Wednesday, October 07, 2015

A few more words about good brains and bad brains

My previous Good Brain / Bad Brain post may have been a little out there, so here are four brief comments.

(1) HCP database.  The entire Human Connectome Project database (ConnectomeDB) is an amazing resource that's freely available (more details in Van Essen et al., 2013, 2015).

(2) Good reporting / bad reporting.  Smith et al. (2015) are to be commended for such an impressive body of work.1  But I still think it was remiss to report a population along a judgmental good/bad binary axis in a cursory manner. The correlation/causation conundrum needs more of a caveat than:
These analyses were driven by and report only correlations; inferring and interpreting the (presumably complex and diverse) causalities remains a challenging issue for the future.
...or else you're confronted with press coverage like this:
Are some brains wired for a lifestyle that includes education and high levels of satisfaction, while others are wired for anger, rule-breaking, and substance use?

“Wired” implies born that way no effects of living in poverty in a shitty neighborhood.

Oh, and my flippant observation about the wine cooler/malt liquor axis wasn't actually a major player in the canonical correlation analysis. But race and ethnicity information was indeed collected (but not used: “partly because the race measure is not quantitative, but consists of several distinct categories”).

(3) Ethics!  This brings up the larger issue of ethics. A whole host of personal participant information (e.g., genomics from everyone, including hundreds of identical twins) is included in the package. From Van Essen et al. (2013):
The released HCP data are not considered de-identified, insofar as certain combinations of HCP Restricted Data (available through a separate process) might allow identification of individuals as discussed below. It is accordingly important that all investigators who agree to Open Access Data Use Terms consult with their local IRB or Ethics Committee to determine whether the research needs to be approved or declared exempt. If needed and upon request, the HCP will provide a certificate stating that an investigator has accepted the HCP Open Access Data Use Terms. Because HCP participants come from families with twins and non-twin siblings, there is a risk that combinations of information about an individual (e.g., age by year; body weight and height; handedness) might lead to inadvertent identification, particularly by other family members, if these combinations were publicly released.


Important Notice to Recipients and System Administrators of HCP Connectome In A Box Hard Drives

Thank you for acquiring a Connectome-in-a-Box that contains HCP image data.  This provides an easy and efficient way to transfer large HCP datasets to other labs and institutions wanting to process lots of data, especially when multiple investigators are involved. With it comes a need to insure compliance with HCP’s Data Use Terms as well as any institutional requirements.

And any participant in the study can look at the results and infer, because of their regular cannabis use and their father's history of heavy drinking, that they must have a “bad brain.” Do the investigators have an obligation to counsel them on what this might mean (and what they should do)? Yeah, stop smoking cigarettes and pot, but there's not much they can do about their father's substance abuse or their fluid intelligence.

(4) Biology.  Finally, I'm not sure what the finding means biologically. Across a population, there's a general mode of functional connectivity while participants lie in a scanner with nothing to do. That falls along an axis of “positive” and “negative” traits. And this pattern of correlated hemodynamic activity across 30 node-pair edges means....... what, exactly?

Every person's connectome is unique (“I am my connectome” for the thousandth time).2  But this mantra more commonly refers to the fine-grained structural connectome. You know, the kind that will live forever and be uploaded to a computer (see Amy Harmon's article on The Neuroscience of Immortality, which caused quite a splash).

What is the relationship between resting state functional connectivity and the implementation of thought and behavior via neural codes? This must be exceptionally unique for each person. We know this because even in lowly organisms like flies, neurons in an olfactory region called the mushroom bodies show a striking degree of individuality in neural coding across animals.3 
At the single-cell level, we show that uniquely identifiable MBONs [mushroom body output neurons, n=34] display profoundly different tuning across different animals, but that tuning of the same neuron across the two hemispheres of an individual fly was nearly identical.

In other words, a fly's unique olfactory experience shapes the response properties of a tiny set of neurons, even for animals reared under the same conditions. “In several cases, we even recorded on the same day from progeny of the same cross, raised in the same food vial” (Hige et al., 2015).

Fig. 5c (Hige et al., 2015).

I never know what to do with information like this, especially in the context of human brains, good and bad.....  Maybe: Are some fly MBONs wired for a wild lifestyle of apple cider vinegar?


1 Or maybe the result was a massive case of confirmation bias, as suggested in a private comment to me.

2 See this book review for an opposing view.

3 fly paper via @fly_papers (also @debivort and @neuroecology).

4 Also see Neurocriminology in prohibition-era New York.

Subscribe to Post Comments [Atom]

Wednesday, September 30, 2015

Good Brain / Bad Brain

'Wiring diagrams' link lifestyle to brain function

Human Connectome Project finds surprising correlations between brain architecture and behavioural or demographic influences.

The brain’s wiring patterns can shed light on a person’s positive and negative traits, researchers report in Nature Neuroscience1. The finding, published on 28 September, is the first from the Human Connectome Project (HCP), an international effort to map active connections between neurons in different parts of the brain.

What are some of these surprising conclusions about the living human brain?

Good Brain / Bad Brain

Smith et al. (2015):
“We identified one strong mode of population co-variation: subjects were predominantly spread along a single 'positive-negative' axis linking lifestyle, demographic and psychometric measures to each other and to a specific pattern of brain connectivity.”

Well. This sounds an awful lot like the Hegemony of the Western Binary as applied to resting state functional connectivity to me...

And hey, looks like IQ, years of education, socioeconomic status, the ability to delay reward, and life satisfaction give you a good brain.

