Neurofractal

[neurofractal-blog]

The Nature Paper Rant & the Trouble with Hyping Autism Research

So earlier this week I went on a little bit of a twitter rant about a recently published and widely publicised letter to Nature. The study used MRI to collect structural brain images from 106 infants with an older ASD-sibling and 42 “low-risk” infants. The authors observed brain overgrowth in several parts of the cortex at very early ages (6 & 12 months) in the high-risk infant group. Furthermore, using this structural brain information, a machine learning algorithm was able to classify those infants who went on to diagnosed with autism at 24 months (15 in total) with around 80% accuracy.

Right I'm about to start a rant here brace yourselves...1/n https://t.co/7PMrIO4Nw5

— Robert (@neurofractal) February 15, 2017

I could give a detailed account of the merits of this paper (which are many), alongside a few technical critiques which might be of interest to students/academics in the field. But I’m not. That’s not what I was upset about. What concerned me was the process by which this information was reported to the general public and the effect this might have on autism support groups and parents more generally.

 Let me start out with a couple of quotes:

 “Brain scans can detect autism long before any symptoms start to emerge, say scientists” BBC News Website

 “Scientists have successfully detected autism in babies using an MRI scan.” Daily Mail

Brilliant! We can all stop doing research because early ASD diagnosis has been solved. Apart from the fact that autism diagnosis was predicted in just 15 infants, there is no causative mechanism which links brain volume and autistic behaviour and the classification accuracy was 80% which is nowhere near high enough for effective clinical use. This is a classic case of seriously over-hyped science.

I get that reporting science to the general public is hard, especially with brain imaging and autism research. There are technical terms, statistical concepts and contradictory findings which are hard to convey. However, I think that people deserve a balanced view which appreciates the complexities of the field and emphasises how much work still needs to be done. Where are the phrases “may become part of autism diagnosis in future”, “promising early results”, “important first steps”? This black and white reporting has serious implications for discussions in autism support groups and the mind-set of parents. During my PhD, I have encountered parents who show me newspaper articles about over-hyped autism ‘cures’, ‘treatments’ and ‘diagnosis’, and hope that by getting a brain scan they can understand why their child has autism or get a faster diagnosis. In my opinion, the BBC and Daily Mail articles, and others like them, lead to this kind of all-or-nothing thinking within the autism community. It is irresponsible and frankly unethical to simplify and hype-up research findings in the hope that a few more people click your link. I was particularly angered by the BBC News Science headline “Autism detectable in brain long before symptoms appear” because the BBC is seen as such a reliable and impartial source of knowledge. I really hope that future articles on the site reporting on autism will use more appropriate language.

Let me finally offer my own mini write-up in the hope that at least some autism support groups and parents see the Nature letter in the correct context:

Autism is a complex neurological condition with no known ‘cause’. The prediction of an autism diagnosis or autism-related behaviours before 24 months is crucial because it allows families and clinicians to seek appropriate interventions at a critical time during development. 

In the future, this may become possible through a careful combination of genetic information, brain scanning and behavioural assessment. Unfortunately, this is not possible at the present time. Our knowledge about the signs of autism before 24 months are poorly understood. There is a complex interplay between genetic and environment factors, and how this shapes the brain’s overall shape and functional working during development.

This is why the recent letter to Nature by Hazlett & colleagues linking brain volume with autism is important. It builds upon previous work linking atypical brain growth with autism using a larger number of people and data from very young infants. Furthermore, a machine learning algorithm showed that this overgrowth could be be predictive of autism diagnosis on an individual basis. The 80% classification rate is higher than previous work and would be exciting if replicated and extended using more data.

Ultimately studies like this one will help us to understand and predict autism before 24 months. But we aren’t there quite yet.

Some other useful readings:

The last paragraph of the article itself (wonder how many reporters read this?) 

Twitter thread by @DrBrocktagon 

NHS News Report

Guardian News Report

ITV News report

[neurofractal-blog]

The Atypical Spread of Neural Activity in Autism

Last year a study published in BRAIN by Khan et al., (2015) reported atypical neural responses in a group of young ASD (mean age = 11.6) participants in response to 25Hz vibrotactile stimulation. The authors showed that feedforward connectivity from primary somatosensory cortex (S1) to secondary somatosensory cortex (S2) was increased in the ASD sample. This was accompanied by reduced 25-to-50Hz phase-locking in the ASD cohort, which the authors argued was due to the lack of recurrent activity at the local level.

