Improved Head Direction Command Classification ...

Proceedings of the 28th IEEE EMBS Annual International Conference New York City, USA, Aug 30-Sept 3, 2006

SaD15.1

Improved Head Direction Command Classification using an Optimised Bayesian Neural Network Son T. Nguyen],

Hung T. Nguyen

I, Philip

B. Taylor], James Middleton'

I Key

2

University Research Centre for Health Technologies, Faculty of Engineering, University of Technology, Sydney, NSW, AUSTRALIA Rehabilitation Research Centre, The University of Sydney, NSW, AUSTRALIA

Abstract-Assistive technologies have recently emerged to improve the quality of life of severely disabled people by enhancing their independence in daily activities. Since man)' of those individuals have limited or non-existing control from the neck downward, alternative hands-free input modalities have become very important for these people to access assistive devices. In hands-free control, head movement has been proved to be a very effective user interface as it can provide a comfortable, reliable and natural way to access the device. Recently, neural networks have been shown to be useful not only for real-time pattern recognition but also for creating user-adaptive models. Since multi-layer perceptron neural networks trained using standard back-propagation may cause poor generalisation, the Bayesian technique has been proposed to improve the generalisation and robustness of these networks. This paper describes the use of Bayesian neural networks in developing a hands-free wheelchair control system. The experimental results show that with the optimised architecture, classification Bayesian neural networks can detect head commands of wheelchair users accurately irrespective to their levels of injuries.

Index Terms- Bayesian neural networks, control, head movement, power wheelchairs. 1.

II.

hands-free

INTRODUCTION

INeE many of disabled people have limited or nonexisting control from the neck downward, alternative hands-free input modalities have become very important for them to access assistive devices. In hands-free control, head movement has been proven to be a very effective user interface as it can provide a comfortable, reliable and natural way to access the device. The need to design challenging systems that are able to detect head direction commands of various forms of power wheelchair users leads to the use of neural networks [1-3]. In such a system, a trained multi-layer perceptron neural network is responsible for classifying input head movement signals into four output classes corresponding to forward, backward, left and right head movement in real-time. Since multi-layer perceptron neural networks trained using the Bayesian technique (Bayesian neural networks) can have superior generalisation properties than standard multi-layer

S

1-4244-0033·3/06/$20.00

©2006 IEEE.

perceptron neural networks in case of fmite available training data, they have been used to develop effective head movement classifiers in our hands-free power wheelchair control systems [4, 5]. In this paper, we continue to optimise the classification Bayesian neural networks used to classify head direction commands of wheelchair users. The structure of the paper is organised as follows. The Section II provides an overview of a head movement-based wheelchair control system. In Section III, the formulation of classification Bayesian neural networks is briefly described. Section IV presents the selection of an appropriate architecture of classification Bayesian neural networks for this application by comparing the log evidence of different candidature architectures and how a disabled user-adaptive head movement classification can be performed. Section V provides a discussion and conclusion for this research. SYSTEM OVERVIEW

The wheelchair platform is based on a commercial powered wheelchair. The movement of the user's head is detected by analysing data from a dual-axis accelerometer installed in a cap worn by the user. The head movement data was collected with a sampling period of 100 ms. A pretrained classification Bayesian neural network is used to classify four gestures in real-time, corresponding to commands for forward, backward, left and right. The start of the head movement to control the travel direction of the wheelchair is determined to be the point where the deviation from the neutral position reached 25% of the maximum value on the relevant axis. The classification of the movement was determined as the first classification made by the Bayesian neural network after the start of the movement. The input to the Bayesian neural network is comprised of a window of 20 samples from each axis. The computer interface module consists of the feedback to the user: real-time graphical displays of the accelerometer data allow the user to track the deviation of their head from the neutral position. Boolean outputs from the classifier and the numerical values of the Bayesian neural network inform the user of how the classifier is interpreting their head movement. The success of the system heavily depends on

5679

the network training. III.

vector of weights and biases in group g.

CLASSIFICA nON

BAYESIAN

NEURAL

NETWORKS

Bayesian learning of multi-layer perceptron neural networks is performed by considering Gaussian probability distributions of the weights which can give the best generalisation [6, 7]. In particular, the weights w in network X are adjusted to their most probable values given the training data D. Specifically, the posterior distribution of the weights can be computed using Bayes' rule as follows

In network training, the hyperparameters are initialised to be arbitrary small values. The cost function is then minimised using an advanced optimisation technique. When the cost function has reached a local minimum, the hyperparameter .;g (g = 1,..., G ) must be re-estimated. This task requires function:

computing

the Hessian matrix of the cost

(5) (w

I D X)

p,

=

p(D I w,X)p(w p(DIX)

I X)

(1)

where where p(D

I w,X)

is the likelihood function, which contains

identity matrix, which selects weights in the g th group. The

information about the weights from observations and the prior distribution p( w I X) contains information about the weights from denominator, p(D

I X),

background is known

knowledge. The as the evidence for

network X. Given a set of candidate networks

H is the Hessian matrix of ED and I g is the

number of 'well-determined'

weights

Yg in group

g

IS

calculated based on the old value of .;g as follows [7]

(g = l, ...,G) (6)

Xi having different

numbers of hidden nodes, the posterior probability of each network can be expressed as

The new value of the hyperparameter estimated as [7]

(2)

If the networks are assumed to be equally probable before any data is observed, then p(Xi) is the same for all networks. Since p(D) does not depend on the network, then the probable network is the one with the highest evidence p(D I Xi)' Therefore, the evidence can be used to compare and rank different candidature networks. Regularisation can be used to prevent any weights becoming excessively large, which can lead to poor generalisation. For a multi-layer perceptron neural network classifier with G groups of weights and biases, a weight decay penalty term proportional to the sum of squares of the weights and biases is added to the data error function ED to

.;g is then re-

(g = 1,..., G) (7)

The hyperparameters need to be re-estimated several times until the cost function value ceases to change significantly between consecutive re-estimation periods. After the network training is completed, the values of parameters Yg and .;g are then used to compute the log evidence of network Xi having M hidden nodes as follows

[8,9]

L:-gW G

lnEv(X;)=-S+

g=l

obtain the cost function

+

2

1 In';g --lnIAI+lnM!+Mln2 2

L:~42 ...!!... ) - Gln(lnn)

(8)

G

g=l

Yg

G

S = ED +

L:';gE

(3)

Wg

g=l

(g

=

1,...,G) (4)

where S is called the cost function, .;g is a non-negative scalar, sometimes knows as a hyperparameter, ensuring the distribution of weights and biases in group g and w g is the

where Wg is the number of weights and biases in group g, and n is set to be 103 [9]. However, n is a minor factor because it is the same for all models and therefore does not effect to the relative comparison of log evidence of different network architectures. Equation (8) is used to compare different networks having different numbers of hidden nodes. The best network will be selected with the highest log evidence.

