signalov, kar ne omogoča analize meritev v realnem času, zato je bil postopek modificiran v sekvenčno različico CKC, imenovano sekvenčna CKC (sCKC). Njeno bistvo je, da dela iterativno in posodablja komponente iz formule CKC med meritvami ob zajemu vsakega novega nabora vzorcev. V magistrski nalogi smo izboljšali algoritme, vgrajene v sCKC, in delovanje nove zasnove preizkusili v različnih, zahtevnih razmerah. Najprej smo preverili vpliv števila vzorcev, ki so vključeni v inicializacijski del postopka. Izhajali smo iz CKC in ugotavljali, kako kratki so lahko signalni odseki, da so dekompozicijski rezultati še zadovoljivi. Pokazalo se je, da CKC da zadovoljive rezultate, če so signali dolgi vsaj 2 do 3 s, medtem ko se število zaznanih motoričnih enot (ME) pri signalih, daljših od 5 s, ne spreminja. Nato smo preverili, kako dobro se sCKC obnese pri dekompoziciji sintetičnih in realnih signalov EMG. V vseh primerih je bil signalom dodan šum različnih moči, opredeljen z razmerjem signal-šum (SNR). Pri razcepu sintetičnih EMG smo primerjali rezultate sCKC in referenčne metode LMMSE (Linear Minimum Mean Square Error). Za ocenjevanje sprejemljivosti dekomponiranih vlakov inervacijskih impulzov smo uporabljali dve meri: senzitivnost (število pravilno postavljenih impulzov) in delež napačno postavljenih impulzov. V vseh šumnih primerih je sCKC zaznala število ME, primerljivo s številom ME pri CKC, to pa je med 5 in 10 ME. Preizkušali smo tudi vpliv števila vzorcev, vključenih v posamezen posodobitveni korak, in ugotovili, da število vzorcev v posodobitvenem koraku vpliva izključno na čas izvajanja sCKC, ki se z večanjem števila vzorcev povečuje s kubom. Z analizo dekompozicije sintetičnih EMG smo lahko nazorno pokazali, da CKC in sCKC uspešno odkrijeta motorične enote (ME), ki so najbliže merilnim elektrodam. Oddaljenost oziroma globina razpoznanih ME v mišici je večja, če je SNR večji. Izboljšano sCKC smo preizkusili tudi z realnimi signali EMG. Izmerjeni so bili pri krčenju dveh različnih mišic, in sicer biceps brachii in tibialis anterior. Za referenčno metodo je služila CKC, saj LMMSE zahteva apriorne informacije o odzivih ME, ki pri realnih signalih niso znani. Ponovno se je sCKC postavila ob bok CKC glede na število zaznanih ME, ki smo jih zaznali med 3 in 9. Edina razlika je, da sCKC ustvari več nepopolnih dekompozicij, ki jih je glede na vse zaznane ME okoli 20 %. Izhodiščna zahteva pri razvoju sCKC je bila, da deluje realnočasovno. Zato smo z analitičnimi izračuni časovne zahtevnosti in izmerjenimi časi posameznih delov algoritma sCKC pokazali, da je sCKC ob pravilni izbiri števila vzorcev v posodobitvi bistveno hitrejša od CKC, vendar na žalost še ne izpolnjuje pogojev za realnočasovno obdelavo.
Keywords:razcep sestavljenih signalov, kompenzacija konvolucijskih jeder, iterativno računanje matričnih inverzov, Sherman-Morrisonova formula, površinski elektromiogrami
Year of publishing:2010
Publisher:[V. Glaser]
Source:Maribor
UDC:621.391:61(043)
COBISS_ID:14776342 Link is opened in a new window
NUK URN:URN:SI:UM:DK:MD9I0ADM
Views:1778
Downloads:86
Metadata:XML RDF-CHPDL DC-XML DC-RDF
Categories:KTFMB - FERI
:
  
Average score:(0 votes)
Your score:Voting is allowed only for logged in users.
Share:AddThis
AddThis uses cookies that require your consent. Edit consent...

Hover the mouse pointer over a document title to show the abstract or click on the title to get all document metadata.

Secondary language

Language:English
Title:VALIDATION OF SEQUENTIAL DECOMPOSITION ON COMPOSITE SIGNALS USING CONVOLUTION KERNEL COMPENSATION
Abstract:The analysis of bioelectrical signals that can be measured on the human body is an important component of medical diagnosing. Clinical diagnosis for many muscular and nerve diseases can be set much more reliable if the contribution of particular parts of muscles are established in the common bioelectrical signal called electromyograms (EMG). The System Software Laboratory developed a decomposition procedure for EMG signals. It is based on the inverse correlation matrix and called Convolution Kernel Composition (CKC). The method is very successful and has been thoroughly clinically tested. CKC operates on longer signal segments, which prevents it to perform in real-time. Therefore, the method was modified and a sequential version was derived under the name sequential CKC (sCKC). The advantage of sequential CKC is that it works iteratively by updating the components of the CKC formula along with the measurements, whenever a new set of samples is available. In this masters thesis, we proposed improvements the algorithms in sCKC and tested them in different complicated situations. First test intended to assess the influence of the number of samples in the initialization part of the algorithm. We derived from CKC in order to determine the smallest length of signals that are properly decomposed. We found out that CKC decomposes correct pulse trains if signals are longer than 2 s, while the decomposition results remain unchanged for the signal length above 5 s. Next, sCKC was tested on synthetic and real signals. In all the cases noise was added with several different signal-to-noise ratios (SNR). In all cases where synthetic signals were used the sCKC results were compared to LMMSE (Linear Minimum Mean Square Error) decompositions. Two performance metrics were used: sensitivity (the number of properly placed pulses) and the false alarm rate (the number of misplaced pulses). In all noisy cases sCKC performed with the same recognition rate as CKC, which means 5 to 10 detected motor units (MU). The influence of the number of samples in each update step was also tested, where it was proved that the number of samples in each update step influences only the computational time, which increases cubically with the number of samples, and not the decomposition quality. Further analysis also showed that both sCKC and CKC decompose pulse trains from MUs that are closer to the electrodes, where the depth of recognized MUs increases with higher SNRs. The upgraded sCKC was then tested on real signals measured from two different muscles: Biceps Brachii and Tibialis Anterior. In this case CKC was used as a reference method, because LMMSE needs prior information on MUs’ responses, which is not available for real signals. Again the sCKC detected as many MUs as CKC, i.e. between 3 and 9. The only difference was that sCKC produced some incomplete decomposition totalling in 20% according to the number of correct detections. The initial requirement in developing sCKC was that it operates in real time. Therefore we used the analytical calculations and measured the time complexity of individual parts of the sCKC algorithm. We showed that sCKC works much faster than the CKC if the optimum number of samples for updating is introduced. However, it still does not qualify for real-time processing when executed on today’s workstations.
Keywords:compound signal decomposition, Convolution Kernel Compensation, iterative matrix inverse computation, Sherman-Morrison formula, surface electromyograms


Comments

Leave comment

You have to log in to leave a comment.

Comments (0)
0 - 0 / 0
 
There are no comments!

Back
Logos of partners University of Maribor University of Ljubljana University of Primorska University of Nova Gorica