Sunday 8 June 2014


Mapping the Brain Connectome





Montreal Neurological Institute 
McGill University, Biomedical Engineering




OVERVIEW: The study of macroscopic neural connectivity using neuroimaging has exploded in recent years, with applications in many areas of clinical and basic neuroscience.  These approaches yield metrics of information flow across a network that are not accessible with focal metrics such as functional activation, metabolism or anatomical morphometry. However, there remain fundamental issues, both technical and conceptual, in reducing connectivity information from different imaging techniques into a holistic model of neural connectivity.  We will discuss different forms of connectivity, as defined by structural and functional correlation (MRI, fMRI, PET) and DTI tractography, with illustrations in normal and disordered brain.
READINGS:
    He, Y., & Evans, A. (2010). Graph theoretical modeling of brain connectivityCurrent opinion in neurology, 23(4), 341-350.
    Bullmore, E. T., & Bassett, D. S. (2011). Brain graphs: graphical models of the human brain connectomeAnnual review of clinical psychology, 7, 113-140.
    Sporns, O., Tononi, G., & Kötter, R. (2005). The human connectome: a structural description of the human brainPLoS computational biology, 1(4), e42.


27 comments:

  1. So are the male-female differences in brain connectivity because of average size differences also the same for bigger and smaller people if the same gender?

    ReplyDelete
  2. My question for Professor Evans is related to the kind of neural computational vehicle on which connectome analysis is centered. Does focusing on neuronal activation, and thus on neuronal spiking rates, displace emphasis on others kinds of information processing ongoing in the brain, carried out by non-spiking vehicles, e.g. hormones, electrical charge in dendrites, etc.? Are there plans to integrate an analysis of these non-spiking vehicles and their effects on cognition in the various projects the map the human connectome? How can these factors be represented in a graph-theoretic model?

    ReplyDelete
  3. Avec tous ces études sur le fonctionnement du cerveau, je voudrais connaître le but ultime de cette recherche. Que pouvons-nous faire avec cette information?

    Disons que nous maîtrisons la compréhension du cerveau et que nous réussissons à le reproduire sur un ordinateur (après plusieurs années), qu'est-ce que cela nous donnera? Est-ce qu'on aurait des robots qui sont aussi intelligent (voire même plus) qu'un humain tel qu'on ne pourrait pas faire la différence entre les deux? Est-ce que le but de la recherche?

    ReplyDelete
  4. I was thinking of relations between connectomics and web science. It seems that the only one is related to network analytic methods / graph theory that can be used to try understand both. But I am not sure this is strong enough relation...

    ReplyDelete
    Replies
    1. What about potential similarities between our brain and a "global brain" in terms of organizational properties and functionality?

      Delete
    2. I think both the Web and the human brain are very complex things that we can analyze with the same advanced tools (graph theory). However, most other similarities between the Web and the brain (and the Global Brain) probably arise from analyzing them in the same manner and are more metaphorical than realistic.

      Delete
  5. J'ai touché à un peu de théorie sur les réseaux de neurones dans le cadre d'apprentissage machine. Les notions du cerveau humain apportées par M.Evans m'ont fait beaucoup pensé aux réseaux de neurones en apprentissage machine.

    Est-ce que l'implémentation de ces réseaux des neurones correspondent au cerveau humain? Ou bien les réseaux de neurones représentent une infime partie (négligeable) du cerveau humain?

    En général, en quoi ils diffèrent?

    ReplyDelete
  6. With the connectomics , we can imagine controlling the web’s growth in the same way that is done in our brain, for more efficient.

    ReplyDelete
  7. Really classy talk. I was wondering is there any connectome data set/data set projects with socio-demographic data? Or better, questionnaires? I think there's a case to make for integrating this kind of data.

    ReplyDelete
    Replies
    1. This is certainly happening in future initiatives. For example, an upcoming Alzheimer's disease neuroimaging initiative in Canada is interested in demographic and lifestyle considerations.

      Delete
    2. There is one for ADHD: http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3433679/

      Delete
  8. Professor Evans, Do you pretend to have a web site about your studies on 4D model of brain connectivity? Thank you.

    ReplyDelete
    Replies
    1. [French "prétend" does not mean "pretend" in English: the meaning of the above is clearer of you delete "pretend to".]

      Delete
  9. It was interesting to hear about the Big Brain, CBRAIN, and stereotaxic space contributing to the omniscient state Heylighen described the global brain as having.

    It seems like a substantial issue assembling the connectomes from different modalities and discovering information at the mind. We discussed this issue in Professor Harnad's class and I think I still lean toward Fodor's opinion on neuroimaging: http://www.lrb.co.uk/v21/n19/jerry-fodor/diary

    ReplyDelete
    Replies
    1. What do you think Jerry Fodor would say about data-driven methods? Does his criticism of a scientist without a hypothesis and only a camera still detract from the information that can, perhaps only, be gleaned from these methods? I'd advocate being open to allowing a picture to emerge from the data rather than clinging too rigidly to hypotheses derived from previous thinking.

      However, whether this teaches us anything about the mind rather than merely elucidating the organizational properties of the brain depends on your position regarding the importance of brain research to cognitive science. I'm inclined to think it is important, but maybe less important than most people think and a little more important than Professor Harnad's class would have us believe. Most connectome initiatives have been more faithful to the research programme of neuroscience, neurology and psychiatry, though, which even Fodor would admit understanding brain properties are of value to.