5680

IV.

EXPERIMENTS

AND RESULTS

A. Data Acquisition Data was collected from eight adults, aged between 19 and 56, with approval from the UTS Human Research Ethics Committee and informed consent from the volunteers. Of these, four had high-level spinal cord injuries (C4 and C5) and were not able to use a standard joystick to control a wheelchair. The remaining four did not have conditions affecting their head movement. Data for each person was collected in two periods of ten minutes, with the user being prompted to give a specified movement every 6 seconds. Each specified movement was chosen randomly from the following: forward, backward, left and right. The extracted movement samples of those users are shown in Table 1.

The performances of the trained networks were tested using a test set taken from the first set of movement samples of user 7 and 8. As shown in Fig.l, the networks having three hidden nodes have the highest evidence. This means that three hidden nodes are sufficient to solve the problem. As shown in Fig.2, these networks also have low test errors (misclassification percentages). The optimisation of Bayesian neural network architecture is extremely important for on-line network training systems because incorporating with the quasi-Newton training algorithm it can contribute to the least network training time while still maintaining the best generalisation for the network. TABLE! EXTRACTED MOVEMENT SAMPLES

B. Optimisation of Network Architecture The recorded movements of user 7 and 8 were randomly divided into two sets. Each set contained 20 samples of each movement. Training data was taken from the recorded movements of users 1,2,3,4,5 and 6, corresponding to 480 samples. Different Bayesian neural networks with varying numbers of hidden nodes were trained to select the optimal network architecture. These networks have the following specification: •

•

four hyperparameters

~1'

~2'

~3

and

~4

Forward 20 20 20 20 20 20 20 20

User I 2 3 4

5 6 7 8

Rizht 20 20 20 20 20 20 20 20

3.

4.

CS C4 C4 CS

the magnitudes of the weights on the connection from the input nodes to the hidden nodes, the biases of the hidden nodes, the weights on the connection from the hidden nodes to the output nodes, and the biases of the output nodes; 41 inputs, corresponding to 20 samples from x axis, 20 samples from y axis and one augmented input with a constant value of 1; four outputs, each corresponding to one of the classes: forward, backward, left and right movement.

4

5

Number of hidden nodes

Fig.l. Log evidence versus number of hidden nodes: The solid curve shows the evidence averaged over the ten networks.

60-------;----;::=:=M===1:;-! 11

c

The weights and biases in four different groups were initialised by random selections from zero-mean, unit variance Gaussians and the initial hyperparameters were chosen to be small values. The network was trained to minimise the cost function S using the quasi-Newton algorithm [5]. When the network training had reached a local minimum, the values of the hyperparameters were reestimated according to equation (6) and (7). Steps 2 and 3 were repeated until the cost function value was smaller than a pre-determined value and did not change significantly in subsequent re-estimations.

~

~

;; >-

I

2'

3h

M =4

!

W

M =

I I

M=

50~

40~

2.

-

to constrain

For a given number of hidden nodes, ten networks with different initial conditions were trained. The training procedure was implemented as follows: 1.

Iniurv Level -

200

3

•

Left 20 20 20 20 20 20 20 20

Backward 20 20 20 20 20 20 20 20

•

:J

~::

o

M

. 30~

=

I

I 7

i '

:!

I

"

o I

i

o

:=c:- t ••

I

-I

10:..

-I

I !

-600

·500

-400

-300

-200

-100

100

Log Evidence

Fig.2. Test error versus log evidence: Every symbol appears ten times, once for each of the ten networks trained.

5681

C. Head Movement Classification

V. DISCUSSION AND CONCLUSION

1) Experiment 1: A Bayesian neural network classifier having three hidden nodes was trained using the training data described in section IV -B. The performance of the trained network was tested using the first set of movement samples of user 7 and 8. The confusion matrix in Table II shows how accurately the trained network classifies each type of movement and the trained network can classify all samples in the test set with an accuracy of 85%. 2) Experiment 2: More training data was taken from the second set of movement samples of user 7 and 8. The Bayesian neural network was trained using the same procedure in Experiment 1. The performance of the trained network was also tested using the first set of movement samples of user 7 and 8. Similarly, a confusion matrix used to evaluate the performance of the trained network as seen in Table III and the network can classify all samples in the test set with a success rate of93.75%. The classification results of Experiment 1 and 2 are summarised in Table IV. Especially, it can be seen that very high sensitivity (true positive) and specificity (true negative) have been achieved for Experiment 2. This means that the performance of the trained network has been significantly improved as more movement samples have been included to train the network. TABLE II CONFUSION MATRIXINEXPERIMENT I Predicted Classification Movement Forward Actual Classification

Backward Left Right Accuracy (%)

Forward Backward 15 0 0 4 0

17 0 0

Left I

3 16 0

Right 4 0 0 20

85

TABLE III CONFUSION MATRIXINEXPERIMENT II Predicted Classification Movement Actual Classification

Forward Backward Left Right Accuracy (%)

Forward Backward 18 0 0 2 0

19 0 0

The results obtained show that Bayesian neural networks can be used to classify head movement accurately. The use of three hidden nodes is an optimal choice for the network architecture as it and the fast training quasi-Newton algorithm can make a significant reduction of on-line network training time. This number of hidden nodes guarantees the best generalisation of the trained network. When the Bayesian neural network was trained using head movement data of the six users, it can classify head movements of two new disabled persons with 85% accuracy. However, if the network was trained further with additional head movement samples of those two disabled persons, it can classify their head movement with a high accuracy of 93.75%. In other words, the network is able to provide an on-line adaptation to the head movement of new disabled wheelchair users. ACKNOWLEDGMENT This research was supported partly by the ARC LIEF grant (LE045408l), ARC Discovery grant (DP0666942) and Key University Research Centre for Health Technologies, UTS. REFERENCES [I] T. Joseph and H. Nguyen, "Neural network control of wheelchairs using telemetric head movement," Proceedings of the 20th Annual International Conference of the IEEE, Engineering Biology Society, vol. 5, pp. 2731 - 2733, 1998.

