Yiheng Tu, PhD, on Innovations in Migraine Research
The Migraine Research Foundation estimates that migraine affects nearly 39 million adults and children in the United States. With this high prevalence, it is imperative that diagnostic tools are able to detect migraine so that health care providers can treat migraine quickly and accurately.
Yiheng Tu, PhD, who is a research fellow in the Department of Psychiatry at Massachusetts General Hospital in Boston, answered our questions about the innovations happening in migraine research that will help improve diagnostics and treatment. He also presented this topic at the 2019 Migraine Symposium.
NEUROLOGY CONSULTANT: What new innovations in research did you discuss during your session?
Yiheng Tu: In general, I talked about two innovations: neuroimaging and machine learning. Neuroimaging techniques, such as electroencephalography (EEG), magnetic resonance imaging (MRI), and magnetoencephalography (MEG), provide an opportunity to noninvasively investigate the pathophysiology of migraine. For example, I discussed how using MRI can find structural and functional abnormalities in cortical and subcortical brain areas in patients with migraine during different disease phases.
Machine learning is a cutting-edge technique in which computers learn from data and make predictions without being explicitly programmed. There is a growing trend of applying machine learning to neuroimaging data to provide a complete description of the relevant neuropathology of a variety of brain diseases to be clinically actionable. Through training and testing, a machine-learning-based brain model can provide useful information for clinical decision making.
NEURO CON: What was the most common, or most surprising, question you received after this session?
YT: The most important or inspiring question I received was from Dr Frederick Godley. After I introduced our recent work in developing an MRI-based biomarker for migraine that can provide complementary information for clinical decision making, Dr Godley raised concerns regarding additional MRI costs for patients. How to translate research findings from the lab to the bedside is an important and critical question that we need to take into careful consideration.
NEURO CON: What research are you currently working on?
YT: My research projects cover a variety of topics in translational neuroimaging, computational neuroscience, and neural engineering. I am currently building a framework for using neuroimaging and machine learning techniques for clinical diagnosis and predicting treatment responses. I am particularly interested in the inter-individual variability of treatment responses and the customization of treatments based on a patient’s brain measurements—an endeavor central to precision medicine.
NEURO CON: What knowledge gaps still exist in this area?
YT: There are two important knowledge gaps. First, to serve as a biomarker for a disease (eg, migraine), the machine-learning-based brain model should have high sensitivity and specificity. Sensitivity relates to how robustly the model responds to migraine, while specificity relates to whether the model only responds to migraine and not to other conditions. Second, generalizability of brain models is another frontier. Brain models that are useful for translation must be generalized across contexts (eg, laboratories, scanners) and populations (ie, must be able to be applied on new individuals).
NEURO CON: What is the key takeaway for our audience of neurologists?
YT: The trend of big data and the boost in computational capacity provide unprecedented opportunities to close the gap between basic (ie, understanding the neuropathophysiology of brain diseases) and translational studies (ie, developing tools that are useful for clinical decision making and therapeutic development).
For more content on migraine, visit our Resource Center.