“You can distinguish people with successful traits and successful lives versus those who are not so successful,” [Marcus Raichle] says.

The authors used canonical correlation analysis (CCA) to estimate how 280 demographic and behavioral subject measures and patterns of brain connectivity co-varied in a similar way across subjects (Smith et al., 2015):
“This analysis revealed a single highly significant CCA mode that relates functional connectomes to subject measures (r = 0.87, P < 10−5 corrected for multiple comparisons across all modes estimated).”

And who is not so “successful” (at least according to their chaotic and disconnected brains)?

Regular pot smokers:  “ of the negative traits that pulled a brain farthest down the negative axis was marijuana use in recent weeks.”  Cue up additional funding for NIDA:  “...the finding emphasizes the importance of projects such as one launched by the US National Institute on Drug Abuse last week, which will follow 10,000 adolescents for 10 years to determine how marijuana and other drugs affect their brains.”

But what about wine coolers??

Why am I asking this?? Because in the subject measures, it was a little obvious that malt liquor was considered separately from beer/wine coolers. {Who drinks wine coolers? Who drinks malt liquor?}

In terms of alcohol content, the distinction is silly these days, since you can buy craft beers like Boatswain Double IPA (8.4% alcohol) for $2.29 at Trader Joe's. Unless those questions were retained as a code for race and socioeconomic status...


I'm getting way off track here. My point is that presenting correlational HCP data in a binary manner without any sort of social context isn't a very flattering thing to do.

I am my connectome,” says Sebastian Seung. What about the 460 participants in the study? What about you?


Reardon S (2015). 'Wiring diagrams' link lifestyle to brain function. Nature News. doi:10.1038/nature.2015.18442

Smith SM, Nichols TE, Vidaurre D, Winkler AM, Behrens TE, Glasser MF, Ugurbil K, Barch DM, Van Essen DC, Miller KL. (2015). A positive-negative mode of population covariation links brain connectivity, demographics and behavior. Nat Neurosci. 2015 Sep 28. doi: 10.1038/nn.4125.

“As a black woman interested in feminist movement, I am often asked whether being black is more important than being a woman; whether feminist struggle to end sexist oppression is more important than the struggle to racism or vice versa. All such questions are rooted in competitive either/or thinking, the belief that the self is formed in opposition to an other...Most people are socialized to think in terms of opposition rather than compatibility. Rather than seeing anti-racist work as totally compatible with working to end sexist oppression, they often see them as two movements competing for first place.”

bell hooks, Feminist Theory: From Margin to Center

Subscribe to Post Comments [Atom]

Saturday, September 26, 2015

Neurohackers Gone Wild!

Scene from Listening, a new neuro science fiction film by writer-director Khalil Sullins.

What are some of the goals of research in human neuroscience?
  • To explain how the mind works.
  • To unravel the mysteries of consciousness and free will.
  • To develop better treatments for mental and neurological illnesses.
  • To allow paralyzed individuals to walk again.

Brain decoding experiments that use fMRI or ECoG (direct recordings of the brain in epilepsy patients) to deduce what a person is looking at or saying or thinking have become increasingly popular as well.

They're still quite limited in scope, but any study that can invoke “mind reading” or “brain-to-brain” scenarios will attract the press like moths to a flame....

For example, here's how NeuroNews site Brain Decoder covered the latest “brain-to-brain communication” stunt and the requisite sci fi predictions:
Scientists Connect 2 Brains to Play “20 Questions”

Human brains can now be linked well enough for two people to play guessing games without speaking to each other, scientists report. The researchers hooked up several pairs of people to machines that connected their brains, allowing one to deduce what was on the other's mind.
. . .

This brain-to-brain interface technology could one day allow people to empathize or see each other's perspectives more easily by sending others concepts too difficult to explain in words, [author Andrea Stocco] said.

Mind reading! Yay! But this isn't what happened. No thoughts were decoded in the making of this paper (Stocco et al., 2015).

Instead, stimulation of visual cortex did all the “talking.” Player One looked at an LED that indicated “yes” (13 Hz flashes) or “no” (12 Hz flashes). Steady-state visual evoked potentials (a type of EEG signal very common in BCI research) varied according to flicker rate, and this binary code was transmitted to a second computer, which triggered a magnetic pulse delivered to the visual cortex of Player Two if the answer was yes. The TMS pulse in turn elicited a phosphene (a brief visual percept) that indicated yes (no phosphene indicated a “no” answer).

Eventually, we see some backpedalling in the Brain Decoder article:
Ideally, brain-to-brain interfaces would one day allow one person to think about an object, say a hammer, and another to know this, along with the hammer's shape and what the first person wanted to use it for. "That would be the ideal type of complexity of information we want to achieve," Stocco said. "We don't know whether that future is possible." 

Well, um, we already have the first half of the equation to some small degree (Naselaris et al. 2015 decoded mental images of remembered scenes)...

But the Big Prize goes to.... the decoders of covert speech, or inner thoughts!! (Martin et al. 2014)

Scientists develop a brain decoder that can hear your inner thoughts

Brain decoder can eavesdrop on your inner voice

Listening to Your Thoughts

The new film Listening starts off with a riff on this work and spins into a dark and dangerous place where no thought is private. Given the preponderance of “hearing” metaphors above, it's fitting that the title is Listening, where fiction (in this case near-future science fiction) is stranger than truth. The hazard of watching a movie that depicts your field of expertise is that you nitpick every little thing (like the scalp EEG sensors that record from individual neurons). This impulse was exacerbated by a setting which is so near-future that it's present day.