in Medicine

and

[2J P. B. Taylor and H. T. Nguyen, "Performance of a head-movement interface for wheelchair control," Proceedings of the 25th Annual [3]

International Conference of the IEEE Engineering in Medicine and Biology Society, vol. 2, pp. 1590 - 1593,2003. H. T. Nguyen, 1. M. King, and G. Knight, "Real-time head movement

system and embedded Linux implementation for the control of power wheelchairs," 26th Annual International Conference of the Engineering in Medicine and Biology Society, vol. 2, pp. 4892 - 4895, 2004. [4] S. T. Nguyen, H. T. Nguyen, and P. Taylor, "Hands-Free Control of Power Wheelchairs using Bayesian Neural Networks," Proceeding of IEEE Conference on Cybernetics and Intelligent Systems, pp. 745 749,2004. [5J S. T. Nguyen, H. T. Nguyen, and P. Taylor, "Bayesian Neural Network Classification of Head Movement Direction using Various Advanced Optimisation Training Algorithms," Proceedings of the first IEEE I

Left I

Right

I

o

[6J

18 0

o

pp.448-472,1992b. [7J D. MacKay, "The Evidence Framework Applied to Classification Networks," Neural Computation, vol. 4, pp. 720 -736, 1992. [8] W. D. Penny and S. J. Roberts, "Bayesian neural networks for classification: how useful is the evidence framework," Neural Networks, vol. 12,pp. 877--892, 1999. [9J H. H. Thodberg, "A review of Bayesian neural networks with an application to near infrared spectroscopy," IEEE Transactions on Neural Networks, vol. 7, pp. 56 - 72,1996.

93.75

TABLE IV SENSITIVITY AND SPECIFICITY OFTHENETWORK Experiment Sensitivity Specificity 0.95 0.85 0.9375 0.97917 2

I

20

5682

RAS-EMBS International Conference on Biomedical Robotics and Biomechatronics, Tuscany Italy, 2006,2006. D. J. C. MacKay, "A practical Bayesian Framework for Backpropagation Networks," Computation and Neural Systems, vol. 4,

Acceptance 2006) Home

Start

Acceptance

Decision and Reviews of Submission 126 for EMBC06 (July 4,

,A.ccess \f'Jorbpace

PIN Refresh Logout Feedback

decision for this submission

Decision

Accepted as Contributed Paper

Cover message

Dear Prof. Nguyen Congratulations!!

Your paper entitled:

Improved head direction command classification optimised Bayesian neural network, (126)

using an

has been accepted for presentation at The 28th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, to be held August 30 September 3, 2006 in New York, New York, USA. To allow for possibly minor revisions of papers suggested by reviewers, there is a two-step notification process. This notification advises of the accept status of your paper. Once your final manuscript is submitted, a second, formal announcement will be sent advising you of the paper presentation type (poster/oral) and schedule (date, time, room location). This information will be available on the conference web site within the next two weeks. Please take note of the following

important

deadlines:

10 July 2006: Last date to withdraw papers, complete registration payment, upload your final (possibly revised) "IEEE compliant" paper and complete electronically the IEEE copyright form. 10 July 2006: Last date for registration refund. Refunds will not be issued for authors who have uploaded their papers. 1 August Deadline

2006: for Hotel reservations

For any questions, please contact Jodi Strock at the EMBS Executive Office ([email protected]). Be sure to note "EMBC 2006" in the subject line of your email. If you

require a formal acceptance letter or invitation letter for Visa purposes, please send your request to Jodi Strock by 1 August 2006. We look forward to welcoming

you in New York City!

Sincerely, Andrew

F. Laine and Ki H. Chon

EMBC 2006 Program Chairs

P.S. Due to the extra hard work of your program committee we were able to send out these acceptance letters out sooner than previously announced. Comments to the author Reviews of this submission Reviewer number

Review

10

There are no reviews to be shown

Comments to the author

Attachment

Conference Committee Conference Chair

International Program Committee Chair:

John Clark, USA

Co-Chairs:

Sergio Cerutti, Italy Kenji Sunagawa, Japan

Atam Dhawan, New Jersey Institute of Technology

Conference Co-Chair Metin Akay, Arizona State University Members:

Program Chair Andrew

F. Laine, Columbia University

Program Co-Chair Ki Chon, SUNY-Stony Brook

Publication Andreas

Hielscher,

Chair

Columbia University

Local Arrangement

Chair

Helen Lu, Columbia University

Student Activities

Chair

Jorge E. Monzon, Northeastern National University

Student Activities Co-Chairs Barbara Oakley, Oakland University Helen Lu, Columbia University Tara Alvarez, New Jersey Institute of Technology Anant Madabhushi, Rutgers University Student Program Co-Chairs Winston Tran, University of Southern California Aimee Betker, University of Manitoba Guruprasad Madhavan, State University of New York Lisa Lazareck, University of Oxford Student Advisors Jorge E. Monzon, Northeastern National University Ahmed Abdelal Morsy, Cairo University Joaquin Azpiroz-Leehan, UAM Ziad Abu-Faraj, American University College of Science and Technology Semahat Demir, National Science Foundation Darrin Hanna, Oakland University Student Project Coordinators Faustina Hwang, University of Cambridge Yunfeng Wu, Beijing University of Posts and Telecommunications Jennifer Flexman, University of Washington Student Paper Competition Chairs Christopher James, University of Southampton Joaquin Azpiroz-Leehan, UAM

