Machine Learning SIG

From MEG Core
Jump to navigation Jump to search

Objectives

Advance general knowledge of machine learning techniques within the MEG community. Discuss journal articles, replicate techniques on NIH data, develop new ML techniques at NIH.

Format

  1. Specific Projects (Weekly)
    1. Code Review
    2. Project updates
    3. Question and Answer Clinic for users
  2. General (Monthly)
    1. Journal Club
    2. Hackathons – implement novel technique from JC with provided data
    3. Tutorial Workshops - instruct worked out examples with provided code/data
    4. General ML training
      1. Parameter tuning
      2. Model optimization
      3. Techniques
      4. Toolbox tutorials (Scikit-learn / keras)

Analysis Types

  1. Decoding[1]
    1. Multivariate time series classification between conditions [2][3]
      1. Filtering Issues[4]
    2. Temporal Generalization [5]
    3. Realtime - Brain computer interface / neurofeedback
  2. Subject classification
    1. eg. Healthy Control vs Major Depressive Disorder
    2. What are the significant features (brain regions, Hz)
    3. Automated diagnosis of TBI from MEG low frequency signals [6]
  3. Prediction of future condition / Biomarkers
    1. What signals predict conversion from mild cognitive impairment to Alzheimer’s Disease [7] [8]
    2. What signals predict recovery from traumatic brain injury
    3. What signals predict poor outcome for epilepsy surgery
  4. Multimodal Integration [9]
    1. Timing derived from MEG / localization from fMRI [10]
    2. Signal comparison between naturalistic viewing data MEG/fMRI
    3. Combining DTI and MEG [11]
    4. Multimodal imaging classification [12]
  5. Signal classification
    1. Artifact [13] [14]
    2. Signal of interest – Spike detection for epilepsy [15] [16] [17]
    3. Signal/sequence predicts correct response
  6. Temporal learning models – RNN DL / markov model [18]
  7. Inferences from deep learning models
    1. CNN on images vs. CNN of brain responses to same images [19][20]

Preliminary Resources

General

SciKit Learn
Keras

MEG

MNE Python Decoding MNE Python Decoding at Source
RealTime MEG
Deep Learning Decoding

NIH

NIMH Machine Learning in Neuroimaging
NIH-AI
Biowulf Deep Learning Course
Biowulf DeepLearning Tools

Relevant Papers

Language Decoding

  1. [1] Deconstructing multivariate decoding for the study of brain function: Martin N.Hebart, Chris I.Baker
  2. [2] Encoding and Decoding Neuronal Dynamics: Methodological Framework to Uncover the Algorithms of Cognition
  3. [3] Decoding Dynamic Brain Patterns from Evoked Responses: A Tutorial on Multivariate Pattern Analysis Applied to Time Series Neuroimaging Data
  4. [4] High-pass filtering artifacts in multivariate classification of neural time series data
  5. Characterizing the dynamics of mental representations: the temporal generalization method - J-R.King, S.Dehaene
  6. [5] An automatic MEG low-frequency source imaging approach for detecting injuries in mild and moderate TBI patients with blast and non-blast causes
  7. [6] Aberrant MEG multi-frequency phase temporal synchronization predictsconversion from mild cognitive impairment-to-Alzheimer's disease: Sandra Pusil, Stavros I. Dimitriadis, María Eugenia López, Ernesto Pereda, Fernando Maestú
  8. [7] Deep-MEG: spatiotemporal CNN features and multiband ensemble classification for predicting the early signs of Alzheimer’s disease with magnetoencephalography: Antonio Giovannetti, Gianluca Susi, Paola Casti, Arianna Mencattini, Sandra Pusil, María Eugenia López, Corrado Di Natale & Eugenio Martinelli
  9. [8] A M/EEG-fMRI Fusion Primer: Resolving Human Brain Responses in Space and Time: Radoslaw M.Cichy Aude Oliva
  10. [9] Resolving human object recognition in space and time: Radoslaw Martin Cichy, Dimitrios Pantazis & Aude Oliva Nature Neuroscience volume 17, pages455–462 (2014)
  11. [10] Using joint ICA to link function and structure using MEG and DTI in schizophrenia J.M.Stephen, B.A.Coffman, R.E.Jung, J.R.Bustillo, C.J.Aine, V.D.Calhoun
  12. [11] Multimodal Classification of Schizophrenia Patients with MEG and fMRI Data Using Static and Dynamic Connectivity Measures: Mustafa S. Cetin, Jon M. Houck, Barnaly Rashid, Oktay Agacoglu, Julia M. Stephen, Jing Sui, Jose Canive, Andy Mayer, Cheryl Aine, Juan R. Bustillo and Vince D. Calhoun
  13. [12] MEGnet: Automatic ICA-based artifact removal for MEG using spatiotemporal convolutional neural networks Alex H.Treacher et al.
  14. [13] ICLabel: An automated electroencephalographic independent component classifier, dataset, and website: Luca Pion-Tonachini, Ken Kreutz-Delgado, Scott Makeig
  15. [14] EMS-Net: A Deep Learning Method for Autodetecting Epileptic Magnetoencephalography Spikes: Li Zheng, Pan Liao, Shen Luo, Jingwei Sheng, Pengfei Teng, Guoming Luan, Jia-Hong Gao
  16. [15] Automated Detection of Epileptic Biomarkers in Resting-State Interictal MEG Data: Miguel C. Soriano, Guiomar Niso, Jillian Clements, Silvia Ortín, Sira Carrasco, María Gudín, Claudio R. Mirasso and Ernesto Pereda
  17. [16] A Long Short-Term Memory neural network for the detection of epileptiform spikes and high frequency oscillations: A. V. Medvedev, G. I. Agoureeva & A. M. Murro
  18. [17] Task-Evoked Dynamic Network Analysis Through Hidden Markov Modeling: Andrew J. Quinn, Diego Vidaurre, Romesh Abeysuriya, Robert Becker,Anna C. Nobre1, and Mark W. Woolrich
  19. [18] Comparison of deep neural networks to spatio-temporal cortical dynamics of human visual object recognition reveals hierarchical correspondence: Radoslaw Martin Cichy 1,2 , Aditya Khosla 1 , Dimitrios Pantazis 3 , Antonio Torralba 1 & Aude Oliva 1
  20. [19] Performance-optimized hierarchical models only partially predict neural responses during perceptual decision making: Laura Gwilliams, Jean-Rémi King