Embed Size (px)
Citation preview
Standardization of Brain-Computer Interfaces and Terminology: Sharing a Model to Share Methods and ToolsLUIGI BIANCHI, PH. D. ‘TOR VERGATA’ UNIVERSITY OF ROME, ITALYEMAIL: [email protected]
Typical BCI system structureMultidisciplinarity•Psicology•Engineering•Neurology•Computer Science•Information Theory•Neuroscience•Robotics•…
Classification•Cortical area recorded•Extracted features•Feedback sensory modality•Application•Input signals•…
(Static) Functional Model of a BCI
ClassifierExtraction of the
features of interest and
translation into Logical Symbols
(LSs)
Control InterfaceTranslation of LSs into Semantic Symbols (SSs) by means of encoding strategies
Application ControlEncoding of SSs into
commands towards output devices
Collection StageEEG, ECoG, fMRI, fNIRS
Mason et al.,” Ann. Biomed. Eng., 2005.Bianchi et al, “TNSRE”, 2007.
.01011. ‘B’,’C’,’I’
Characterizing the Transducer: The Extended Confusion Matrix
a b g ?
a 56 9 4 7b 2 64 3 4g 1 3 62 6
CodificaChar E1 E2 E3 E4 E5 E6 E7 Freq (Eng)
Freq (Ita)
A aaa ggg aab baa baa bag bg 6.543 9.528B aab ggb aag bbg ggag aaaga aagb 1.250 0.793C aag gga aba bga gba aabg aaba 1.963 3.743D aba gbg abb bbb bgb abb bab 3.712 2.991E abb gbb abg aab ab bg bb 10.080 9.642F abg gba aga agg gbb aabb agga 1.707 0.903G aga gag agb gba gbg aaba gaba 1.739 1.587H agb gab agg abb bbb agb aga 5.109 1.041I agg gaa baa gaa bag baa gba 5.489 8.478J baa bgg bab gbg ggga aaaag abbagb 0.123 0.018K bab bgb bag bgb ggbb aaabb gagb 0.773 0.008L bag bga bba abg bgg aba gab 3.369 5.141M bba bbg bbb bag gab aagb ggb 2.037 2.180… ... ... ... ... ... ... ... ... ...Space ggg aaa aaa aaa aa g aa 19.104 18.114Len 3.000 3.000 3.000 3.000 2.696 2.685 2.657 - -
Enc. Encoded Text a b g Len
E1 bbbggggabagbabbgggbbaaggabaababagabbgggbbgabgggggabagbabbgggbbbaggagaagbgab 20 26 29 75
E2 bbbaaaagbgabgbbaaabbggaagbggbgbgagbbaaabbagbaaaaagbgabgbbaaabbbgaagaggabagb 29 26 20 75
E3 bbgaaagagaggabgaaabbbbaaabbabbbbaabgaaabgaagaaaagagaggabgaaabbgbaaagbagggag 35 20 20 75
E4 agaaaaabaabbaabaaabaggaabbbbbbabgaabaaaaagaggaaaabaabbaabaaaagagaagbaabbaba 44 21 10 75
E5 bbaaaagbbbabaagabbagbgbbgbbggabaababgbbaaagbbbabaabbabaggbgbbbag 22 29 13 64
E6 agggbbagbbggaagbbaaabbabbababggbabaabbgbbagbbggaggbaaaabaagbbb 21 25 16 62
E7 bbaaaagagabbaaggbgbababbabgabbbaaggaggaaaagagabbaabbagbagabaagaag 30 19 16 65
Simulations
LogicalAlphabet
SemanticAlphabet
Encoder
ControlInterface
Transducer
ExtendedConfusion
MatrixSemantic
Input Sequence
Simulator
LogicalOutput
Sequence
LogicalAlphabet
SemanticAlphabet
Encoder
ControlInterface
Transducer
ExtendedConfusion
MatrixSemantic
Input Sequence
Simulator
Logical Output
Sequence
ReliableMetrics
LogicalAlphabet
Semantic Alphabet
Encoder
ControlInterface
Transducer
ExtendedConfusion
Matrix
Optimization
BetterOverallSystem
Performances
ReliableMetrics
P300 Confusion exampleu Performance evaluation
u Characters per secondu Selections per secondu Words per secondsu Bit-rateu Information Transfer Rateu Accuracyu Efficiencyu Mutual Information
u Protocol descriptionu Trial? Session? Run?
