Razvoj Kompjuterske Vizije za Upravljanje Funkcionalnom Električnom Stimulacijom

Abstract

Sistemi za funkcionalnu električnu stimulaciju (FES) se u terapiji gornjih ekstremiteta, kod osoba koje su usled oštećenja centralnog nervnog sistema uzrokovanog povredom kičmene moždine ili moždanim udarom delimično izgubili moć voljne kontrole pokreta, izdvajaju kao potencijalno najadekvatnija rehabilitaciona tehnika. Putem sistema za FES se električni impulsi odgovarajućih oblika (amplitude, trajanja i frekvencije) dovode do motornog sistema (mišića) koji su zaduženi za pokrete koje subjekat nije u stanju samostalno da uradi, sa ciljem da se datoj osobi omogući izvršavanje nekog funkcionalnog zadataka. Da bi to bilo moguće, sam FES sistem u sebi mora da uključi interpretaciju komandi pacijenta ili terapeuta, regulaciju stimulacionih parametara, generisanje strujnih impulsa i interfejs za prenos energije do senzorno-motornog sistema subjekta koristeći implantibilne ili površinske elektrode.

Kada je reč o gornjim ekstremitetima funkcionalni zadaci koji se postavljaju pred subjekta podrazumevavaju najčešće manipulaciju predmetima, pri čemu korisnik bira tip hvata koji odgovara predmetu i kontroliše trenutke zatvaranja i otvaranja ruke kako bi obavio željenu funkciju. Definisanje i realizacija upravljanja kakvo bi subjektu pružio dovoljno fleksibilnosti uz odgovajući nivo jednostavnosti se i dalje smatra jedinim od otvorenih pitanja. U postojećim FES sistemima korisnik manuelno bira funkciju koristeći prekidač ili sličan periferal računara. Ovaj metod je nepogodan, jer je neophodno da se korinik koncentriše na funkciju koju želi da obavi, a ne na komandu koju treba da pošalje sistemu da bi obavio datu funkciju. U literaturi su prikazana idejna rešenja koja zamenjuju ovaj konvencionalni način komunkiacije i primenjuju neki oblik automatizacije. Rešenja su ponudjena kroz komande glasom, udisanjem i izdisanjem, primenom elektrofizioloških signala (elektromiografija - EMG), senzora pokreta i sile, međutim svi ovi sistemi su i dalje podrazumevali interakciju od strane subjekta koja ga zapravo kroz taj dodatni korak odvraća od zadatka koji je pred njega postavljen, a osim toga ometa pravilnu senzomotornu spregu pri obavljanju odabrane funkcije. Primena mozak-računar (brain-computer ili brainmachine) interfejsa je tema kojoj je u današnje vreme posvećena velika pažnja, a podrazumeva direktno prepoznavanje namere iz elektroencefalograskih (EEG) signala. Ovo, logično rešenje, medjutim, ne pruža odgovarajući nivo pouzdanosti kako bi sistemi postali deo kliničke prakse.

Istraživanja u motornoj kontroli su pokazala da je prirodna kontrola bazirana na vizuelnoj percpeciji, i da na osnovu iskustva osoba bira način hvata i korišćenja objekta od interesa. Ova činjenia je našla primenu u istraživanjima u domenu proteze šake i ruke, i razvijeni su sistemi koji su zasnovani na pojednostavljenom modelu percepcije. Predmet istraživanja predstavljenog u ovoj tezi je unapređenje ove metodologije za automatsku kontrolu sistema za FES gornjih ekstremiteta zasnovanog na kompjuterskoj viziji. U skladu sa tim, definisane su postavke sistema za kompjutersku viziju koja koristi mogućnosti Kinect kamere za odabiranje tipa hvata, ali i povratnu spregu u toku stimulacije. Cilj istraživanja je razvoj sistema (od ideje do inovacije) koji bi se mogao primeniti u kliničkom okruženju za restoraciju hvata i to sa minimalnim angažmanom korisnika rehabilitacionog pomagala. U radu su prikazani novi algoritmi za procenu pozicije i orijentacije ruke, u realnom vremenu, na osnovu slike sa Kinect kamere i rezultati testiranja sistema na zadatku kontrole pozicije podlaktive u odnosu na nadlakticu (stimjulacija fleksora i ekstenzora lakta).