The paper also seemed to indicate that very early (30-70ms) evoked activity in response to the vibrotactile stimulation was increased in the ASD cohort. To further address this finding, Khan & colleagues (2016) applied a measure called spatio-temporal divergence (S-T Div) to study onset response dynamics within S1. This metric essentially estimates the spatial extent of the onset response combined with the velocity of the spread of this activity. Importantly, S-T Div is not affected by field spread, and robust bootstrap statistics can be performed to correct for multiple comparisons.

The results of this analysis indicated that the spatio-temporal spread of evoked S1 activity in response to vibrotactile stimulation was much greater for the ASD compared with the typically developing control group (see below). Furthermore S-T Dev values could be successfully used to classify ASD vs TD participants using a linear discriminant analysis.

These results imply that connectivity between the thalamus and primary sensory cortices is increased in autism, adding to evidence that increased feedforward connectivity is a hallmark of ASD. The authors also hypothesise that this uncontrolled feedforward activity may be the underlying basis of reported hyper/hyposensitivities in autism, consistent with findings from fMRI/EEG and theories of excitation-inhibition. Having said this, the authors did not explicitly measure sensory atypicalities, only reporting correlations with touch and behavioural ADOS scores. In particular, more work is needed to verify the claim that sensory hyposensitivies in autism arise as a result of “generalized down regulation” mechanisms. Future studies should therefore use qualitative and quantitative measures of sensory hypo/hyper-sensitivity to subdivide ASD samples into hypo/hyper subgroups, followed by measures to estimate feedforward and/or compensatory sensory processing.

Overall, this is a particularly strong study because it makes full use of MEG’s excellent spatiotemporal properties, reporting on well-defined neural activity within an isolated system. Estimates of connectivity relying upon more direct measures of neural activity can only help clarify the basis of atypical connectivity in autism and help resolve inconsistencies in the literature. It will also be interesting to see if these results generalise to other sensory systems e.g. the visual and auditory domain (disclaimer – I’m trying to do this at the moment in my PhD!). However, it should be noted that these systems are notably more “complicated” than the somatosensory system, and results will be much trickier to interpret.

[neurofractal-blog]

The Mind in the Eyes Test

As part of my first PhD experiment, I have decided to collect some behavioural data using the Mind in the Eyes Task. This is a classic test of theory of mind / emotional recognition abilities. In the task you are presented with pictures of people’s eyes and then asked to choose an emotional word which best describes the emotion they are feeling. Clear differences have been found in accuracy between ASD and control participants. 

Example stimuli from my version of the Mind in the Eyes task

To save time during my data analysis, I decided to program a version of this paradigm using Psychopy - a python-based open-source stimulus presentation program. I am planning to run the paradigm on my laptop and participant responses will be analysed automatically.

The stimuli, code and instructions for my version of the Mind in the Eyes task can be found here https://goo.gl/X9i1tE - I hope someone finds this useful in the future :)

[neurofractal-blog]

Six Months into my PhD...

So it’s been six months since I started my PhD, and I thought it would be a good idea to write a blog post in order to reflect on some of the stuff I’ve learnt and share what I’ve found out (disclaimer – nothing new!).

Finding #1 - I process visual information in the right bit of the brain!

My PhD project is essentially testing the idea that autism or autism spectrum disorders (ASD) can be characterised by patterns of atypical brain connectivity. Specifically, people with ASD are thought to show long-range under-connectivity, but local hyperconnectivity (which also ties in quite nicely with the weak central coherence account of ASD). These alterations to brain connectivity are thought to underlie the atypical development of complex cognitive skills like social cognition, working memory and language comprehension, and could also result in repetitive behaviours/restricted interests. In a nutshell, I am interested in whether children and adolescents with ASD exhibit common patterns of functional brain connectivity which link sensory processing, cognition and ‘resting-state’ activity. To address this research question, I am using a non-invasive brain imaging technique called magnetoencephalography (MEG), which combines excellent temporal resolution with increasingly sophisticated source localisation techniques. One of the advantages of MEG is that you can explore the mechanisms underlying functional connectivity associated with neuronal oscillations and the interactions between these oscillations. We are currently recruiting both ASD and non-ASD participants aged 14-20 for a research study at the Aston Brain Centre – so if you know anyone in the Birmingham area interested in getting a brain scan please do share this advert.