Motivation for BCI Standardsu Too much time spent to «synchronize» the terminology
among groupsu Too often impossible to compare systemsu Too much time spent to reproduce/adapt tools and
methodsu Too much effort spent to share resourcesu Virtually impossible to allow interoperability among
existing systems.u …u CAN A USER REALIZE WHICH BCI IS MORE SUITABLE FOR
HIS NEEDS?
u THIS DRAMATICALLY LIMITS PROGRESS!
IEEE-Standards Association
• standards must provide clear advantages to the whole community, including patients, manufacturers, scientists and health professionals, in terms of quality, safety and efficiency;
• they must be minimally “invasive” with existing systems. In other words, actual system should not need to be largely modified in order to made them standard compliant;
• they should not reinvent the wheel, but possibly make use of already available standards
• they must be easy to understand;
• they should not represent a limitation to the implementation of new paradigms or an obstacle to innovation;
IEEE P2731 Unified Terminology for BCI - WG
iMeet discussions
What are the areas/boundaries of P2731?
• Janis Peksa
• Zygmunt Ryznar
• Surendra Tipparaju
• Chuck Easttom
• Guillermo Sahonero Alvarez
• Jeremy Gleick
• Ali Hossaini
1.Hardware 2.Software 3.Protocol 4.Performance 5.Ethics 6.Application 7.File Format 8.Methods9.Data Transfer Protocol10.Security
Q: Is the protocol section referring to users' training procedures? A: No, I meant P300, SSVEP, C-VEP, etc..
CONCLUSION: The protocol term seems too wide.
Chuck EasttomI think it is probably beyond our bounds to standardize the format of data? Medical data has HL7 format, and many other areas have specific data formats that allow interoperability. But that seems as if such a project would be a separate standards group in and off itself.
Guillermo Sahonero AlvarezI agree. Perhaps, we could establish some aspects that data format should have instead of standardizing the format of data strictly?
What are the areas/boundaries of P2731?
Dealing with file formats
iMeet discussions
1.Healthy users 2.Patients 3.Caregivers 4.Acquaintances 5.Medical doctors 6.Neurologists 7.Psychologists 8.Engineers 9.Computer Scientists 10.Programmers 11.Manufacturers 12.Technicians 13.Ethical Committees 14.Bio engineers15.Electrical engineers16.Neuroscience researchers17.…
Chuck EasttomDavide ValerianiZygmunt RyznarMirza Ishraq YeahiaGuillermo SahoneroAlvarez Surendra TipparajuAli Hossaini
iMeet discussions
Who should benefit from P2731?
Ali Hossaini
TARGET STAKEHOLDERS 1. Users: Healthy users, patients, caregivers 2. Computer Scientists: Programmers, database designers 3. Medical Doctors 4. Researchers: Neuroscientists, psychologists, behavioral scientists 5. Engineers: Bioengineers, electrical engineers 6. Manufacturers 7. Technicians: 8. Policymakers: Ethical committees, privacy committees 9. Designers: UI / interface designers, artists, game designers 10. Marketing
iMeet discussions
Davide Valeriani
I understand the importance of capturing the different stakeholders, but we should not go into too fine-grained details.
Proposal for the creation of subgroups
1) BCI Vocabulary (glossary): start collecting available definitions: most of them are widely
accepted….. make them understandable by all stakeholders, possibly defining the
same term with different languages, such as the one of the user or the computer scientist. Some of the definitions will be paradigm specific (e.g. iteration in a P300), some others will be more general (dependent BCI or reactive BCI). With this simple output, we could also gain some visibility.
2) BCI Functional model. I think this is also very important: having a good functional model would ease the description of a BCI, especially regarding methods, setup, hardware, etc...
3) BCI Description. methods (e.g. performance evaluation), what should be stored into a file (precedes the definition of a file
format)-> P2731 compliant what should be transmitted, ….
For example, we could establish that a BCI system is P2731 compliant if some
iMeet discussions
www.draw.io
u DRAW.IO
www.draw.io
Thank you! 28
P2731Vice-chair: Carole CareySecretary: Pradeep BalacandranTom Thompson
IEEE EMBC
‘19
An example: a P3 SpellerEncoder
R1·C1 = AR1·C2 = B
...R3·C4= P
OutputWrite an email
P...
S.G. Mason et al.,” Ann. Biomed. Eng., 2005.L.R. Quitadamo et al, Neuroinformatics,, 2008.
An example: a visual ERP-based protocol
P300