Alternate abstract:

Functional electrical stimulation (FES) based systems for the treatment of upper extremity in people who have, due to the central nervous system leasion, caused by spinal cord injury or stroke, to some extent lost the voluntary control offer great potential as a rehabilitation technique to improve the lost functionality. In a FES system electrical pulses of appropriate electrical parameters (shape, amplitude, duration and frequency) are transmitted to the motor system (muscles) that are responsible for the movements that the subject is not able to perform in a satisfactory manner, with an aim to enable the person perform some functional task with the help of this external muscle activation. In order to function, FES system needs to include a mechanism for interpretation of the patient or therapist commands, control of stimulation parameters, generation of adequate electrical pulses and an interface for transfer of electrical energy to the sensory-motor system of the subject through the implantable or surface electrodes.

Considering the upper extremity treatment, the functional tasks the subject is performing are mostly related to reaching, grasping and manipulating with objects of daily living. In order to perform desired function with the help of electrostimulation, user must, therefore, select the adequate grasp type in respect to the object of interest and timely trigger the stimulation to perform hand opening and closing. The implementation of the control mechanism that would provide the user with the adequate flexibility and the desired level of simplicity is still considered an open question. In conventional FES systems user needs to select the grasp type and trigger the stimulation manualy via keybord, joystick or similar peripheral device. This method is inappropriate because the user should concentrate solely on performing the desired function and not on communicating the desired command to the stimulation system. A variety of methods for ease and automation of the control mechanism can be found in litterature. Control mechanisms based on voice commands, electromyographic (EMG) control, motion sensors and force transducers were developed and tested, however, all these systems still required interaction from the subject that considers additional steps which may distract the subject from the task at hand and could result with the development of the incorrect sensorimotor loop while practicing the desired function. The methodology of the brain-computer interface relys on identifying the intentions of the subject directly from electroencephalography (EEG) signals and, thus, represents the most appropriate solution to the problem. However, this solution still does not provide the adequate level of reliability in order to be used in daily clinical practice.

Previous research in motor control has shown that natural control is based on visual perception, and that the choice of adequate grasp type and appropriate object interaction is based on experience. This fact was first exploited in the field of prostheses hands and arms, where several systems that are based on simplified model of perception were developed. The main topic of the research presented in this thesis is the further improvement of this methodology to enable automatic control of the upper extremity FES system. With this in mind, we have defined the system setup for computer vision that guarantees easy integration of Kinect camera in the standard clinical environment and requires minimal interaction from the user. We developed novel algorithms for the real-time assessment of the position and orientation of the hand based on images from the Kinect camera and we have tested their work for closed loop control of hand position through electrical stimulation of the elbow flexor and extensor muscles (biceps and triceps).

We have introduced the concept of artificial perception and developed computer vision algorithms for detection and identification and objects that the user can interact with, along with the estimation of the intention of the subject, with the aim to enable automatic selection of the appropriate grasp type during the FES therapy. Using the developed algorithms and the wireless communication interface we managed to close the loop in the system for FES assisted grasping. Automatic selection of stimulation parameters by the artificial perception algorithm and timely triggering of the chosen stimulation pattern based on the estimation of the position and orientation of the hand was implemented. Performed tests on three hemiplegic patients confirmed that the use of the proposed methodology reduced the time required for performing the grasping exercises and could thereby shorten and simplify the rehabilitation process.

Computer vision system that enables automatic selection of the appropriate stimulation sequence and can timely activate stimulation during the FES assisted grasping exercise can be considered a direct result of the research presented in this thesis. Most of the developed computer vision algorithms that are result of this research and are in detail described in this thesis could prove useful in various other aspects of the rehabilitation technologies (e.g. assessment of recovery and the augmented reality applications). The main scientific contribution of the research described in this thesis is the design of the control models that are based on biological system and consider sequential integration of artificial perception. Application of this methodology for external control of electrostimulation in the neurorehabilitation system should be considered a significant outcome of applied research.