Ewart Carson, UK Jean-Yves Chapelon, France Patrick Clarysse, France Jean Louis Coatrieux, France Paolo Dario, Italy Andre Dittmar, France Huanqing Feng, China Dimitris, Fotiadis, Greece Alejandro Frangi, Spain Walter Herzog, Canada Dirk Hoyer, Germany Rolf Johansson, Sweden Richard Jones, New Zealand Yasemin Kahya, Turkey Peter Kneppo, Czech Republic Periklis Ktonas, Greece Vi nod Kumar, India Nigel Lovell, Australia Eric McAdams, North Ireland Jorge E. Monzon, Argentina Ozan Ozdamar, USA Thomas Penzel, Germany Jose Principe, USA Laura Roa, Spain Christian Roux, France Maria Ruano, Portugal Wim Rutten, The Netherlands Emilio Sacristan, Mexico Jasjit Suri, USA Nitish Thakor, USA Giuseppe Tritto, France Max Viergever, The Netherlands Andreas Voss, Germany Young Ro Yoon, Korea Josh Zeevi, Israel Y.T. Zhang, Hong Kong Heqin Zhou, China

EtttB

--(0---

Dear Colleagues, It is my great pleasure to welcome you to New York City and the the IEEE Engineering in Medicine and Biology Society (EMBS).

zs" Annual

International Conference of

Over 1,700 technical papers will be presented in oral and poster sessions during the Conference . • 'urnerous workshops and special sessions from Virtual Rehabilitation to Systems Biology and the latest in .nedical devices are available throughout the Conference. We have many distinguished plenary and keynote speakers in addition to over 15 Mini-Symposia presentations. Our keynote speaker is Dr. Nora D. Volkow, Director of the NIH National Institute of Drug Abuse (NIDA). This outstanding conference was organized and its program developed through the dedicated and tireless efforts of Drs. Atam Dhawan, Andrew Laine, Ki Chon and Metin Akay and their EMBC'06 Conference Committee. I would like to extend my sincere thanks to all the volunteers who have contributed to the success of the EMBC'06. With this depth and breadth in the EMBC'06 program, I am sure that you will enjoy learning about cuttingedge research, development, translational efforts and clinical applications in all major areas of biomedical engineering. EMBS encourages you to participate in the extra activities provided for attendees, particularly the student events and Graduates of the Last Decade (GOLD). This is an extraordinary opportunity to meet the future biomedical engineers and emerging leaders around the world. I also hope that you have some time to enjoy the attractions of our host city and celebrate the life and vibrancy of the Big Apple, New York City . . would like to thank NJIT, Columbia University, SUNY Stony Brook and Rutgers University for their support of EMBC'06. In particular, EMBS recognizes all its volunteer members, who work hard to make EMBS events the successes they are, year in and year out. Very special thanks to NJIT and IBM for its generous sponsorship of the Student Activities of EMBC'06. EMBS, the EMBC'06 Conference Committee and our local Chapters welcome you to the EMBC'06 Conference! We trust that you will have a productive, informative and enjoyable time in Manhattan. Best regards,

Yongmin Kim IEEE EMBS President, 2005-2006

Engineering in Medicine and Biology EMS Executive Office • 445 Hoes Lane. Piscataway, New Jersey 08854 USA Phone +1 732981 3433 • Fax +1 7324656435 • E-mail [email protected] • www.embs.org

Dear Colleagues: On behalf of the organizing committee, it is our pleasure to welcome you to the 28th Annual International Conference of the IEEE Engineering in Medicine and Biology Society and to the city of New York, the Big Apple. As you explore New York City, you will find that it represents a fusion of cultures from all parts of the world. It is indeed a great place for us to meet for the largest international conference of biomedical engineers and scientists in the world. Last several decades have witnessed the continued revolution and scientific advances in technology development in the field of medicine and biosciences improving our understanding of physiological processes from molecular to organ levels for better health care. As the Engineering Revolution in BioMedicine continues, the scientific program of the EMBC 2006 presents an outstanding spectrum of over 1700 research papers from more than fifty countries worldwide, emerging from multidisciplinary areas such as biology, physics, chemistry, electrical engineering, mechanical engineering, chemical engineering, nanotechnology, computational sciences and many more to focus on the challenges of developing future technologies for biomedical engineering and medicine. Weare proud to bring you a great synergy of expertise from internationally renowned scientists and pioneers in medicine, biomedical engineering and biosciences through keynote and plenary speakers representing academic, industry, health care, and government sectors. Special pre-conference workshops, mini-symposiums and special sessions throughout the conference are focused on addressing the current and future trends and challenges in scientific research, technology development, and clinical practice. The scientific program is targeted to provide tangible benefits and networking opportunities to current researchers; it also provides a tremendous environment for younger scientists and students to learn about future prospects and professional development activities. The keynote address by Nora Volkow, M.D., Director of the National Institute of Drug Abuse will emphasize the importance of the development of technologies for understanding the physiological processes from molecular to behavioral levels. In addition, the scientific program includes plenary talks from Michael Phelps, Ph.D., Director of Crump Institute of Molecular Imaging, UCLA; Van Mow, Ph.D., Director, Liu Ping Laboratory for Functional Tissue Engineering, Columbia University; John Donoghue, Ph.D., Chair, Dept. of Neuroscience, Brown University; Jennifer West, Ph.D., Director, Institute of Biosciences and Bioengineering, Rice University; Joseph Jasinski, Ph.D., Director, Life Sciences, IBM; David Goldenberg, Sc.D., M.D., Chairman, Immunomedics, Inc; and Ernest Takafuji, M.D., MPH, Director, Biodefense Research, NAlD, NIH. In addition to the outstanding scientific program, we hope that you will find time to explore New York City and the surrounding areas. New York City is a great place to tour and explore on your own. Please take advantage of this opportunity during your stay in the City. We hope that you will enjoy the charming multi-cultural atmosphere of Times Square and stimulating environment of the Hudson river, and will join the feeling, "I Love New York ". Regards,

Atam P. Dhawan, Ph.D., Conference Chair New Jersey Institute of Technology

Andrew F. Laine, Ph.D., Program Chair Columbia University

Ii- If Metin Akay, Ph.D., Conference Co-Chair Arizona State University

~

Ki H.Chon, Ph.D., Program Co-Chair State University of New York at Stony Brook

11:30-11:45 Hardware Considerations of a Spatial Filter for Decorrelating High-Density Multielectrode Neural Recordings, pp. 1244-1247 Thomson, Kyle Michigan State Univ. Oweiss, Karim Michigan State Univ. 11:30-11:45 Affective State Control for Neuroprostheses, Doyle, Thomas E. Kucerovsky, Z. leta, Adrian

pp. 1248-1251 McMaster Univ. Univ. of Western Ontario Murray State Univ.