From Marilyn Monroe Neurons to Carbon Nanotubes

But there were many things I did like about Listening.1  In particular, I enjoyed the way the plot developed in the second half of the film, especially in the last 30 minutes. On the lighter side was this amusing scene of a pompous professor lecturing on the real-life finding of Marilyn Monroe neurons (Quian Quiroga et al., 2005, 2009).

Caltech Professor: “For example, the subject is asked to think about Marilyn Monroe. My study suggests not only conscious control in the hippocampus and parahippocampal cortex, when the neuron....”

Conversation between two grad students in back of class: “Hey, you hear about the new bioengineering transfer?” ...

Caltech Professor: “Mr. Thorogood, perhaps you can enlighten us all with Ryan's gossip? Or tell us what else we can conclude from this study?”

Ryan the douchy hardware guy: “We can conclude that all neurosurgeons are in love with Marilyn Monroe.”

David the thoughtful software guy: “A single neuron has not only the ability to carry complex code and abstract form but is also able to override sensory input through cognitive effort. It suggests thought is a stronger reality than the world around us.”

Caltech Professor: “Unfortunately, I think you're both correct.”

Ryan and David are grad students with Big Plans. They've set up a garage lab (with stolen computer equipment) to work on their secret EEG decoding project. Ryan the douche lets Jordan the hot bioengineering transfer into their boys' club, much to David's dismay.

Ryan: “She's assigned to Professor Hamomoto's experiment with ATP-powered cell-binding nanotube devices.” [maybe these?]

So she gets to stay in the garage. For the demonstration, Ryan sports an EEG net that looks remarkably like the ones made by EGI (shown below on the right).

Ryan reckons they'll put cell phone companies out of business with their mind reading invention, but David realizes they have a long way to go...

Jordan the hot bioengineering transfer:  “Your mind can have a dozen thoughts in a millisecond 2 [really? how can you possibly assert this?] but it takes you five seconds to say 'hi sexy'?”

Ryan the douchy hardware guy: “It's not perfect.”

Jordan: “It's crap.”


Jordan points out the decoding algorithm's response time is way too slow to be useful, and that recording from “a thousand neurons” 3 isn't enough... “you have to open the books.” David points out they're not neurosurgeons (who would implant intracranial electrodes for ECoG).

Jordan: “You don't need surgery... you need nanotubes.”

...and this leads to the most ridiculous scenario: intrathecal administration of said nanotubes [along with microscopic transistors to form molecular electrodes] via lumbar puncture (spinal injections) performed by complete novices wielding foot long needles. [direct administration into the cerebrospinal fluid bypasses difficulties with the impermeable blood brain barrier.] But if you can get through that, and the heavy handed use of color filters... will be transported to the Red Room, where scary bald men “listen” to every thought [the direct brain-to-brain communication is one way only to avoid that nasty "circular feedback loop"].

Then more THINGS happen. It's not perfect. But it's not crap. I thought Listening was worth $4.99.

Available on Amazon and Vimeo.

Sometimes even The Neurocritic is willing to suspend disbelief...

Further Reading

Brain decoding: Reading minds: 2013 Nature News story by Kerri Smith.
“By scanning blobs of brain activity, scientists may be able to decode people's thoughts, their dreams and even their intentions.”

Neuroscience: ‘I built a brain decoder': BBC Future
“What are you looking at? Scientist Jack Gallant can find out by decoding your thoughts, as Rose Eveleth discovers.”

Brain Decoding Project: mouse hippocampus
---A BRAIN Project: Brain Activity Mapping of Neural Codes for memory

One more step along the long road towards brain-to-brain interfaces: Nice blog coverage of the 20 Questions study by Pierre Mégevand.

Meet the Hackers Who Are Decrypting Your Brainwaves: Oh no they're not. But an interesting piece on the DIY EEG movement.


1 Some of the dialogue and the interpersonal relationships? Not as much.

2 Dozens of thoughts in 1/1000 of a second?? Perhaps she's being hyperbolic here... Well, popular lore says we have 70,000 thoughts per day, which comes out to only 0.8101851851851852 thoughts per second. But this is also absurd, since we haven't yet defined what a “thought” even is. Interesting factoid: the Laboratory of Neuroimaging (LONI) at UCLA has taken credit for this number. But they did offer some caveats:
*This is still an open question (how many thoughts does the average human brain processes in 1 day). LONI faculty have done some very preliminary studies using undergraduate student volunteers and have estimated that one may expect around 60-70K thoughts per day. These results are not peer-reviewed/published. There is no generally accepted definition of what "thought" is or how it is created. In our study, we had assumed that a "thought" is a sporadic single-idea cognitive concept resulting from the act of thinking, or produced by spontaneous systems-level cognitive brain activations.
theoracleofdelphi-ga had some interesting thoughts on the matter:
So there's the heart of the problem: No one really knows what the biological basis for a 'thought' is, so we can't compute how fast a brain can produce them. Once you figure out the biological basis for a thought (and return from the Nobel ceremony) you can ask the question again and expect a reasonable scientific answer.

In the mean time, you could probably get a bunch of psychologists to argue about the definition of a thought for a while, and get a varying set of answers that depend highly on the definitions.
Oh, I think they said also 30 thoughts per second at another time in the movie...