Since starting in Birmingham, I have had to come to terms with the fact that PhD students do not spend their entire time doing research. Training courses, ethics applications, writing, IT and admin have all taken up a far larger proportion of my time than expected. Being a PhD student means being in charge of almost every aspect of your research project, including all the (boring) admin stuff. This is cool but scary at the same time – I’m always worried I’ve made a mistake and will end up with no data to analyse!

Having said this, I have done some research and got some data(!). In order to test my MEG paradigm, I collected data from 8 non-ASD participants in our lab. I analysed the data in a MATLAB toolbox called Fieldtrip, initially designed at the Donders Institute in Nijmegen. This was a struggle at first, mainly because I had never programmed in Matlab. However, once I got over the initial learning curve, Fieldtrip turned out to be a really useful tool for MEG data analysis due to it’s flexibility, and implementation of non-parametric statistics. You can customise each step of your MEG analysis using ‘configuration options’ and also write batch scripts for group analysis. This takes away some of the tedious button-heavy user-interfaces prevalent in neuroimaging software. I’m currently writing a few tutorials for Elekta-Fieldtrip data analysis, so if anyone is interested in these then feel free to email – link.

My ‘pipeline’ for MEG data analysis for data collected using an Elekta Neuromag scanner

I recently shared the results of my pilot study at MEG UK, held at the University of York. You can read my poster here if you wish… but basically I found what I wanted! High frequency gamma-band and lower frequency alpha-band activity associated with visual and auditory processing localised to the right bits of the brain (occipital and superior temporal cortex). This was followed by time-frequency analysis in source space which, again, produced predictable results. The data I’m most excited about is the group-level time-frequency plot from my auditory stimuli. As you can see below the clicktrain presented to participants produces a really nice steady state response in auditory cortex, at the expected 40Hz. The next step is going be the application of more complex connectivity metrics to the data, for example peak gamma amplitude, DSI and phase-amplitude coupling.

Reflecting on the first six months of my PhD, I think the most important thing I have learnt is that collaboration in research is essential. The neuroimaging centre I have been working at has seen a number of people switch to analysing data in Fieldtrip. I would not have been able to learn Matlab and Fieldtrip so fast were it not for other people sharing their scripts and their time with me, both at Aston and at other MEG labs around the world. Yes, research is ultimately a personal exercise, but collective effort helps to speed up research and makes things more fun along the way. For this reason, I have been very keen to share my scripts on the lab WIKI and organise discussion sessions so that people can talk about data analysis issues and share ideas. I hope I will remember to continue this even when I have less time than I do now!

[neurofractal-blog]

Autism Brain Connectivity Talk

Just thought I’d share these slides I made for a talk I'm giving about brain connectivity in autism. The focus is very much on electrophysiological measures of brain connectivity and a novel way to view findings from the field.

[neurofractal-blog]

Phase Amplitude Coupling (PAC) – a mechanism for the dynamic coordination of neural activity?

Over the past decade or so, ultra-slow fluctuations (<0.1Hz) in the fMRI BOLD response have been used to reveal the intrinsic architecture of the human brain. It is now well established that there are separable resting state networks (RSNs), which are consistent across time and subjects (Yeo et al, 2011). Slow-wave low-frequency oscillations have also been predominant in the study of electrophysiological resting-state activity, especially in recent MEG studies (e.g. Brookes et al., 2011). Yet most neuroscientists would agree that brain activity cannot be described simply by the slow synchronisation between regions within a network. Brain activity is fast and dynamic, reflecting periods of very rapid local processing within specific brain regions. Our current understanding of resting-state activity therefore lacks an appreciation of the dynamicity of brain function. The relatively new study of phase amplitude coupling (PAC) provides an ideal mechanism to explain such dynamic processing within the context of large-scale neurocognitive networks. I will admit that when I first came across PAC it seemed too mathematical and abstract to be applied to the study of cognition. However, three months into my PhD, having read many more oscillation/PAC papers, I’m beginning to really excited about the idea.