ThBP11 EMG Analysis

Westside Ballroom South (Poster Session)

10:45-11:00 Multi-Receiver Precision Decomposition Nawab, Syed Hamid /otiz, Robert Oe Luca, Carlo J.

of Intramuscular

11:00-11 :15 How Much Can We Trust the Electromechanical Using Electromyography?, pp. 1256-1259 Conforto, Silvia Mathieu, Pierre A. Schmid, Maurizio Bibbo, Daniele Florestal, Joel R. D'Alessio, Tommaso

EMG Signals, pp. 1252-1255 Boston Univ. Boston Univ. Boston Univ.

Delay Estimated by Univ. Roma TRE Univ. de Montreal, Canada Univ. Roma TRE Univ. Roma TRE Univ. de Montreal, Canada Univ. Roma TRE

11:15-11 :30 Resolution of Superpositions in EMG Signals Using Belief Propagation: Results for the Known Constituent Problem, pp. 1260-1263 ETH Zurich Koch, Volker Maximillian VA Palo Alto Health Care System McGill, Kevin Loeliger, Hans-Andrea ETH Zurich

1 .. ';P12 Neural Engineering

Westside Ballroom South (Poster Session)

10:45-11 :00 A Four Command BCI System Based on the SSVEP Protocol, pp. 1264-1267 Maggi, Luca Parini, Sergio Piccini, Luca Panfili, Guido Andreoni, Giuseppe

Pol. Pol. Pol. Pol. Pol.

di di di di di

Milano Milano Milano Milano Milano

11:00-11:15 A Comparison of Neural Feature Extraction Methods for Brain Machine Interfaces, pp. 1268-1272 Gilmour, Tim The Pennsylvania State Krishnan, Lavanya The Pennsylvania State Gaumond, Roger The Pennsylvania State Clement, Ryan The Pennsylvania State 11:15-11:30 Low-Cost Motivated Rehabilitation System for Post-Operation Exercises* Novak, Daniel Brutovsky, Jan

Univ. Univ. Univ. Univ.

Czech Tech. Univ. in Prague Czech Tech. Univ. in Prague

1:30-11 :45 Behavior Study of the Effects of Visual Feedback on Motor Output, pp. 1273-1276 Hou, Wensheng Bioengineering Inst. of Chongqing Univ. Zheng, Jun Queens Coil. - City Univ. of New York Jiang, Yingtao Univ. of Nevada Shen,Shan Un~.ofSurrey Sterr, Annette Univ. of Surrey Szameitat, Andre J. Univ. of Surrey van Loon, Monique Univ. of Surrey 1:45-12:00 'lnit« Element Modeling of Electric Field Effects of TASER Devices on Nerve and Muscle, pp. 1277-1279 Panescu, Dorin St. Jude Medical Kroll, Mark Cal Pol. Univ. San Luis Obispo Efimov, Igor Washington Univ. Sweeney, James Arizona State Univ. I

2:00-12:15 I Fmri Study of the Cross-Modal Interaction in the Brain with an Adaptable EMG Prosthetic land with Biofeedback, pp. 1280-1284 Hernandez Arieta, Alejandro Univ. of Tokyo Kato, Ryu The Univ. of Tokyo Yokoi, Hiroshi Univ. of Tokyo Arai, Tamio The Univ. of Tokyo 2:00-12: 15 ittects of Prolonged Stimulation at the Electrode-Retina Interface, pp. 1285-1287 Ray, Aditi Univ. of Southern California Chan, Leanne Univ. of Southern California Thomas, Biju Univ. of Southern California Weiland, James Univ. of Southern California 2:00-12:15 ';urative Effect of General Geomagnetic Therapy, pp. 1288-1290 Rotov, Anton Garilevich, Boris Zubkov, A. D. Olefir, Yu. V. Andriyanov, Yuri 12:00-12:15 rwo Channel EEG Thought Pattern Classifier, pp. 1291-1294 Craig, Daniel Ashley Nguyen, Hung Burchey, Harold Andrew

Central Clinical Air Forth Central Clinical Air Forth Central Air-Force Clinical Central Air Force Clinical

Hospital Hospital Hospital Hospital TRINITI

Univ. of Tech. Sydney Univ. of Tech. Sydney Univ. of Tech. Sydney

12:00-12:15 =Iectrooculogram Based System for Computer Control Using a Multiple Feature ';Iassification Model, pp. 1295-1298 Kherlopian, Armen Columbia Univ. Columbia Univ. Sajda, Paul Univ. of California, San Diego Gerrein, Joseph Columbia Univ. Vue, Minerva Columbia Univ. Kim, Kristina New York Medical Coil. Kim, Ji Won Columbia Univ. Sukumaran, Madhav

09: 15-09:30 Modulation in Spinal Circuits and Corticospinal Connections Following Nerve Stimulation and Operant Conditioning, pp. 2138-2141 Thompson, Aiko K. Wadsworth Center, NY State Dept. Health Univ. of Alberta Stein, Richard B. Wadsworth Center, NY State Dept. Health and SUNY Albany Chen, Xiang Yang Wolpaw, Jonathan R. Wadsworth Center, NY State Dept. Health and SUNY Albany 09:30-09:45 Electrical Stimulation Augmented Rehabilitation of Hemiparetic Gait* Sinkjaer, Thomas 09:45-10:00 Transcutaneous Electrical Stimulation Therapy Applications, pp. 2142-2145 Popovic, Milos R.

Aalborg Univ.

Technology for Functional Electrical

10:00-10:15 Huf)rid Assistive Systems for Rehabilitation: i rapy in Hemiplegics, pp. 2146-2149 Popovi j, Dejan Popovi_;, Mirjana

Univ. of Toronto Lessons Learned from Functional Electrical Aalborg Univ. Aalborg Univ.

