SFE PRESENTATION

IMT AtlantiqueCIBR Okubo LabZiqiao WANG

data science on neuro-behavioral data

Related research

Project development

Mission

Content

Working enviroment

Methodology of neural science and data science

Evaluation

Modeling

Data

Project

Background

Conclusion

Solution

Problem

Working enviroment

Methodology of neural science and data science

Background

CIBR, Chinese institute for brain research

Okubo lab

Machine learningEngineer

Post-doc

Brain research

Artificial intelligence

Materials research

Biological experiments

...

Data scientist

Working enviroment

Data science methodology

Neuro-science methodology

Methodology

Evaluation

Modeling

Data

Project

Solution

Problem

Every body movement, from raising a hand to smiling, involves a complex interaction between the central nervous system (brain and spinal cord), nerves, and muscles. Damage to or malfunction of any of these components may result in a movement disorder.

Every year, around the world, between 250 000 and 500 000 people suffer a spinal cord injury (SCI). -- WHO

https://medicine.umich.edu/sites/default/files/content/downloads/NSCISC%20SCI%20Facts%20and%20Figures%202021.pdf

Problem

Muscle effector signals

Brain electrical signals

Problem

+ : 1) More accurate 2) More felxible

Non-invasive Exoskeleton

https://www.nature.com/articles/s41586-023-06094-5 -- CHUV 2022/WEBER GILLES

How to restore the walking ability of SCI patients through BMI technology ？

https://unwire.hk/2019/10/06/mind-controlled-exoskeleton-helped-a-man-with-paralysis-walk-again/fun-tech/

Invasive Implants

BMI applications

Health

...

Education

Game andentertainment

Brain machine interface

Short term objective

Long term objective

https://www.nature.com/articles/s41586-023-06094-5

...

10

Solution

Data flow

Muscle stimulation

Input ECoG signals

Movement intention

Encoding

Decoding

Motion capture system

Movement intention

Input ECoG signals

11

EEG : electroencephalography (EEG) ECoG : electrocorticography (ECoG)

Data understanding

mice_0604-2023_0704-AM

09h0009h0309h05......

(time)

(Animal_id, date, slot)

Session

Trials

Data terminology

Data set

12

Data understanding

Need : work with other labs to prepare dataDifficulties :1) Presented as code : hard to understand for collaborator2) Presented as figures : lack of flexibility and efficiencySolution : develop an app could generate figures by choosing params

Time (s)

Streamlit data viewer

13

Statistical analysis

Visualize channel impedance changes over time

Choose an appropriate impedance threshold by comparing the voltage range

To remove abnormal channels

Abnormal channel analysis

Statistical analysis

Visualize the correlation between different channels of ECoG signals and behavior data

Visualize the time distribution of trials in a session

To verify the quality of data

To remove abnormal trials

Correlation analysis

14

Trial distribution analysis

Statistical analysis

0 50 100 150 200 250 300Frequency (Hz)

Common reference average

Statistical analysis

Example of real data

Wavelet transform

Envelope

Morlet wave

Signal

Morlet wavelet

+ : 1) Keep time and frequency information at the same time2) Provides variable resolution.

Wavelet transform

15

Data preparation

120Hz

1kHz

Find cloest

Label behavoir data

Time stamp M

...

Time stamp 2

Time stamp 1

Time stamp N

...

Time stamp 2

Time stamp 1

Input ECOG data

Input-label alignment

one session

Train set Test set

0 10 15 Time (min)

Train-test split

Final data

16

Data preparation

Final model

Problem : There are many features -> many coefficients -> collinearit problemOrdinary linear regression could easily overfit.

p = (number of frequency features) x (number of time features) x (number of channels)

Solution:(1) First reduce the dimension of X from p to t -- Maximum the covariance between X' and Y(2) Then do ordinary linear regression between X' and Y

X'

linear dimensionality reduction

17

Partial Least regression

Modeling

Hyper parameter tuning

18

Model accuracy evaluation

Evaluation

https://www.nature.com/articles/s41586-023-06094-5#Fig4

Model stability evaluation

https://www.nature.com/articles/s41586-023-06094-5#Fig3

Bussiness evaluation

19

Evaluation

Related research

Project development

Mission

1) Help Wu lab to prepare the CIBR Mice food tracking data set2) Develop data analysis tools to help biological scientist to check the data

Two locate methods,Operation 1 : first align the label and input data then apply wavelet transformOperation 2 : first apply wavelet transform then align the label and input wavelet features

Collaboration with other labs

1) Security management2) Front-end GUI design3) Back-end Algorithm development

Streamlit development

21

Preprocessing optimization

Project development

​Fundamental of Machine learningbased lab meeting

22

Paper based-lab meeting

Related research

Conclusion

Integrate societal issues into an environment professional

Reflect on myself, my knowledge and my experiences

Communcation

Ability used

Design and create systems and organizations

Data science skill improvementBasic knowledge enhancementLearn more about the neural science...

Personal gain

Phase 1 : Project approvalPhase 2 : Model evaluation using RIKEN datasetPhase 3 : CIBR data set preparation and analysisphase 4 : School evaluation

Time table about the internship

24

Conclusion

25

[1] Tonio Ball, Markus Kern, Isabella Mutschler, Ad Aertsen, and Andreas Schulze-Bonhage. Signal quality of simultaneously recorded invasive and non-invasive eeg. Neuroimage, 46(3) :708–716, 2009. [2] Zenas C Chao, Yasuo Nagasaka, and Naotaka Fujii. Long-term asynchronous decoding of arm motion using electrocorticographic signals in monkey. Frontiers in neuroengineering, 3 :1189, 2010. [3] Agnar Höskuldsson. Pls regression methods. Journal of chemometrics, 2(3) :211–228, 1988. [4] Henri Lorach, Andrea Galvez, Valeria Spagnolo, Felix Martel, Serpil Karakas, Nadine Intering, Molywan Vat, Olivier Faivre, Cathal Harte, Salif Komi, et al. Walking naturally after spinal cord injury using a brain–spine interface. Nature, pages 1–8, 2023. [5] John W McDonald and Cristina Sadowsky. Spinal-cord injury. The Lancet, 359(9304) : 417–425, 2002. [6] JF Soechting and M Flanders. Moving in three-dimensional space : frames of reference, vectors, and coordinate systems. Annual review of neuroscience, 15(1) :167–191, 1992. 40

Reference

Overall data preparation workflow

Appendix

26

Thank you for your attention