By Margaret Y Mahan
In a keynote at the 7th annual Neuroinformatics Congress in Leiden, the Netherlands on August 25, Dr. Michael Milham discussed how resting state fMRI (R-fMRI) has emerged as an approach for psychiatric biomarker identification by enabling the uncovering of human connectome variations that are associated with diagnostic status.
However, as Dr. Milham stressed in his talk, even as it enters the mainstream, this approach still faces many challenges. He reviewed these challenges, provided potential solutions, and offered concepts for discovery and validation of biomarkers that can be expanded to alternative modalities, such as magnetoencephalography or diffusion weighted imaging.
With principal interests in clinical neuroscience and pediatric neurology, Dr. Milham seeks to discover the neural basis of psychiatric disorders by looking at functional and structural connectivity in the brain. Currently, he is the founding director for the Center for the Developing Brain at the Child Mind Institute in New York, NY, and the Deputy Director of Human Imaging at the Nathan S. Kline Institute for Psychiatric Research. He is also cofounder of the 1000 Functional Connectomes Project (FCP), which was followed by the International Neuroimaging Datasharing Initiative (INDI). In his research and advocacy work, Dr. Milham has been a driving force behind the push for research aimed at biomarker identification in the human brain using MRI technologies.
How can biological tests for psychiatry be created?
Dr. Milham acknowledged two major obstacles to further advances in biological psychiatry: the lack of a biological gold standard in diagnostic prediction, and growing concerns regarding the validity of the current categorical classification system. But he also laid fault for lack of progress on the current research culture; i.e., significance chasing with underpowered studies, approximate replications, and extreme comparisons.
He pointed to recent open science initiatives, the big data model, and Research Doman Criteria Project (RDoC) as presenting possible solutions to these challenges. RDoC, for example, could lead to a neuroscience-based classification system for psychiatry, since one of its aims is to enable researchers to rethink the scientific approach by letting the data guide them to ask a range of questions based on big data instead of traditional hypothesis testing.
What Exactly Do We Mean by Biomarkers?
Dr. Milham defined a biomarker as an indicator that allows us to track the status of a process in the brain. He pointed out that these indicators can have clinical utility even if they are difficult to understand from a neuroscientific perspective. In a clinical context, he described a biomarker as an objective test to be used for psychiatric disease diagnosis: for determining the presence or absence of disease, the staging of disease, the patient’s risk or prognosis, and for the prediction and monitoring of clinical response to an intervention.
But will biomarkers be clinically useful? To determine this, Dr. Milham offered, validity, reliability, specificity, and sensitivity first need to be established. Then, and more importantly, he said, widespread availability of a biomarker is necessary for clinical utility to be recognized.
The Case for fMRI Biomarker Identification in Clinical Practice
Particularly in pediatrics, much of the patient’s history comes from parents and teachers. In some cases, patient history alone does not provide enough information for clinicians to provide proper diagnosis and treatment. While the need for treatment may be implied, these discrepancies are often difficult to interpret, which gives rise to the need for biological tests.
“The idea is not to use brain imaging as a means of screening every child,” Milham said. Comparing future uses of psychiatric biomarkers to the diagnostic process in medicine, he pointed to the obvious. “Not every person that walks into their doctor’s office with cold symptoms gets laboratory testing. It is when the diagnosis or next best treatment decision is unclear that we are aiming to have imaging used to facilitate clinical practice.”
Challenges for fMRI Researchers Seeking Clinically Meaningful Biomarkers
In his remarks on research challenges, Dr. Milham noted three main sources of confusion inhibiting the identification of biomarkers.
(1) The significance levels for an effect will be different depending on whether it is clinically useful or just scientifically informative.
(2) Biomarkers are associative but not necessarily causal.
(3) There is no best modality for biomarker discovery.
Another area of confusion is when to deploy a biomarker, given that the current validation techniques can still be confounded.
To overcome these sources of confusion for clinically meaningful biomarker identification, Dr. Milham said we should be more critical with current tests, namely regarding validity and reliability. This was an area he and his coauthors investigated in a 2011 PLOSONE paper, titled Graph Theoretical Analysis of Functional Brain Networks: Test-Retest Evaluation on Short and Long-Term Resting State Functional MRI Data. PLoS ONE 6(7): e21976. doi:10.1371/journal.pone.0021976).
Zeroing in on this topic in his INCF keynote, Milham offered further practical insights into how researchers can establish validity and reliability.
For validity tests:
- Scanner standardization will not suffice since there is also the need to control for other variables, such as sleep and substance use (e.g., caffeine intake, prescription medications, illicit substances), which may impact results. And, correcting for these and other artifacts (i.e. head motion) is hardly universal.
- Need to develop highly consistent networks from independent modalities, such as R-fMRI and magnetoencephalography (MEG).
For reliability tests:
- Test-retest is not enough since artifacts can be reliable too.
- For establishing reliability, need to use openly available datasets from test-retest studies, or initiatives such as the Consortium for Reliability and Reproducibility recently released via INDI as a possible avenue.
The Aims of Collaborative Research
In the last portion of his talk, Dr. Milham discussed neuroimaging using big data in terms of acquiring massive datasets with standardized measurements and sampling the multiple –omics levels in a single dataset. These datasets need data intensive hardware resources, quality control metrics, and data driven exploratory techniques. With the cost of big data neuroimaging, he stressed that data sharing is a must and standardization is essential.
In the near future, Dr. Milham suggests, collaborative research should aim to: (1) give the processing pipelines a common challenge, (2) give the databases a common challenge, and (3) use comparative connectomics.
Overall, the keynote by Dr. Milham provided valuable insight into the need for biomarkers in psychiatry and the roles of big data, open science, and data sharing in their discovery.
For more information on FCP and INDI, I encourage you to visit their website. The NITRC houses neuroinformatics tools, data, and resources providing a community platform for open science and data sharing.
The views expressed in this post belong to its author and are not necessarily those of PLOS.
Margaret Y. Mahan is a Clinical Research Physiologist at the Brain Sciences Center in Minneapolis while also working on a PhD in Biomedical Informatics and Computational Biology at the University of Minnesota. Her focus is on developing analytical methods for integrating data from multimodal sources in order to understand and identify normal brain mechanisms and their alteration in brain diseases.