10:15-10:30 Dynamic Balance Training with Sensory Electrical Stimulation in Chronic Stroke Patients, pp. 2150-2153 Worms, Georg Univ. of Glasgow Matja;i_;, Ziatko Inst. for Rehabilitation, Republic of Slovenia Gollee, Henrik Univ. of Glasgow Cikajlo, Imre Inst. for Rehabilitation, Republic of Slovenia Goljar, Nika Inst. for Rehabilitation, Republic of Slovenia Hunt, Kenneth J. Univ. of Glasgow FrA03 Neural Networks

and Support Vector Machines in Biological

Signal Processing

SoHo-Herald (Poster Session)

09:00-09:15 Multiple ECG Beats Recognition in the Frequency Domain Using Grey Relational Analysis, pp. 2154-2158 Lin, Chia-Hung Kao-Yuan Inst. of Tech. J, Yi Chun National Cheng Kung Univ. Chen, Yung-Fu China Medical Univ. Chen, Tainsong National Cheng Kung Univ. 09:15-09:30 Robustness of Support Vector Machine-Based Classification Kampouraki, Argyro Nikou, Christophoros Manis, George 09:30-09:45 Intelligent Arrhythmia Detection and Classification Azemi, Asad Sabzevari, Vahid Reza Khademi, Morteza Gholizade, Hossein Kiani, Arman Dastgheib, zeinab

of Heart Rate Signals, pp. 2159-2162 Univ. of loannina Univ. of loannina Univ. of loannina

Using ICA, pp. 2163-2166 Penn State Univ. Islamic Azad Univ. of Mashhad Ferdowsi Univ. of Mashhad ferdowsi Ferdowsi Univ. of Mashhad, Iran Ferdowsi Univ.

09:45-10:00 Adaptive Neuro-Fuzzy Inference System for Analysis of Doppler Signals, pp. 2167-2170 Ubeyli, Elif Derya TOBB Ec. and Tech. Univ.

0:00-10:15 :ffect of Feature and Channel Selection on EEG Classification, AI-Ani, Ahmed AI-Sukker, Akram

pp. 2171-2174 Univ. of Tech. Sydney Univ. of Tech. Sydney

0:15-10:30 ~educing the Number of Channels for an Ambulatory Patient-Specific EEG-Based Epileptic seizure Detector by Applying Recursive Feature Elimination, pp. 2175-2178 Glassman, Elena Guttag, John

MIT MIT

0:15-10:30 )etection of Bursts in the EEG of Post Asphyctic Newborns, pp. 2179-2182 Lofhede, Johan Univ. Call. of Bods Univ. Call. of Boris Lofgren, Nils Thordstein, Magnus Unit of Clinical Neurophysiology, Sahlgrenska Univ. Hospital Flisberg, Anders Goteborg Univ. Kjellmer, Ingemar Goteborg Univ. Lindecrantz, Kaj Univ. Coil. of Bods 10:15-10:30 ';reating a Nonparametric Brain-Computer 'rediction Preprocessing, pp. 2183-2186 Coyle, Damien McGinnity, T. Martin Prasad, Girijesh

Interface with Neural Time-Series

10:15-10:30 :arly Driver Fatigue Detection from Electroencephalography Veural Networks, pp. 2187-2190 King, Leslie Nguyen, Hung Lal, Sara 10:15-10:30 'Jnspoken Vowel Recognition PoosapadiA~unan, Sridhar Kant Kumar, Dinesh Yau, Wai Chee Weghorn, Hans

Using Facial Electromyogram,

Univ. of Ulster Univ. of Ulster Univ. of Ulster Signals Using Artificial Univ. of Tech. Sydney Univ. of Tech. Sydney Univ. of Tech. Sydney pp. 2191-2194 RMIT Univ. RMIT Univ. RMIT Univ. BA-Univ. of cooperative education

10:15-10:30 Patients on Weaning Trials from Mechanical Ventilation Classified with Neural Networks and Feature Selection, pp. 2195-2198 Giraldo, Beatriz Tech. Univ. of Catalonia (UPC) Arizmendi, Carlos Tech. Univ. of Catalonia (UPC) Romero, Enrique Tech. Univ. of Catalonia (UPC) Tech. Univ. of Catalonia (UPC) Alquezar, Rene Tech. Univ. of Catalonia (UPC) Caminal, Pere Benito, Salvador Hospital de la Santa Creu i Sant Pau Hospital Univ. de Getafe Ballesteros, D 10:15-10:30 Speech Sound Classification and Detection of Articulation Machines and Wavelets, pp. 2199-2202 Georgoulas, George Georgopoulos, Voula Stylios, Chrysostomos

Disorders with Support Vector Univ. of Patras Tech. Educal Inst. of Patras Univ. of Patras

10:15-10:30 The Effect of Electrode Displacements on Pattern Recognition Based Myoelectric Control, pp. 2203-2206 Hargrove, Levi Univ. of New Brunswick Englehart, Kevin Univ. of New Brunswick Hudgins, Bernard Univ. of New Brunswick 10:15-10:30 Novel Method of Using Dynamic Electrical Impedance Signals for Noninvasive Diagnosis of Knee Osteoarthritis, pp. 2207-2210 SGGSIE& T, Nanded and CBME, Indian Inst. of Tech. Gajre, Suhas CBME, Indian Inst. of Tech. Anand,Sneh PMR, All India Inst. of Medical Sciences New Singh, U CBME, Indian Inst. of Tech. Saxena, Rajendra FrA04 Micro Robotics

Delhi Delhi Delhi Delhi

Ziegfeld (Oral Session)

09:00-09:15 l .euter Locomotive Park, Sukho Hyunjun, Park Sungjin, Park Jee, Changyeol Kim, Jinseok Kim, Byungkyu

Microrobot for Gastrointestinal

Tract, pp. 2211-2214 Korea Inst. of Science & Tech. KIST KIST KIST Korea Inst. of Science and Tech. Korea Inst. of Science & Tech.

