Francesco Grussu - A study on a bidirectional brain-machine interface inspired by corticospinal control of movement

Attenzione: apre in una nuova finestra. PDFStampaE-mail

Architettura e Ingegneria - Tesi di Laurea

Valutazione attuale: / 12

A study on a bidirectional brain-machine interface inspired by corticospinal control of movement.

A brain-machine interface (BMI) is a device which creates a direct communication between the nervous system (NS) and an artificial system. Different kinds of BMIs exist although many efforts still have to be made in order to develop clinical practice relying on such a kind of devices, which mainly aim at restoring functions of patients suffering from amputations, locked-in syndrome or other kinds of motor or perceptive impairments.

A number of classifications for BMIs exist, according to the criterion which is embraced to group the devices. A first classification relies on the neural signal which is recorded; it groups BMIs in Electroencephalographic (EEG) BMIs, Electrocorticographic (EcoG) BMIs or intracortical BMIs. A second classification identifies input BMIs and output BMIs. Input BMIs deliver patterns of electrical stimulation to Central NS (CNS) or Peripheral NS (PNS) in order to restore a particular impaired function, such as Cochlear implants help in restoring auditory functions. Output BMIs instead record the neural activity and decode it for clinical purposes or to predict the stimuli which elicited the activity itself. An output BMI might be regarded as closed-loop (bidirectional) or open-loop whether the NS gets any feedback information back dealing with the state of the decoding (closed-loop) or not (open-loop). In particular, the feedback might be merely visual or might be provided as a pattern of electrical stimulation.

The aim of the thesis is to study a bidirectional intracortical BMI on rats whose decoding procedure is inspired by the corticospinal control of movement. In particular, according to recent findings, it is assumed that the control policy implemented by the CNS is in the form of convergent muscular force fields characterized by an equilibrium point which is the target position of the limb. These force fields, whose origin can be found in the spinal circuitry, might be regarded as motion primitives and their combination by the CNS might generate complex limb movements. The CNS therefore may encode the limb movement as a Cartesian trajectory (like a moving target point corresponding to the limb effector) as the spinal cord is able to translate such an abstract representation into patterns of muscular forces acting on the limb itself and leading it towards the target. Nonetheless, such a control policy, relying on force fields, is stable and robust towards sudden and undesired perturbations.

The procedure characterizing the BMI studied in the thesis is called dynamic shaping and a BMI based on this procedure is called dynamic BMI (dBMI), as stated in previous works by the supervisors. Its general block diagram is shown in figure 1.

The dynamic shaping characterizing a dBMI

According to the dynamic shaping procedure in figure 1, the neuronal activity of a motor region of the rat brain, represented as the time-dependent firing rate of the recorded units, is acquired by a microelectrode array (a) and decoded by the motor interface (b) which provides a force as output, in order to emulate the spinal cord which translates cortical commands into muscular forces for limbs. The decoded force acts on a dynamical system (c) which, in the simple case presented in the thesis, is simulated and made of a point mass moving through a viscous medium. The state of the system is encoded by the sensory interface as a pattern of Intracortical Microstimulation (ICMS) (d) which is applied via a second array to a sensory region of the rat brain (e) as an artificial feedback signal.

The thesis deals with two main topics. On the one hand, an experimental set-up is developed using commercial hardware and software products. The set-up acquires the intracortical neural activity and, at the same time, is able to deliver up to eight different patterns of ICMS. On the other hand, a new decoding algorithm named IsoANN is studied in order to implement the transformation provided by the motor interface. The new algorithm estimates a map linking the neural responses to a convergent elastic force field during a calibration stage using the dimensionality reduction technique by Tenenbaum and colleagues known as Isomap. Afterwards, the map is implemented with one or more artificial neural networks (ANNs) to make the dBMI operate. The performance of the IsoANN decoding is compared with the one of a previous procedure based on principal component analysis (PCA). For such a purpose, a number of off-line simulations are run on real spike data previously acquired from anaesthetised Long-Evans rats in acute experimental sessions approved by the Italian Ministry of Health (D.Lgs 116/92).

The new set-up has been successfully tested and validated in some experimental sessions.

The IsoANN algorithm, instead, has improved the performance of the decoding stage, since the speed of convergence of the point mass is greater than the one achieved with the previous decoding procedure relying on PCA. Furthermore, in a preliminary feasibility study also included in the thesis Isomap has proved to preserve the intrinsic variability of the calibration neuronal responses better than PCA.

Full references are provided in the thesis bibliography.

Elementi di innovazione introdotti dallo studio. A mio giudizio, il lavoro rivela come una nuova strategia di controllo per interfacce cervello-macchine basata sui campi di forze possa condurre allo sviluppo di una nuova classe di dispositivi, le dynamic brain-machine interface (dBMI), ovvero a una nuova famiglia di neuro-protesi user-friendly che svincolerebbero l’utente da un controllo volitivo costante e faticoso durante l’esecuzione di semplici movimenti stereotipati.

Francesco Grussu


Tesi di Laurea Specialistica

Autore: Francesco Grussu
Relatore: Alessandro Vato
Università: Università degli Studi di Genova
Facoltà: Facoltà di Ingegneria
Corso: Laurea Magistrale in Bioingegneria
Data di Discussione: 09/03/2012
Voto: 110 cum laude
Disciplina: Dispositivi neuro-elettronici e interfacce cervello-macchina
Tipo di Tesi: di Ricerca
Altri Relatori: Marianna Semprini
Lingua: Italiano
Grande Area: Area Scientifica
Dignità di Stampa: Si
In Collaborazione con: Istituto Italiano di Tecnologia, Genova
Settori Interessati: Progettazione di neuro-protesi per amputati o per pazienti affetti da sindrome locked-in

Pubblicata in: