neuroimaging research: a beginner’s reference guide
One of the main things that I advocate for is science-based activism in the neuro-disorders world.
But I’ve realized that there are barriers to instituting this approach.
Namely: scientific misunderstanding and miscommunication.
There are so many misconceptions floating around the neuro-disorders sphere. Those based on brain connectivity neuroimaging studies alone are too numerous to be adequately addressed in just one article - I barely know where to start within that relatively narrow subspecialty alone, let alone the broader areas of all neuroscience or psychology technical approaches.
How did this happen?
Broadly speaking: specialists aren’t very good at speaking in language that most nonspecialists can understand. This is true for many fields. Computational neuroscience is certainly no exception.
Becoming a neuroscientist takes an incredible level of training. To even become a competent non-doctor research assistant takes years of training on the technology and methods used. In order to run a lab a person must go way beyond this - they need a PhD, an MD, or both.
During this time of training, neuroscientists move further into a world of deeply layered, complex, and (to them, anyway) enthralling ideology and debate. This world has its own set of language, social politics, etc. that diverge in significant ways from those of the regular world. And unlike the world of clinical neurology, that of neuroscience infrequently intersects with a demographic which requires communication via non-specialized language. On top of all of that, the nature of actual job descriptions within neuroscience tends to attract people whose intelligence is weighted towards the analytical rather than verbal. Computational neuroscientists are talented at coding complicated mathematical programs to mine information, and at telling a story with data, graphs, and figures. An ability to accurately translate those complex ideas into simple language isn’t a job requirement.
Yet, the average patient in many neurological disorder populations is asked to understand this highly technical niche scientific writing without the benefit of any of the training or background of the target audience. This leads to a LOT of misunderstandings.
I can’t address everything here (this would be a book, not an article), but let me attempt to clear up some common points of confusion:
Types of neuroimaging techniques
There are several different types of neuroimaging techniques. The ones most commonly used in contemporary neuroimaging research are Computerized Tomography (CT), Positron Emission Tomography (PET), and Magnetic Resonance Imaging (MRI).
Computerized Tomography
Computerized tomography is a structural imaging method in which many x-rays are taken of a body region from all angles and then combined by a computer into a three-dimensional image.
Positron Emission Tomography
Positron emission tomography is designed to measure metabolic activity of tissue. It involves the consumption, injection, or inhalation of a radioactive tracer and subsequent measurement of radioactive signal from body tissue. Depending on the exact tracer administered, PET can be used to measure different metabolic processes in different types of tissue. In neuroscience studies it is often used as a way to measure neuroinflammation.
Magnetic Resonance Imaging
Magnetic resonance imaging is the technique I can speak about in the greatest detail, since MRI brain connectivity research is the technique that I’ve had experience working with.
MRI is a technique where a magnetic field is generated that is able to interact with molecules within the human body. The exact nature of the magnetic field can be changed in order to cause different types of molecules to react in different ways. In this manner, signal coming from different types of body tissues are able to be detected, captured, and analyzed to produce detailed photograph-like images (on their own, these images are sometimes called “structural MRIs”, or sMRI). There are many different applications of MRI which are commonly seen in research - they are all based on the same basic principle, and are achieved by combining different manipulations of the magnetic field in one scanning session in order to capture different kinds of images that can be combined through later analysis to reveal more complex data about tissue composition and activity.
Functional Magnetic Resonance Imaging
Functional MRI (fMRI) is achieved by taking an sMRI, then following with a sequence designed to view the activity of molecules commonly found in blood. Because active brain regions generate more blood flow than nonactive brain regions, snapshots captured in rapid succession of blood flow allow for a reasonably accurate quantification of regional brain activity over that specific period of time.
Diffusion Tensor Imaging
Diffusion Tensor Imaging (DTI) involves following an sMRI with a sequence which captures the activity of water molecules. Because water molecules tend to move along axon bundles (long projections from brain cells) in a set direction, analyzing a rapid succession of these water molecule snapshots allows for the general approximation of axon bundle location and orientation. This allows neuroscientists to render detailed 3-dimensional models which show the highest probable locations for physical connections between brain regions.
Neuroimaging studies use different combinations of these MRI approaches.
Common applications of neuroimaging techniques in research
There are a variety of ways that neuroimaging can be applied to studies. At simplest, structural images are analyzed to look for large abnormalities in brain tissue, such as lesions, tumors, cysts, angiomas, etc. Newer technologies allow for more complex analyses. Advanced MRI techniques can now, for example, be used to look for microstructural tissue abnormalities which aren’t visible on standard clinical scans. fMRI can be used to analyze patterns of regional brain activation in order to determine whole-brain connectivity, regional correlations, etc. fMRI and DTI can be combined in order to perform in-depth analyses about the relationships between brain structural and functional changes. PET can be used to look for metabolic changes in brain tissue which can indicate inflammation.
It is worth noting that many techniques being applied to contemporary neurodisorders research are incredibly new, and many of the technologies are still under development and validation stages of research. Much of the research taking place is exploratory, and truly robust bodies of research are not ubiquitous. There are many emerging ideas which can often conflict with each other. None should be taken for granted.
Study design
Not all studies are set up the same way, and this is important to note when interpreting research. There are many types of study designs and many factors which can influence study outcomes, validity, etc. (in undergrad I took a semester-long college course on this topic alone). This is not an exhaustive list of everything to consider for all studies. But, familiarity with these basic categories will help you to consume research in a more informed manner.
Experimental versus nonexperimental
Experimental design:
Perhaps the most important thing to note is that contrary to how science is spoken about in the general population, not all scientific research is experimental research. A true experiment is designed to demonstrate a causal relationship. This type of study involves a control group (a group of participants in which all factors are held exactly the same as in the experimental group, except this group does not receive the experimental treatment). An example that many people would be familiar with is the placebo group in an experimental drug study.