The classic Yeo et al., (2011) RSNs.

Terminology

Frequency describes the rate of the oscillation

Phase describes where you are in the cycle of an oscillation (i.e. at the maximum amplitude, zero amplitude or somewhere in between)

Amplitude describes the strength of the oscillation (i.e. the maximum value of the oscillation)

The concept of PAC is actually very simple. Neuronal oscillations are often split up into frequency bands (e.g. theta, alpha, gamma…) all with their own unique role in cognitive processing. There are many established methods to determine whether the phase of distinct frequency bands is synchronous across the brain – for example coherence, Granger causality and transfer entropy (all outlined in this recent review by Andre Bastos & Kan-Mathijs Schoffelen). However, evidence is accumulating that instead of acting in isolation, certain neuronal oscillations interact with each other in functionally significant ways. One of the most common examples of this is PAC, which is a coupling (or statistical dependence) between the phase of a slow oscillation (e.g. theta) and the amplitude of another faster oscillation (e.g. gamma). During PAC, brain activity is coordinated in such a way that gamma is only observed during certain periods of the slower frequency oscillation, producing a rhythmical pattern of gamma-band responses (see figure below).

For me, one of the most exciting things about PAC is that it has been observed during intracranial recordings of human brain activity, and can be detected using EEG and MEG (Hyafil et al., 2015). We are therefore actually studying something directly physiological about the brain, rather than relying on a correlational measure of brain activity like the BOLD response (sorry for the jibe blobologists, but it’s got to be said). Furthermore, PAC has been linked to the efficiency of learning, memory and language processing (Canolty & Knight, 2006), making this physiological mechanism relevant to the study of cognition.

Illustration to demonstrate the concept of PAC, taken from Voytek et al., (2010). In one experiment from this paper the authors use ECOG recordings whilst patients perform a visual task. The time-frequency plot clearly shows a rhythmical pattern of gamma-band activity (50-250Hz), closely following the phase of an alpha rhythm.

Unsurprisingly, to calculate PAC you need to extract both the phase of the slow wave oscillation and the amplitude of the higher frequency oscillation. You then apply a statistical metric to determine how related these two measure are. There are many algorithms to do this, with perhaps the most popular being the Modulation Index (Canolty et al., 2006). However, it is important to note that there are many disagreements in the literature about which metric is the most statistically robust (for more on this see the recent critical review of cross-frequency coupling by Aru & colleagues).

So what does this all of this have to do with the brain? Well, broadband gamma is thought to reflect synchronised cortical activity within specific local circuits, possibly mediated via interneurons. Put simply, gamma-band responses currently represent the best candidate to capture basic neural computation at a local level. The statistical coupling between this high frequency activity and a lower frequency oscillation provides an ideal coordinating mechanism, such that several ‘active’ regions are bound into a mutually co-operative network. This is demonstrated below. According to the so-called ‘synchronised gating model’, when two regions are functionally coupled their phases align and allow rapid & high-frequency gamma-band activity to become coordinated between distant cortical regions. However, if two regions are out of phase, or if there is no broadband gamma, then those regions will not be functionally connected. There is preliminary evidence that failure to establish such synchronous neural activity may underlie several neurological conditions like Parkinson’s, autism and schizophrenia (Voytek & Knight, 2015).

Illustration of the so-called ‘synchronised gating’ hypothesis. Taken from Florin & Baillet (2015).

Two recent papers have provided strong support for this model of dynamic brain activity. Firstly, Florin & Baillet (2015) collected resting-state MEG data from 12 participants, and mapped this to source space using a minimum norm estimate. The authors were able to demonstrate that PAC is ubiquitously present across the cortex with the phase of delta, theta and alpha oscillations all related to the amplitude of gamma. Interestingly there was no spatial pattern to this activity. They then computed a model of MEG-PAC based on the time-series of the MEG data (outlined in the paper). By projecting this model into source-space, the authors were able to extract many well-established resting-state networks, which closely resembled those generated by the more common method outlined in Brookes et al., 2011 (see figure below). This suggests that during periods of synchronous activity within neuro-cognitive networks, the phase of low frequency oscillations is coupled to the amplitude of gamma oscillations.