3 Yeah, here's the “one electrode, one neuron” fallacy. The reality is that a single EEG electrode records summed, synchronous activity from thousands of neurons, at the very least.


Herff C, Heger D, de Pesters A, Telaar D, Brunner P, Schalk G, Schultz T. (2015). Brain-to-text: decoding spoken phrases from phone representations in the brain. Front Neurosci. 9:217.

Liu H, Agam Y, Madsen JR, Kreiman G. (2009). Timing, timing, timing: fast decoding of object information from intracranial field potentials in human visual cortex. Neuron 62(2):281-90.

Martin S, Brunner P, Holdgraf C, Heinze HJ, Crone NE, Rieger J, Schalk G, Knight RT, Pasley BN.  (2014). Decoding spectrotemporal features of overt and covertspeech from the human cortex. Front Neuroeng. 7:14.

Naselaris T, Olman CA, Stansbury DE, Ugurbil K, Gallant JL. (2015). A voxel-wise encoding model for early visual areas decodes mental images of remembered scenes. Neuroimage 105:215-28.

Pasley BN, David SV, Mesgarani N, Flinker A, Shamma SA, Crone NE, Knight RT, Chang EF. (2012). Reconstructing speech from human auditory cortex. PLoS Biol. 10(1):e1001251.

Quian Quiroga R, Kraskov A, Koch C, Fried I. (2009). Explicit encoding of multimodal percepts by single neurons in the human brain. Curr Biol. 19(15):1308-13.

Quiroga RQ, Reddy L, Kreiman G, Koch C, Fried I. (2005). Invariant visual representation by single neurons in the human brain. Nature 435(7045):1102-7.

Stocco, A., Prat, C., Losey, D., Cronin, J., Wu, J., Abernethy, J., & Rao, R. (2015). Playing 20 Questions with the Mind: Collaborative Problem Solving by Humans Using a Brain-to-Brain Interface PLOS ONE, 10 (9) DOI: 10.1371/journal.pone.0137303

Subscribe to Post Comments [Atom]

Monday, September 14, 2015

Mind Reading in the Red Room of "Listening"

"How am I supposed to work knowing that guy is listening to every thought that's going through my head? This is insane..."

David Thorogood and Ryan Cates are poor but brilliant Cal Tech grad students in Listening, a new neuro science fiction film by writer-director Khalil Sullins. Their secret garage lab invention of direct brain-to-brain communication has been hijacked by the CIA, who put it to nefarious use.

I'll take a closer look at the neuroscience (good and bad) in the next post.

Excessive use of filters? Perhaps...

Subscribe to Post Comments [Atom]

Monday, August 31, 2015

Cats on Treadmills (and the plasticity of biological motion perception)

Cats on a treadmill. From Treadmill Kittens.

It's been an eventful week. The 10th Anniversary of Hurricane Katrina. The 10th Anniversary of Optogenetics (with commentary from the neuroscience community and from the inventors). The Reproducibility Project's efforts to replicate 100 studies in cognitive and social psychology (published in Science). And the passing of the great writer and neurologist, Oliver Sacks. Oh, and Wes Craven just died too...

I'm not blogging about any of these events. Many many others have already written about them (see selected reading list below). And The Neurocritic has been feeling tapped out lately.

Hence the cats on treadmills. They're here to introduce a new study which demonstrated that early visual experience is not necessary for the perception of biological motion (Bottari et al., 2015). Biological motion perception involves the ability to understand and visually track the movement of a living being. This phenomenon is often studied using point light displays, as shown below in a demo from the BioMotion Lab. You should really check out their flash animation that allows you to view human, feline, and pigeon walkers moving from right to left, scrambled and unscrambled, masked and unmasked, inverted and right side up.

Biological Motion Perception Is Spared After Early Visual Deprivation

People born with dense, bilateral cataracts that are surgically removed at a later date show deficits in higher visual processing, including the perception of global motion, global form, faces, and illusory contours. Proper neural development during the critical, or sensitive period early in life is dependent on experience, in this case visual input. However, it seems that the perception of biological motion (BM) does not require early visual experience (Bottari et al., 2015).

Participants in the study were 12 individuals with congenital cataracts that were removed at a mean age of 7.8 years (range 4 months to 16 yrs). Age at testing was 17.8 years (range 10-35 yrs). The study assessed their biological motion thresholds (extracting BM from noise) and recorded their EEG to point light displays of a walking man and to scrambled versions of the walking man (see demo).

Behavioral performance on the BM threshold task didn't differ much between the congenital cataract (cc) and matched control (mc) groups (i.e., there was a lot of overlap between the filled diamonds and the open triangles below).

Modified from Fig. 1 (Bottari et al., 2015).

The event-related potentials (ERPs) averaged to presentations of the walking man vs. scrambled man showed the same pattern in cc and mc groups as well: larger to walking man (BM) than scrambled man (SBM).

Modified from Fig. 1 (Bottari et al., 2015).

The N1 component (the peak at about 0.25 sec post-stimulus) seems a little smaller in cc but that wasn't significant. On the other hand, the earlier P1 was significantly reduced in the cc group. Interestingly, the duration of visual deprivation, amount of visual experience, and post-surgical visual acuity did not correlate with the size of the N1.

The authors discuss three possible explanations for these results:
(1) The neural circuitries associated with the processing of BM can specialize in late childhood or adulthood. That is, as soon as visual input becomes available, initiates the functional maturation of the BM system. Alternatively the neural systems for BM might mature independently of vision. (2) Either they are shaped cross-modally or (3) they mature independent of experience.

They ultimately favor the third explanation, that "the neural systems for BM specialize independently of visual experience." They also point out that the ERPs to faces vs. scrambled faces in the cc group do not show the characteristic difference between these stimulus types. What's so special about biological motion, then? Here the authors wave their hands and arms a bit:
We can only speculate why these different developmental trajectories for faces and BM emerge: BM is characteristic for any type of living being and the major properties are shared across species. ... By contrast, faces are highly specific for a species and biases for the processing of faces from our own ethnicity and age have been shown.

It's more important to see if a bear is running towards you than it is to recognize faces, as anyone with congenital prosopagnosia ("face blindness") might tell you...


1 Troje & Westhoff (2006):
"The third sequence showed a walking cat. The data are based on a high-speed (200 fps) video sequence showing a cat walking on a treadmill. Fourteen feature points were manually sampled from single frames. As with the pigeon sequence, data were approximated with a third-order Fourier series to obtain a generic walking cycle."


Bottari, D., Troje, N., Ley, P., Hense, M., Kekunnaya, R., & Röder, B. (2015). The neural development of the biological motion processing system does not rely on early visual input Cortex, 71, 359-367 DOI: 10.1016/j.cortex.2015.07.029

Links to Pieces About Momentous Events

Remembering Katrina in the #BlackLivesMatter Movement by Tracey Ross

Hurricane Katrina Proved That If Black Lives Matter, So Must Climate Justice by Elizabeth Yeampierre

Project Katrina: A Decade of Resilience in New Orleans by Steven Gray

Hurricane Katrina, 10 Years Later, Buzzfeed's Katrina issue

ChR2: Anniversary: Optogenetics, special issue of Nature Neuroscience

ChR2 coming of age, editorial in Nature Neuroscience

Optogenetics and the future of neuroscience by Ed Boyden

Optogenetics: 10 years of microbial opsins in neuroscience by Karl Deisseroth

Optogenetics: 10 years after ChR2 in neurons—views from the community in Nature Neuroscience

10 years of neural opsins by Adam Calhoun

Estimating the reproducibility of psychological science in Science

Reproducibility Project: Psychology on Open Science Framework

How Reliable Are Psychology Studies? by Ed Yong

The Bayesian Reproducibility Project by Alexander Etz

A Life Well Lived, by those who maintain the Oliver Sacks, M.D. website.

Oliver Sacks, Neurologist Who Wrote About the Brain’s Quirks, Dies at 82, NY Times obituary

Oliver Sacks has left the building by Vaughan Bell

My Own Life, Oliver Sacks on Learning He Has Terminal Cancer

Subscribe to Post Comments [Atom]

Sunday, August 09, 2015

Will machine learning create new diagnostic categories, or just refine the ones we already have?

How do we classify and diagnose mental disorders?

In the coming era of Precision Medicine, we'll all want customized treatments that “take into account individual differences in people’s genes, environments, and lifestyles.” To do this, we'll need precise diagnostic tools to identify the specific disease process in each individual. Although focused on cancer in the near-term, the longer-term goal of the White House initiative is to apply Precision Medicine to all areas of health. This presumably includes psychiatry, but the links between Precision Medicine, the BRAIN initiative, and RDoC seem a bit murky at present.1

But there's nothing a good infographic can't fix. Science recently published a Perspective piece by the NIMH Director and the chief architect of the Research Domain Criteria (RDoC) initiative (Insel & Cuthbert, 2015). There's Deconstruction involved, so what's not to like? 2

ILLUSTRATION: V. Altounian and C. Smith / SCIENCE

In this massively ambitious future scenario, the totality of one's genetic risk factors, brain activity, physiology, immune function, behavioral symptom profile, and life experience (social, cultural, environmental) will be deconstructed and stratified and recompiled into a neat little cohort. 3

The new categories will be data driven. The project might start by collecting colossal quantities of expensive data from millions of people, and continue by running classifiers on exceptionally powerful computers (powered by exceptionally bright scientists/engineers/coders) to extract meaningful patterns that can categorize the data with high levels of sensitivity and specificity. Perhaps I am filled with pathologically high levels of negative affect (Loss? Frustrative Nonreward?), but I find it hard to be optimistic about progress in the immediate future. You know, for a Precision Medicine treatment for me (and my pessimism)...

But seriously.

Yes, RDoC is ambitious (and has its share of naysayers). But what you may not know is that it's also trendy! Just the other day, an article in The Atlantic explained Why Depression Needs A New Definition (yes, RDoC) and even cited papers like Depression: The Shroud of Heterogeneity. 4

But let's just focus on the brain for now. For a long time, most neuroscientists have viewed mental disorders as brain disorders. [But that's not to say that environment, culture, experience, etc. play no role! cf. Footnote 3]. So our opening question becomes, How do we classify and diagnose brain disorders neural circuit disorders in a fashion consistent with RDoC principles? Is there really One Brain Network for All Mental Illness, for instance? (I didn't think so.)

Our colleagues in Asia and Australia and Europe and Canada may not have gotten the funding memo, however, and continue to run classifiers based on DSM categories. 5 In my previous post, I promised an unsystematic review of machine learning as applied to the classification of major depression. You can skip directly to the Appendix to see that.

Regardless of whether we use DSM-5 categories or RDoC matrix constructs, what we need are robust and reproducible biomarkers (see Table 1 above). A brief but excellent primer by Woo and Wager (2015) outlined the characteristics of a useful neuroimaging biomarker:
1. Criterion 1: diagnosticity

Good biomarkers should produce high diagnostic performance in classification or prediction. Diagnostic performance can be evaluated by sensitivity and specificity. Sensitivity concerns whether a model can correctly detect signal when signal exists. Effect size is a closely related concept; larger effect sizes are related to higher sensitivity. Specificity concerns whether the model produces negative results when there is no signal. Specificity can be evaluated relative to a range of specific alternative conditions that may be confusable with the condition of interest.

2. Criterion 2: interpretability

Brain-based biomarkers should be meaningful and interpretable in terms of neuroscience, including previous neuroimaging studies and converging evidence from multiple sources (eg, animal models, lesion studies, etc). One potential pitfall in developing neuroimaging biomarkers is that classification or prediction models can capitalize on confounding variables that are not neuroscientifically meaningful or interesting at all (eg, in-scanner head movement). Therefore, neuroimaging biomarkers should be evaluated and interpreted in the light of existing neuroscientific findings.

3. Criterion 3: deployability

Once the classification or outcome-prediction model has been developed as a neuroimaging biomarker, the model and the testing procedure should be precisely defined so that it can be prospectively applied to new data. Any flexibility in the testing procedures could introduce potential overoptimistic biases into test results, rendering them useless and potentially misleading. For example, “amygdala activity” cannot be a good neuroimaging biomarker without a precise definition of which “voxels” in the amygdala should be activated and the relative expected intensity of activity across each voxel. A well-defined model and standardized testing procedure are crucial aspects of turning neuroimaging results into a “research product,” a biomarker that can be shared and tested across laboratories.

4. Criterion 4: generalizability

Clinically useful neuroimaging biomarkers aim to provide predictions about new individuals. Therefore, they should be validated through prospective testing to prove that their performance is generalizable across different laboratories, different scanners or scanning procedures, different populations, and variants of testing conditions (eg, other types of chronic pain). Generalizability tests inherently require multistudy and multisite efforts. With a precisely defined model and standardized testing procedure (criterion 3), we can easily test the generalizability of biomarkers and define the boundary conditions under which they are valid and useful.
[Then the authors evaluated the performance of a structural MRI signature for IBS presented in an accompanying paper.]

Should we try to improve on a neuroimaging biomarker (or “neural signature”) for classic disorders in which “Neuroanatomical diagnosis was correct in 80% and 72% of patients with major depression and schizophrenia, respectively...” (Koutsouleris et al., 2015)? That study used large cohorts and evaluated the trained biomarker against an independent validation database (i.e., it was more thorough than many other investigations). Or is the field better served by classifying when loss and agency and auditory perception go awry? What would individualized treatments for these constructs look like? Presumably, the goal is to develop better treatments, and to predict who will respond to a specific treatment(s).

OR should we adopt the surprisingly cynical view of some prominent investigators, who say:
...identifying a genuine neural signature would necessitate the discovery of a specific pattern of brain responses that possesses nearly perfect sensitivity and specificity for a given condition or other phenotype. At the present time, neuroscientists are not remotely close to pinpointing such a signature for any psychological disorder or trait...

If that's true, then we'll have an awfully hard time with our resting state fMRI classifier for neuro-nihilism.


1 Although NIMH Mad Libs does a bang up job...

2 Derrida's Deconstruction and RDoc are diametrically opposed, as irony would have it.

3 Or maybe an n of 1...  I'm especially curious about how life experience will be incorporated into the mix. Perhaps the patient of the future will upload all the data recorded by their memory implants, as in The Entire History of You (an episode of Black Mirror).

4 The word “shroud” always makes everything sound so dire and deathly important... especially when used as a noun.

5 As do many research groups in the US. This is meant to be snarky, but not condescending to anyone who follows DSM-5 in their research.


Insel, T., & Cuthbert, B. (2015). Brain disorders? Precisely. Science, 348 (6234), 499-500 DOI: 10.1126/science.aab2358

Woo, C., & Wager, T. (2015). Neuroimaging-based biomarker discovery and validation. PAIN, 156 (8), 1379-1381 DOI: 10.1097/j.pain.0000000000000223


Below are 34 references on MRI/fMRI applications of machine learning used to classify individuals with major depression (I excluded EEG/MEG for this particular unsystematic review). The search terms were combinations of "major depression" "machine learning" "support vector" "classifier".

Here's a very rough summary of methods:

Structural MRI: 1, 14, 22, 29, 31, 32

DTI: 6, 12, 18, 19

Resting State fMRI: 3, 5, 8, 9 11, 16, 17, 21, 28, 33

fMRI while viewing different facial expressions: 2, 7, 10, 24, 26, 27, 34

comorbid panic: 13

verbal working memory: 25

guilt: 15 (see The Idiosyncratic Side of Diagnosis by Brain Scan and Machine Learning)

Schizophrenia vs. Bipolar vs. Schizoaffective: 16

Psychotic Major Depression vs. Bipolar Disorder: 20

Schizophrenia vs. Major Depression: 23, 31

Unipolar vs. Bipolar Depression: 24, 32, 34

This last one is especially important, since an accurate diagnosis can avoid the potentially disastrous prescribing of antidepressants in bipolar depression.

Idea that may already be implemented somewhere: Individual labs or research groups could perhaps contribute to a support vector machine clearing house (e.g., at NTRIC or OpenfMRI or GitHub) where everyone can upload the code for data processing streams and various learning/classification algorithms to try out on each others' data.

Brain. 2012 May;135(Pt 5):1508-21. doi: 10.1093/brain/aws084.
Multi-centre diagnostic classification of individual structural neuroimaging scans from patients with major depressive disorder.
Mwangi B Ebmeier KP, Matthews K, Steele JD.

Bipolar Disord. 2012 Jun;14(4):451-60. doi: 10.1111/j.1399-5618.2012.01019.x.
Pattern recognition analyses of brain activation elicited by happy and neutral faces in unipolar and bipolar depression.
Mourão-Miranda J Almeida JR, Hassel S, de Oliveira L, Versace A, Marquand AF, Sato JR, Brammer M, Phillips ML.

PLoS One. 2012;7(8):e41282. doi: 10.1371/journal.pone.0041282. Epub 2012 Aug 20.
Changes in community structure of resting state functional connectivity in unipolar depression.
Lord A Horn D, Breakspear M, Walter M.

Neuroreport. 2012 Dec 5;23(17):1006-11. doi: 10.1097/WNR.0b013e32835a650c.
Machine learning classifier using abnormal brain network topological metrics in major depressive disorder.
Guo H Cao X, Liu Z, Li H, Chen J, Zhang K.

PLoS One. 2012;7(9):e45972. doi: 10.1371/journal.pone.0045972. Epub 2012 Sep 26.
Increased cortical-limbic anatomical network connectivity in major depression revealed by diffusion tensor imaging.
Fang P Zeng LL, Shen H, Wang L, Li B, Liu L, Hu D.

PLoS One. 2013;8(4):e60121. doi: 10.1371/journal.pone.0060121. Epub 2013 Apr 1.
What does brain response to neutral faces tell us about major depression? evidence from machine learning and fMRI.
Oliveira L Ladouceur CD, Phillips ML, Brammer M, Mourao-Miranda J.

Hum Brain Mapp. 2014 Apr;35(4):1630-41. doi: 10.1002/hbm.22278. Epub 2013 Apr 24.
Unsupervised classification of major depression using functional connectivity MRI.
Zeng LL Shen H, Liu L, Hu D.

Psychiatry Clin Neurosci. 2014 Feb;68(2):110-9. doi: 10.1111/pcn.12106. Epub 2013 Oct 31.
Aberrant functional connectivity for diagnosis of major depressive disorder: a discriminant analysis.

Neuroimage. 2015 Jan 15;105:493-506. doi: 10.1016/j.neuroimage.2014.11.021. Epub 2014 Nov 15.
Sparse network-based models for patient classification using fMRI.
Rosa MJ Portugal L Hahn T Fallgatter AJ Garrido MI Shawe-Taylor J Mourao-Miranda J.

Proc IEEE Int Symp Biomed Imaging. 2014 Apr;2014:246-249.
Sacchet MD Prasad G Foland-Ross LC Thompson PM Gotlib IH.

Front Psychiatry. 2015 Feb 18;6:21. doi: 10.3389/fpsyt.2015.00021. eCollection 2015.
Support vector machine classification of major depressive disorder using diffusion-weighted neuroimaging and graph theory.
Sacchet MD Prasad G Foland-Ross LC Thompson PM Gotlib IH.

J Affect Disord. 2015 Sep 15;184:182-92. doi: 10.1016/j.jad.2015.05.052. Epub 2015 Jun 6.
Separating depressive comorbidity from panic disorder: A combined functional magnetic resonance imaging and machine learning approach.
Lueken U Straube B Yang Y Hahn T Beesdo-Baum K Wittchen HU Konrad C Ströhle A Wittmann A Gerlach AL Pfleiderer B, Arolt V, Kircher T.

PLoS One. 2015 Jul 17;10(7):e0132958. doi: 10.1371/journal.pone.0132958. eCollection 2015.
Structural MRI-Based Predictions in Patients with Treatment-Refractory Depression (TRD).
Johnston BA Steele JD Tolomeo S Christmas D Matthews K.

Psychiatry Res. 2015 Jul 5. pii: S0925-4927(15)30025-1. doi: 10.1016/j.pscychresns.2015.07.001. [Epub ahead of print]
Machine learning algorithm accurately detects fMRI signature of vulnerability to major depression.
Sato JR Moll J Green S Deakin JF Thomaz CE Zahn R.

Neuroimage. 2015 Jul 24. pii: S1053-8119(15)00674-6. doi: 10.1016/j.neuroimage.2015.07.054. [Epub ahead of print]
A group ICA based framework for evaluating resting fMRI markers when disease categories are unclear: Application to schizophrenia, bipolar, and schizoaffective disorders.
Du Y Pearlson GD Liu J Sui J Yu Q He H Castro E Calhoun VD.

Neuroreport. 2015 Aug 19;26(12):675-80. doi: 10.1097/WNR.0000000000000407.
Predicting clinical responses in major depression using intrinsic functional connectivity.
Qin J, Shen H, Zeng LL, Jiang W, Liu L, Hu D.

J Affect Disord. 2015 Jul 15;180:129-37. doi: 10.1016/j.jad.2015.03.059. Epub 2015 Apr 4.
Altered anatomical patterns of depression in relation to antidepressant treatment: Evidence from a pattern recognition analysis on the topological organization of brain networks.
Qin J, Wei M, Liu H Chen J Yan R Yao Z Lu Q.