Experimental study designs also take extraordinary measures to limit confounding variables (any events or differences between subject groups that could threaten the ability to say that the experimental treatment caused any between-group differences). This includes blind randomization of participants to control versus experimental groups.
Randomization is often done by a computer program and serves the purpose of neutralizing differences between study participants that could interfere with a study’s result.
When this is done blindly, it means that participants, researchers, or both do not know whether an individual is in the control or experimental group. In a double-blind study - often considered the best design - participants do not know what group they are in, and researchers interacting with patients and performing initial data analysis do not know patient group assignment either. This blinding process is important because it protects against the placebo effect and also removes sources of potential bias.
Quasi-experimental design:
Quasi-experimental design is more or less the same as experimental design, except that participant assignment to experimental and control groups are not randomized. This study design is often used when researchers want to maintain as many standards of experimental study as possible but random assignment of participants is not possible (such as when certain characteristics are pre-determined). In this study design type strong efforts are still made to eliminate other confounding factors. Strong inferences can be drawn from well-designed quasi-experimental designs, but it’s difficult to claim causal relationships with certainty due to the possibility of pre-existing participant differences having caused outcomes.
Nonexperimental design:
Nonexperimental study designs are studies which do not follow the rules set for experimental and quasi-experimental designs. Nonexperimental studies can appear remarkably similar to experimental and quasi-experimental designs, but they lack robust efforts to eliminate confounding variables. These types of studies can be useful, but cannot be used to establish causal relationships.
Correlational Research
In correlational research, variables are measured to obtain a set of scores for each study participant. The measurement data are then analyzed to identify any patterns. This type of research can reveal information about relationships between two factors, but it cannot be used to establish a causal relationship. To use one of the most ubiquitous sayings in science, correlation does not equal causation.
Descriptive Research
Descriptive research involves making observations about study participants. Examples of this type of research are case studies, or scientific reports which document observations about a small group of participants but do not involve elements of experimental designs. These types of studies often do not feature study interventions at all - they are often meant to document data as it is in the “real world”, or to document rare phenomena in a singular patient or small group of individuals. This strategy is valuable in terms of documenting phenomena, but data from these types of studies are not generalizable and no causal relationships can be claimed.
Time factors
Studies are conducted over varying amounts of time.
Longitudinal studies take place over a long period of time and capture changes in the same subjects over the course of months to years.
Cross-sectional studies take place over a shorter period of time.
Choices about study length can be for scientific or other reasons - some studies only need to capture a snapshot in time and would not be aided by a longer course. Other times, budgetary concerns often limit the scope and length of studies.
Sample size and generalizability
It’s important to consider the number of participants in a study. Studies which involve many subjects tend to have results that are more generalizable than studies which involve fewer participants. Studies with very few participants can be useful, but with few exceptions should be followed up with research involving more participants in order to verify results. In neuroimaging studies sample sizes do tend to be small due to the extremely high cost of conducting scans. It’s fairly typical for studies to involve a sample in the tens as opposed to hundreds. Well-conducted smaller-scale studies can bear valid insights. However, there are many neuroimaging studies with small sample sizes evey by neuroimaging research standards which are often quoted as unquestionable sources of information, and it is important to guard against this. A sample size of, say, five individuals can be a good start to getting further funding for a larger study but barring extreme circumstances shouldn’t be held as gospel and generalized to a broad population without follow-up study.
Regions of interest
Regions of interest (ROIs) refer to brain areas which are analyzed in a neuroimaging study. These are typically broken down into incredibly small anatomical increments.
While some studies analyze the whole brain, it’s common for researchers to focus analyses on relatively small subsections of the brain. It’s useful to look for documentation of which brain regions researchers included in initial analyses, because this can indicate important things about the study findings and how they can be interpreted. For example in Functional Neurological Disorder, a study design that finds differences in the amygdala after looking at the entire brain is remarkably different than one which finds the same difference after analyzing only a subset of brain regions.
Confirmation bias and its effect on the scientific process
This is less of a note about research structure itself, and more about the nature of science behind the scenes.
Bias is an inescapable facet of human cognition. Scientists desire to be impartial but still exist within social structures, are taught schemas, and form ideas based on past information. This also applies to consumers of science - you, me, and everyone. This can lead to incorrect assumptions, circular reasoning, and neglect of potentially important revelations. To consume scientific writing at a professional and truly informed level one must understand this and watch for biases in scientific writing and in one’s own reactions to findings. Be on guard for signs of confirmation bias from scientists, and be skeptical of your own reactions to scientific writing since it’s inevitable that you will also experience this phenomenon.
Implications for understanding neuroimaging research
Staying informed about research is an amazing privilege that modern patients have which can make a huge difference in the ability to hold doctors accountable and to make informed decisions about one’s own health and care. However, interpreting research must be done with care. Neuroimaging studies are complex and written in technical language for a highly specialized population.
Pay attention to study design - before assuming a finding means that a phenomenon is caused by a specific factor, make sure the study being referenced is experimental. If it’s not, there’s not a scientific basis for assuming causation.
Pay attention to study size. Was a finding made in a cohort of 1000? 100? 40? 5? These sample sizes would indicate extreme differences in the ability to make certain assumptions about study findings.
Watch for biases - both those of scientists, and yourself. Science can only be properly objective when those participating in the scientific process are aware of the natural human tendency to prefer ideas which do not challenge our present understandings. In science, if an idea is never properly challenged then it’s not really verified. Healthy skepticism is a good thing to foster.
So, with all this in mind...happy researching!