The main findings of Florin & Baillet (2015). Five different resting-state maps derived from the MEG-PAC model are shown. These bear a striking resemblance to the standard maps produced using the method outlined in Brookes et al., 2011.

A recent paper by Weaver & colleagues from the University of Washington also lends support to the synchronous gating hypothesis. By using three epilepsy patients, the authors directly compared PAC derived from resting-state ECOG data with resting-state fMRI data. By applying rigorous statistical corrections to a PAC metric, the authors were able to extract fairly reliable resting-state maps from a seed placed within sensorimotor cortex. These maps overlapped significantly with rs-fMRI maps produced by placing a seed in the same area of sensorimotor cortex.

Obviously much more research is needed to verify these claims about PAC and the synchronous gating hypothesis. The previous two studies use fairly small sample sizes, and different measures of PAC to the usual Modulation Index. Furthermore, a recent paper by Aru et al., (2015) suggests that some neuroscience papers may be reporting spurious examples of CFC unrelated to cognitive processing. Despite these methodological concerns, PAC remains a promising theoretical mechanism to explain the dynamic coordination of brain activity both within the resting-state and during task-driven activity.

Further Reading:

Classic Canolty & Knight (2006) TiCS review

Canolty et al., (2006) Science paper

Excellent (Hyafil et al., 2015) review of the different types of cross-frequency coupling, including PAC

Aru et al., (2015) – This paper points out many methodological considerations to avoid spurious cross-frequency coupling

Implementing measures of PAC in BrainStorm

Excellent blog post by Bradley Voytek

[neurofractal-blog]

An illustrated guide to the triad of impairments in autism

Source: http://www.autismtopics.org/

[neurofractal-blog]

5 Things I Learnt During my MSc Course

1.     fMRI is great, but…

fMRI has undoubtedly revolutionised the field of neuroscience, granting scientists the opportunity to study the functional properties of the brain, non-invasively. I used to read fMRI cognitive neuroscience papers and think “this is the way that memory or vision or attention actually works”. BUT, like any research technique fMRI has flaws. It ultimately relies upon an indirect measure of neural activity, bloodflow, rather than measuring the process itself. On top of this, the temporal resolution of fMRI is limited to 1-2 seconds so that information about the dynamic nature of brain activity is lost. As I got into neuroimaging data analysis I also began to realise that many of the statistical analyses employed in early work were a little bit dodgy (double dipping etc.). Don’t get me wrong, I still find fMRI fascinating and think it has a place in cognitive neuroscience... but it has to be combined with other techniques (MEG/EEG for temporal resolution and TMS for causal evidence) to fully address a research question.

2.     Stats and Computers are your friend

Stats * sigh *. Obviously you need a good understanding of statistics to do anything meaningful in science. With a bit of work, I even came to quite enjoy my statistics course (especially the non-parametric stuff). SPSS is confusing and stupid and dull for everyone, so don’t despair – Andy Field is your friend :)

As well as learning about stats, we were taught how to program using Python. As I have mentioned in another blogpost, this really helped me to save time during my MSc project and do more complicated, unconventional analyses. As data analysis is largely computer based, I feel that learning how to use data analysis software in an effective way is one of the most important skills to master for any research job.

3.     Doing nothing can tell you something!  

One of the most astonishing findings to come out of neuroscience in the past few years is that spontaneous brain activity, whilst participants are doing nothing apart from sitting quietly, can be largely segregated into resting state networks. These networks coincide with the major functional systems of the brain – the default mode, motor, visual and dorsal attention networks to name but a few. The functional properties of these networks are also linked to behaviour. For example the rs-connectivity between regions of the superior parietal lobule and temporal lobe can explain individual differences in the length of time between bistable percepts (Baker et al., 2015). This turns the whole premise of subtraction-based cognitive neuroscience research on it’s head – you don’t need a task block and a control block to study the functional properties of the brain anymore, you can just pop people into a scanner for 5 minutes whilst they sit quietly.

View post on imgur.com

Using a complex analysis technique pioneered by Matthew Brookes, the same resting-state networks can be localised using MEG. In the next few years, I anticipate that research into these MEG rs-networks will produce some very informative insights about the dynamics of whole brain cortical activity.    

4.     Psychology experiments are boring (mostly)

There’s a reason why psychology departments introduce mandatory participant pools for undergraduates – taking part in psychological research is dull. One working memory experiment I volunteered for was so tedious I think I might have fallen asleep halfway through. Neuroimaging experiments aren’t much better. You have a choice of sitting in a claustrophobic, noisy magnet (fMRI); getting sticky gel in your hair (EEG); having annoying electrodes on your face (MEG); or getting zapped with a powerful magnet (TMS). When designing an experiment, you have to be aware of these practical constraints. Participant’s aren’t going to perform your task properly if they are uncomfortable or bored. It is therefore vital to incorporate frequent breaks into your experiment and to thoroughly prepare people for what they are going to experience whilst in the scanner.    

5.     Research is collaborative

During my MSc course at York, I began to appreciate how vital collaboration is for science. Whilst most research is conducted in small groups within a department, the exchange of knowledge and ideas across disciplines is vital for most ground-breaking work. From a personal perspective, when I was learning about neuroimaging analysis I was lucky enough to work alongside some very helpful RAs, PhD students and postdocs who had encountered many of the same issues as me. Furthermore, during my MSc project/dissertation, I really benefitted from having two supervisors who were able to address issues from different angles.

As a side note, I would also mention that I’ve found Twitter to be a great way to share research and ideas in a (mostly!) productive manner – I’d advise aspiring researchers to set up an account.

[neurofractal-blog]

Reddit dailyprogrammer - Cellular Automata Rule #90 Python Code

So to keep my python skills up to speed I’ve recently been trying a few challenges from the dailyprogrammer subreddit. If you need some coding practice then the website is a really useful resource - they’ve got easy, intermediate and expert level challenges designed to be completed in a relatively short period of time.

Here, I just thought I’d post my code to a recent challenge (#213) based on the cellular automata rule #90. You can read the instructions here - it basically produces a nice fractal pattern based on an input of 0 & 1s. Any comments / constructive criticism would be much appreciated.  

Input = 00000000000000000000000000000000000000000000000001000000000000000000000000000000000000000000000000

txt = open("CA_2.txt").read() #opens the text file iterations = 25 #set number of iterations n = 0 while n <= iterations: #carries on for the specified no. #of iterations output = '' for number in list(txt): if number == '1': output += '*' elif number == '0': output += ' ' print (output) #prints the line converting 0 -> ' ' #and 1 -> 'x' new_txt = '' new_txt += txt[1] #this applies the cellular for number in range(len(txt)-2): #automata XOR rules if txt[number] == txt[(number+2)]: new_txt += '0' elif txt[number] != txt[(number+2)]: new_txt += '1' new_txt += txt[len(txt)-2] txt = new_txt n = n + 1

Output = http://imgur.com/7tAzaun

[neurofractal-blog]

Plotly

I’ve just found this really cool site called Plotly. It’s a free online resource that produces professional-looking interactive graphs. To try out the software (I apologise in advance for those not interested in cricket*) I made this graph to show the cumulative test average of five English fast bowlers. As you hover over the graph, some pop-up text should appear with the bowler’s average and some information about the opposition.

To produce the graph it was fairly easy to paste in some data from Excel and click the X and Y buttons for each variable (see picture). I also clicked the text option (choose as T) in order to add the information about the opposition to the final graph. One thing that I found slightly annoying was having to name every variable by clicking on a drop-down menu then rename and then edit! The final graph editor was also a bit confusing to navigate, but I’m sure that this will improve with time.

  It also looks like you can integrate Plotly with the matplotlib library of python, Matlab and R, which will make it an excellent resource for producing more complex graphs. My hope is that journals will cotton on to the idea of interactive resources and start to include graphs like this in articles published online. 

* Data were aquired from the ESPN CricInfo website

[neurofractal-blog]

The distributed set of regions involved in semantic cognition, adapted from Binder et al., (2009)

[neurofractal-blog]

Graphic showing the different strands of research on the Default Mode Network. Source

[neurofractal-blog]

Illustration of the ventral (semantic) language network using a tube map. Source

[neurofractal-blog]

Resources for Learning to Program using Python and Code Experiments Using Psychopy

As part of my MSc course, I recently took a module called ‘Programming for Neuroimaging’, which introduced several basic programming skills often employed in modern psychological and neuroscientific research. We primarily used python, which is becoming increasingly popular within the scientific community, as well as the open-source stimulus presentation software Psychopy. Being a self-confessed computer geek, I really enjoyed the module and developed loads of valuable skills that I will definitely be using in the future.