09: 15-09:30 Towards Active Capsular Endoscopy: Preliminary Results on a Legged Platform, pp. 2215-2218 Menciassi, Arianna Scuola Superiore Sant'Anna Stefanini, Cesare Scuola Superiore Sant'Anna Orlandi, Giovanni Scuola Superiore Sant'Anna Quirini, Marco Scuola Superiore Sant'Anna Dario, Paolo Scuola Superiore Sant'Anna 09:30-09:45 Robust Contact Detection in Micromanipulation Wang, Wenhui . 'u, Xinyu .sun, Yu FrA05 Monitoring

Using Computer Vision Microscopy, pp. 2219-2222 Univ. of Toronto Univ. of Toronto Unversity of Toronto Duffy-Columbia

Sensors and Systems I (Oral Session)

09:00-09: 15 Applications of a Textile-Based Wearable System for Vital Signs Monitoring, pp. 2223-2226 Di Rienzo, Marco Fondazione Don Carlo Rizzo, Francesco Fondazione Don C. Gnocchi Meriggi, Paolo Fondazione Don Bordoni, Bruno Fondazione Don Gnocchi Brambilla, Gabriella Fondazione Don C. Gnocchi, Ferratini, Maurizio Fondazione Don C. Gnocchi, Castiglioni, Paolo Fondazione Don C. Gnocchi

Gnocchi ONLUS Gnocchi ONLUS ONLUS ONLUS ONLUS

09: 15-09:30 Measurement of Human Gastric Motility by Near-Infrared Light for the Assessment of Chronic Mental Stress, pp. 2227-2230 Teppei, Hayashi Ritsumeikan Univ. Shiozawa, Naruhiro Ritsumeikan Univ. Makikawa, Masaaki Ritsumeikan Univ.

iaD15 lew Techniques

O'Neill in Sensory Rehabilitation

(Oral Session)

5:45-16:00 mproved Head Direction Command Classification leural Network, pp. 5679-5682 Nguyen, Son Nguyen, Hung Taylor, Philip Middleton, James 6:00-16:15 )ivergence Dynamic Modification Alvarez, Tara Gayed, Bassem

Using an Optimised Bayesian Univ. of Tech. Univ. of Tech. Univ. of Tech. The Univ. of

Sydney Sydney Sydney Sydney

As a Function of Initial Position, pp. 5683-5686 New Jersey Inst. of Tech. New Jersey Inst. of Tech.

6: 15-16:30 '"he Transient Component of Disparity Vergence Maybe an Indication of Progressive .ens Acceptability, pp. 5687-5690 Castillo, Carlos New Jersey Inst. of Tech. Gayed, Bassem New Jersey Inst. of Tech. Pedrono, Claude Essilor International Ciuffreda, Kenneth SUNY - School of Optometry Semmlow, John Rutgers Univ. Alvarez, Tara New Jersey Inst. of Tech. 16:30-16:45 :ffect of Learning on Listening to Ultra-Fast Synthesized Speech, pp. 5691-5694 Nishimoto, Takuya The Univ. of Tokyo Sako, Shinji The Univ. of Tokyo Sagayama, Shigeki The Univ. of Tokyo Ohshima, Kazue Tokyo Woman's Christian Univ. Oda, Koichi Tokyo Woman's Christian Univ. Watanabe, Takayuki Tokyo Woman's Christian Univ. 16:45-17:00 Should Object Function Matter During Modeling of Reach- To-Grasp Selfcare Tasks in ~obot-Assisted Therapy?, pp. 5695-5698 Marquette Univ. Nathan, Dominic Medical Coli. of Wisconsin/Marquette Univ. Johnson, Michelle 17:00-17: 15 J.fybrid Feature Subset Selection for the Quantitative Assessment of Skills of Stroke Patients in t1.ctivity of Daily Living Tasks, pp. 5699-5703 Katholieke Univ. Leuven Van Dijck, Gert K.U. Leuven Van Hulle, Marc Center for Multidisciplinary Approach and Tech. Van Vaerenbergh, Jo SaEP1 Principal Component

Westside Ballroom South and Independent

Component

Analysis

(Poster Session)

16:30-18:00 independent Component Analysis of Parameterized ECG Signals, pp. 5704-5707 Tanskanen, Jarno Mika Antero Tampere Univ. of Tech. Viik, Jari Tampere Univ. of Tech. Hyttinen, Jari Tampere Univ. of Tech.

16:30-18:00 The Preliminary Study on the Clinical Application (Wearable Heart Activity Monitor), pp. 6033-6036 Shin, Kunsoo Kim, Youn Ho Kim, Jong Pal Park, Jae Chan

of the WHAM Samsung Sam sung Samsung Samsung

Advanced Advanced Advanced Advanced

Inst. Inst. Inst. Inst.

of of of of

Tech. Tech. Tech. Tech.

16:30-18:00 Application-Oriented Programming Model for Sensor Networks Embedded in the Human Body, pp. 6037-6040 Catholic Univ. of Golas Barbosa, Talles M. G. de A. Catholic Univ. of Golas Sene Jr, Iwens G. Univ. of Brasilia da Rocha, Adson F. Univ. of Brasilia Assis de Oliveira Nascimento, Francisco Univ. of Brasilia Carvalho, Hervaldo Sampaio Univ. de Brasilia Camapum, Juliana Fernandes " ~0-18:00 Development of a Quantitative In-Shoe Measurement System for Assessing Balance: Sixteen-Sensor Insoles, pp. 6041-6044 Bamberg, Stacy J Morris Dibble, Lee LaStayo, Paul Musselman, Josh Dasa Raghavendra, Swarna Kiran 16:30-18:00 Reduction of Skin Stretch Induced Motion Artifacts in Electrocardiogram Adaptive Filtering, pp. 6045-6048 Liu, Van Pecht, Michael

Univ. Univ. Univ. Univ. Univ.

of of of of of

Utah Utah Utah Utah Utah

Monitoring Using Univ. of Maryland, Coil. Park Univ. of Maryland, Coil. Park

16:30-18:00 An Evaluation of a PTT-Based Method for Noninvasive and Cuff less Estimation of Arterial Blood Pressure, pp. 6049-6052 Teng, Xiaofei The Chinese Univ. of Hong Kong ""'hang, Yuan-Ting The Chinese Univ. of Hong Kong SaEP10 Diagnostic Instrumentation

Westside Ballroom South and Physiological

16:30-18:00 Neural-Network Detection of Hypoglycemic Physiological Parameters, pp. 6053-6056 Nguyen, Hung Ghevondian, Nejhdeh Jones, Timothy

Measurements 2 (Poster Session)

Episodes in Children with Type 1Diabetes Using Univ. of Tech. Sydney AIMedics Pty Ltd Princess Margaret Hospital for Children