Magn Reson Imaging. 2014 Dec;32(10):1314-20. doi: 10.1016/j.mri.2014.08.037. Epub 2014 Aug 29.
Abnormal hubs of white matter networks in the frontal-parieto circuit contribute to depression discrimination via pattern classification.
Qin J, Wei M, Liu H Chen J Yan R Hua L Zhao K Yao Z Lu Q.

Biomed Res Int. 2014;2014:706157. doi: 10.1155/2014/706157. Epub 2014 Jan 19.
Neuroanatomical classification in a population-based sample of psychotic major depression and bipolar I disorder with 1 year of diagnostic stability.
Serpa MH, Ou Y Schaufelberger MS Doshi J Ferreira LK Machado-Vieira R Menezes PR Scazufca M Davatzikos C Busatto GF Zanetti MV.

Psychiatry Res. 2013 Dec 30;214(3):306-12. doi: 10.1016/j.pscychresns.2013.09.008. Epub 2013 Oct 7.
Identifying major depressive disorder using Hurst exponent of resting-state brain networks.
Wei M Qin J, Yan R, Li H, Yao Z, Lu Q.

J Psychiatry Neurosci. 2014 Mar;39(2):78-86.
Characterization of major depressive disorder using a multiparametric classification approach based on high resolution structural images.
Qiu L Huang X Zhang J Wang Y Kuang W Li J Wang X Wang L Yang X Lui S Mechelli A Gong Q2.

PLoS One. 2013 Jul 2;8(7):e68250. doi: 10.1371/journal.pone.0068250. Print 2013.
Convergent and divergent functional connectivity patterns in schizophrenia and depression.
Yu Y Shen H, Zeng LL, Ma Q, Hu D.

Eur Arch Psychiatry Clin Neurosci. 2013 Mar;263(2):119-31. doi: 10.1007/s00406-012-0329-4. Epub 2012 May 26.
Discriminating unipolar and bipolar depression by means of fMRI and pattern classification: a pilot study.
Grotegerd D Suslow T, Bauer J, Ohrmann P, Arolt V, Stuhrmann A, Heindel W, Kugel H, Dannlowski U.

Neuroreport. 2008 Oct 8;19(15):1507-11. doi: 10.1097/WNR.0b013e328310425e.
Neuroanatomy of verbal working memory as a diagnostic biomarker for depression.
Marquand AF Mourão-Miranda J, Brammer MJ, Cleare AJ, Fu CH.

Biol Psychiatry. 2008 Apr 1;63(7):656-62. Epub 2007 Oct 22.
Pattern classification of sad facial processing: toward the development of neurobiological markers in depression.
Fu CH Mourao-Miranda J, Costafreda SG, Khanna A, Marquand AF, Williams SC, Brammer MJ.

Neuroreport. 2009 May 6;20(7):637-41. doi: 10.1097/WNR.0b013e3283294159.
Neural correlates of sad faces predict clinical remission to cognitive behavioural therapy in depression.
Costafreda SG Khanna A, Mourao-Miranda J, Fu CH.

Magn Reson Med. 2009 Dec;62(6):1619-28. doi: 10.1002/mrm.22159.
Disease state prediction from resting state functional connectivity.
Craddock RC Holtzheimer PE 3rd, Hu XP, Mayberg HS.

Neuroimage. 2011 Apr 15;55(4):1497-503. doi: 10.1016/j.neuroimage.2010.11.079. Epub 2010 Dec 3.
Prognostic prediction of therapeutic response in depression using high-field MR imaging.
Gong Q Wu Q, Scarpazza C, Lui S, Jia Z, Marquand A, Huang X, McGuire P, Mechelli A.

Neuroimage. 2012 Jun;61(2):457-63. doi: 10.1016/j.neuroimage.2011.11.002. Epub 2011 Nov 7.
Diagnostic neuroimaging across diseases.
Klöppel S Abdulkadir A, Jack CR Jr, Koutsouleris N, Mourão-Miranda J, Vemuri P.

Brain. 2015 Jul;138(Pt 7):2059-73. doi: 10.1093/brain/awv111. Epub 2015 May 1.
Individualized differential diagnosis of schizophrenia and mood disorders using neuroanatomical biomarkers.
Koutsouleris N Meisenzahl EM Borgwardt S Riecher-Rössler A Frodl T Kambeitz J Köhler Y Falkai P Möller HJ Reiser M Davatzikos C.

JAMA Psychiatry. 2014 Nov;71(11):1222-30. doi: 10.1001/jamapsychiatry.2014.1100.
Brain morphometric biomarkers distinguishing unipolar and bipolar depression. A voxel-based morphometry-pattern classification approach.
Redlich R Almeida JJ Grotegerd D Opel N Kugel H Heindel W Arolt V Phillips ML Dannlowski U.

Brain Behav. 2013 Nov;3(6):637-48. doi: 10.1002/brb3.173. Epub 2013 Sep 22.
A reversal coarse-grained analysis with application to an altered functional circuit in depression.
Guo S Yu Y Zhang J Feng J.

Hum Brain Mapp. 2014 Jul;35(7):2995-3007. doi: 10.1002/hbm.22380. Epub 2013 Sep 13.
Amygdala excitability to subliminally presented emotional faces distinguishes unipolar and bipolar depression: an fMRI and pattern classification study.
Grotegerd D Stuhrmann A, Kugel H, Schmidt S, Redlich R, Zwanzger P, Rauch AV, Heindel W, Zwitserlood P, Arolt V, Suslow T, Dannlowski U.

Subscribe to Post Comments [Atom]

eXTReMe Tracker