Here I thought I’d compile a list of resources/tips, which I have used over the past few months. Hopefully someone else will find these links useful whilst learning to program and/or use Psychopy. For further links to resources, take a look at a similar post by Matt Wall on his excellent Computing for Psychologists blog.

Basic Introduction to Python:

Code Academy (Python) – nice gentle introduction to programming, featuring completely browser-based tutorials

A Non-Programmers Guide To Python – a more traditional textbook-like resource

Tutorials Point – online resource featuring lots of useful examples

Google’s Python Class – fast-paced introduction to python, probably aimed at people with previous programming experience. Contains some good tips/exercises to revisit when you have developed some programming knowledge

Spyder – Powerful Graphical User Interface for scientific programming using python

Example Exercises – some interesting exercises to try if you want a challenge

Psychopy (Coding Mode):

Main website – download Psychopy from here

Documentation & Tutorials – contains the complete Psychopy documentation as well as a couple of tutorials using the coding mode

Experimental Template – Excellent resource for coding complicated experiments using Psychopy, created by Jonas Lindeløv. This template can be used to form the basis of almost any visual psychological experiment.

Tips For When You’re Stuck:

- Look at the error message – where is the problem in your code?

- Check for missing commas, full stops and brackets - so many issues are solved this way!

- Check you have imported the right modules

- Google the problem. Chances are you’re not the only one who has encountered the same issue. The website Stackoverflow can be very useful in certain situations

- Find out what your script is actually doing. Adding a print function often helps to diagnose the issue

If you have a programming question you can reach me on Twitter or leave a comment below. Happy programming!

[neurofractal-blog]

Nice sunny day in York

[neurofractal-blog]

The Bouba/Kiki Effect: Synesthesia or Ideasthesia?

Look at the picture of the two shapes below. Which of these is bouba and which is kiki? Most people, around 98%, immediately choose “kiki” for the jagged star-like shape and “bouba” for the rounded curvy shape. This effect has been demonstrated in many languages and for people of all ages - Maurer, Pathman & Mondloch (2006) even showed that infants as young as two-and-a-half showed the Bouba/Kiki effect , despite their lack of reading abilities.

But why should subjects prefer to match one arbitrary shape and sound over another? Many researchers believed that the Bouba/Kiki effect demonstrated that we all show a little synaesthesia, where sensory inputs involuntarily activate an unrelated sensory experience. Synaesthetes often describe numbers as eliciting colours, or colours as being associated with smells. As “bouba” is said with a rounded, relaxed mouth this may become associated with the smooth rounded-edged shape through synesthesia-like sound symbolism, whilst “kiki” is made up of short, sharp syllables and is therefore likely to be linked with the angular, sharp-edged shape.

However, unlike classical synaesthesia, participants in the Bouba/Kiki experiment are associating a sensory input with a semantic label rather than two independent sensory experiences. Nikolic (2009) therefore recently introduced the idea of ideasthesia – a “phenomenon in which activations of concepts (inducers) evoke perception-like experiences (concurrents).” 

Interestingly the words “bouba” and “kiki” are associated with a wider range of stimuli than previously thought. Heston Blumenthal’s Fat Duck Restaurant has an online game in which you are presented with pictures of various food items and asked whether they are “more Kiki or Bouba”. They claim that strong-tasting foods like lemon is consistently rated as more “kiki” whereas blander items like eggs are seen as “more bouba”. The bouba/kiki effect therefore demonstrates the ability of humans to engage in high-level ideasthesia - linking multiple sensory and semantic experiences together in an abstract manner. The dissociation between ideathesia and synaesthesia can also be seen in people with autism who possess above average levels of synaethesia (e.g. Daniel Tammet, the autistic savant who uses his synaesthesia to perform amazing feats of memory), but below average scores on Bouba/Kiki tests.

Which of these is more kiki and which is more bouba?