      Delete
    2. This brings into question the role of science in our society. Science is the pursuit of knowledge. But, when tax-payer funded, it should have some obvious application.

      It appears to me that a lot of neuroimaging is done to simply map the brain. This has obvious scientific value, but limited immediate application. Do you think researchers mapping the brain are more concerned with future applications, or are they happy with solely expanding human knowledge.

      Delete
  10. Is there any research correlating intelligence level (IQ) or other specific competences with metrics extracted from connectomics data sets?

    ReplyDelete
  11. While Dr. Evans research and databases are very interesting and impressive, the immediate therapeutic application of this knowledge seems limited. If we discover that on average the brains of people with autism are different in some aspect than non-autistics, where does this leave us? Can we change their brains now, or is it too late? I sometimes feel that neuroimaging is leading psychiatry towards treating brains rather than treating people.

    One of the first collaborative neuroimaging sharing projects (the ADHD-200 competition) was concerned with ADHD. Researchers shared about 1000 brain scans from children with and children without ADHD. These neuroimages were labelled as ‘typically developing’, ADHD-inattentive, or ADHD-hyperactive. Next, researchers shared 500 or so more scans, some with and some without ADHD. However, this time they were not labelled. Research groups then devised algorithms to predict whether the children in the last set of brain images had ADHD, and if so what type. Performing at chance would results in 33% correct. Across research groups, the average correct diagnoses was 50%. Thus, at least for ADHD, while group differences may be present, we cannot apply these group level statistics to individual subjects.

    ReplyDelete
    Replies
    1. Maybe the burden lies also in the construct and diagnosis of ADHD rather than neuroimaging alone. Comparing Alzheimer's dementia, Autism and ADHD in terms of diagnostic specificity and sensitivity is already a bad idea before throwing in neuroimaging. It's possible that the characteristics captured by neuroimaging are better suited to identifying certain psychiatric maladies than others.

      Here's an interesting work with "Autism Spectrum Disorder" demonstrating 90% identification accuracy. http://www.pnas.org/content/early/2013/01/09/1214533110.abstract

      Delete
    2. Good points. We must treat each disorder indivudually.

      However, the paper you reference applied the classification methods on the same participants which they developed the classification methods (unlike the ADHD-200 consortium).

      Perhaps they would not reach 90% if they were give 40 scans of new people with and without ASD and ran the same classification algorithm.

      Delete
    3. In the supplementary information, the authors state that "To assess the validity of the procedure, the data were split randomly 1,000 times into a training set (75% of participants) and a test set (25% of participants)". As you pointed out, it's bad practice to test your classifier on the same data it was trained on.

      Delete
    4. Always the supplementary data! Thanks for sharing that Ishan.

      Impressive results

      Delete
  12. I thought it was interesting when Dr. Evans mentioned that increased brain size early in life can lead to a decrease in network efficiency (and perhaps be the cause for autism). In a global mind, there would similarly be physical factors which limit network efficiency, and would influence the type of mind that emerges. A global mind would have intercontinental connections, and perhaps a limiting factor in that network would be the conduction speed of signals at these large distances. Maybe this would mean that connections in the global mind would be concentrated in local hubs, with a fewer number of long range connections to link these hubs together.

    ReplyDelete
  13. J'ai trouvé cette présentation très intéressante, mais elle m'a laissé sur ma faim. Je m'attendais à découvrir des théories ou du moins quelques propositions concernant un modèle possible de la connectivité du cerveau. Cependant tous les variabilités présentées par Alan Evans me font me demander s'il ne faut pas abandonner l'idée d'un modèle unique et proposer plutôt plusieurs modèles?

    ReplyDelete
  14. A lot of research use graph theory and networks modelling to explain how related are brain parts, things, communities, ideas, etc. Then, the (semantic, functional, space) distance between two nodes is a important metric. But, we can observe also the phenomenon of circonvolution in the development of the brain of many animals as a solution to a space constraint. So, how is it possible to model mathematically and computationaly with the graph theory, the topology of nodes in a space that consider the physical dimension of circonvolutions as a solution to solve the ratio similarity/complexity, and that consider the origin of this problem? Maybe answering to these questions would help to manage limits of ours computational resources and complexity of algorithms. I’m sure somebody already ask this question...

    ReplyDelete
  15. Are there any agreements and protocols to make all brain research data openly available? Are there standards of storing and representing data?

    ReplyDelete
  16. I think this has fascinating implications for the web and the global brain in that, if we better understand the substrate on/from which the mind operates, then perhaps we will be better capable of designing machines that can replicate and improve on this biological substrate and in effect have mind-like properties currently in possession of agents. Of course, the neurons themselves are extremely important, and individual nodes on a network such as the web do not function in the same way, but would it not be possible to create a neuron emulation program, do we not already have this technology? With respect to what this tells us about consciousness and the human brain/mind, I am quite concerned about plasticity and individual variation, however. While it may be somewhat useful to attach concepts like "self-referential processing" to a location in the brain, at what resolution do our conceptualizations fail? I think the structural-fuctional gap will be the biggest hurdle.

    ReplyDelete