16:30-18:00 Which Techniques to Improve the Early Detection and Prevention of Pressure Ulcers?, pp. 6057-6060 GEHIN, Claudine INSA Lyon BRUSSEAU, Elisabeth UMR CNRS5515 Meffre, Richard INSA Lyon SCHMITT, Michael INSA Lyon DEPREZ, Jean-Francois UMR CNRS5515 INSERM U630 DITTMAR, Andre INSA Lyon

6:30-18:00

'wo-chennet Bioimpedance Monitor for Impedance Cardiography, pp. 6061-6063 Vondra, Vlastimil Inst. of Scientific Inst. Czech Acad. of Halamek, Josef Inst. of Scientific Inst. Acad. of Sciences ofthe Czech Republic Viscor, Ivo Inst. of Scientific Inst. Czech Acad. of Sciences Jurak, Pavel Inst. of Scientific Inst. Acad. of Sciencesofthe Czech Republic 16:30-18:00 =xtraction of the Intracranial Component from the Rheoencephalographic Signal: !\ New Approach, pp. 6064-6067 Perez, Juan J Pol. Univ. of Valencia Guijarro, Enrique Pol. Univ. of Valencia Sancho, Jeronimo Hospital General Univ. Navarre, Arantxa Hospital General Univ. 16:30-18:00 Design of Rapid Medical Evacuation System for Trauma Patients Resulting from Biological and Chemical Terrorist Attacks, pp. 6068-6071 Frieder, Russell Cal-Zark, LLC Kumaresan, Srirangam Cal-Zark, LLC Sances, Jr., Anthony Cal-Zark, LLC Renfroe, David Cal-Zark, LLC Myers, Will J. Cal-Zark, LLC Harvey, L. William Cal-Zark, LLC 16:30-18:00 The Simulation of Current Generator Design for Multi-Frequency Electrical Impedance Tomograph, pp. 6072-6075 Chen, Cheng-Yu National Cheng Kung Univ. Lu, Yi-Yu National Cheng Kung Univ. Huang, Wen-Lung National Cheng Kung Univ. Cheng, Kuo-Sheng National Cheng Kung Univ. 16:30-18:00 A New Device for Assessing Subjective Haptic Vertical, pp. 6076-6079 Cavalieri, Matteo Reni, Gianluigi Brambilla, Daniele

IRCCS E.MEDEA IRCCS E.MEDEA IRCCS E.MEDEA

16:30-18:00 Discrete Laplacian Recordings of the Electroenterogram from Abdominal Surface in Humans, pp. 6080-6083 Prats-Boluda, Gema Centro de investigaci6n e Innovaci6n en Bioingenieria Martinez-de-Juan, Jose Luis Centro de investigaci6n e Innovaci6n en Bioingenieria Garcia-Casado, Javier Univ. Pol. de Valencia Guardiola, Jose Luis Univ. Pol. de Valencia Ponce, Jose Luis Hospital Univ. La Fe 16:30-18:00 Evaluation of Spasticity in the Ankle of a Hemiplegic Subject Using a Step-Like Response, pp. 6084-6087 Murayama, Daisuke Faculty of science and Tech. Uchiyama, Takanori Keio Univ. Uchida, Ryusei Kanto-Rosai Hospital 16:30-18:00 Sensitivity of Rheoencephalographic Guijarro, Enrique Perez, Juan J Berjano, Enrique Ortiz, Pedro

Measurements to Spatial Brain Electrical Conductivity, pp. 6088-6091 Pol. Univ. of Valencia Pol. Univ. of Valencia Pol. Univ. of Valencia Hospital General Univ.

16:30-18:00 Heart Rate Detection in Low Amplitude Non-Invasive Fetal ECG Recordings, pp. 6092-6094 Peters, Christiaan Hendrik Leonard Maxima Medical Center Vullings, Rik Eindhoven Univ. of Tech. Bergmans, Johannes Wilhelmus Maria Eindhoven Univ. of Tech. Oei, S. Guid Maxima Medisch Centrum, Veld hoven Wijn, Pieter Maxima Medical Center SaEP11 Intelligent

Westside Ballroom South Monitoring

and Analysis

Systems (Poster Session)

16:30-18:00 A Multi-Facets Analysis of the Driver Status by EEG and Fuzzy Hardware Processing, pp. 6095-6100 Faro, Alberto Univ. di Catania Giordano, Daniela Univ. di Catania Spampinato, Concetto Univ. di Catania 16:30-18:00 'lanced Insulin Transport through Skin by High-Voltage Pulse* Zhu, Bing Bao, Jiali Yang, Yuanyuan hou, haifeng

Coil. of medicine science,Zhejiang Zhejiang Coil. of Medicine Science, Zhejiang Coli. of Medicine Science, Zhejiang

Univ. Univ. Univ. Univ.

16:30-18:00 Differentiation of Epileptic Seizures from Voluntary Movements by Quantitative Analysis of Digital Video* Karayiannis, Nicolaos Univ. of Houston Modur, Pradeep Univ. of Louisville Xiong, Yaohua Univ. of Houston Tao, Guozhi Univ. of Houston 16:30-18:00 Embedded Assessment Algorithms within Home-Based Cognitive Computer Game Exercises for Elders, pp. 6101-6104 Oregon Health & Science Univ. Jimison, Holly Oregon Health and Science Univ. Pavel, Michael 16:30-18:00 'twere Enhanced Learning of Cardiac Auscultation, Syed, Zeeshan Curtis, Dorothy Nesta, Francesca Levine, Robert A. Guttag, John SaEP12 Healthcare

pp. 6105-6108 Massachusetts Inst. Massachusetts Inst. Massachusetts General Massachusetts General Massachusetts Inst.

of Tech. of Tech. Hospital Hospital of Tech.

Westside Ballroom South Decision

Support Systems (Poster Session)

16:30-18:00 Risk Factors for Apgar Score Using Artificial Neural Networks, pp. 6109-6112 Ibrahim, Doaa Ottawa Univ. Frize, Monique Carleton Univ. Walker, Robin Children Hospital of Eastern Ontario 16:30-18:00 Probabilistic Framework for Reliability Analysis of Information- Theoretic CAD Systems in Mammography, pp. 6113-6116 Habas, Piotr A. Univ. of Louisville Zurada, Jacek M. Univ. of Louisville Elmaghraby, Adel S. Univ. of Louisville Tourassi, Georgia D. Duke Univ. Medical Center