The bouba/kiki effect is therefore not evidence of synaesthesia per se, but it does demonstrate the remarkable ability of the human brain to link seemingly abstract experiences together in a fairly predictable manner.

Nikolic D (2009). Is synaesthesia actually ideaesthesia? An inquiry into the nature of the phenomenon Proceedings of the Third International Congress on Synaesthesia, Science and Art

[neurofractal-blog]

Brain Connectivity in Autism – Too Much, Too Little or Both?

One of the most influential theories of autism suggests that individuals with Autistic Spectrum Conditions (ASC) have a general deficit in brain connectivity. Evidence from task-based functional connectivity studies over the past decade has more or less supported the disconnection theories (see a nice review by Schipol and colleagues).

    Diagram showing functional under-connectivity between frontal and posterior areas whilst people with autism complete an inferential text comprehension task (taken from Schipol, Keller & Just, 2011).

  Researchers who have used resting-state fMRI to investigate brain connectivity, however, have produced more varied results. In line with under-connectivity theories, Kennedy & Courchesne (2008) found reduced default mode network (DMN) connectivity in a group of adolescent participants with ASC. This has been replicated by other groups investigating the DMN and general brain connectivity using both data driven “independent component analysis” and Region of Interest seed-based analysis. However a study by Di Martino et al., (2011) found hyper-connectivity between the striatum and parts of limbic cortex like the insula and superior temporal gyrus in children and adolescents with ASC. Furthermore Uddin et al., (2013) used rs-fMRI in 20 autistic children aged between 7-11 and found widespread hyper-connectivity within “salience, default mode, motor and visual networks”.

  This confusing picture is compounded by methodological concerns with rs-fMRI. The technique is very sensitive to subject head motion head motion, as first demonstrated by Jonathan Power and colleagues. Even slight movement of the subject’s head can cause the data to show artificial reductions in long-range connections but an increase in short-range brain connections. Furthermore standard motion correction procedures do not seem to eradicate the problem. This is a substantial concern for the field, as participants with ASC are less inclined to sit still in a loud MRI scanner for 20 minutes or more. Another problem is the use of Region of Interest (ROI) analyses in fMRI data processing. If the ROIs do not correctly represent participants with ASC and controls, it can generate a bias, particularly when the brain networks of ASC participants are known to be atypical. Therefore whole-brain “independent component analysis” (ICA) should be the preferred technique for brain connectivity researchers. Despite the methodological concerns surrounding ROI analysis, very few ICA brain connectivity studies have been conducted. Yet another concern for rs-fMRI researchers is the role of physiological differences between ASC participants and typically developing controls. This can influence the BOLD signal generated during scanning and significantly affect the results during data analysis if not controlled for.

  A more general criticism of recent neuroimaging work on autism is that the vast majority of studies have been done on adolescents and young adults who are usually high functioning and male. Whilst the results from these participants are probably the most reliable (less head motion, fewer people on medication etc.), the  findings can bias the field if generalised to ALL people with autism. In the context of brain connectivity this is especially problematic given the large-scale changes to brain connectivity which take place during adolescence. Indeed, brain connectivity studies that have studied younger autistic participants (e.g. Di Martino et al., (2011) and Uddin et al., (2013) discussed earlier) tend to show local brain hyper-connectivity in people with autism. Therefore perhaps the best way to view the confusing field of brain connectivity in autism is from a developmental perspective (see an excellent review article by Uddin and colleagues). That is, in late childhood (around 7-12 years old) people with autism seem to show local hyper-connectivity compared with typically developing children, whilst during adolescence and early adulthood they tend to show hypo-connectivity compared with their peers. This atypical pattern of brain maturation may be sign of reduced synaptic pruning in childhood coupled with reduced stabilisation of core brain networks during puberty, especially the DMN. More work is needed to clarify the exact nature of this dysmaturation and address the methodological concerns surrounding rs-fMRI research. Perhaps the next step for the field is a longitudinal study of brain connectivity in children with autism and typically developing controls.   

Developmental trajectory of brain connectivity in autism. The solid red line shows a scenario in which ASC participants demonstrate a linear dysmaturation in brain connectivity. The dotted line shows a possible quadratic relationship in which ASC brain connectivity drops in adolescence followed by recovery in early adulthood. Taken from Uddin et al., (2013).