One of the winners in the first stage of the $20 million “Head Health Challenge” were a team made up of the University of California, San Francisco (UCSF), San Francisco, CA- UCSF and Ayasdi, a Silicon Valley start-up. This collaboration’s research focused on better identifying patients who were more likely to experience persistent symptoms and need additional support following a concussion or mild traumatic brain injury.
Here, UCSF’s Esther Yuh and Adam Ferguson are joined by Ayasdi’s Pek Lum in an interview to discuss the implications their work would have on understanding how injuries impact connections in the brain. They discuss their work in creating life-changing tools, treatments and devices that can play out in real time.
Q1. Your winning proposal was entitled: ‘Topological analysis of diffusion tensor imaging to diagnose and predict outcome following mild traumatic brain injury.’ Could you describe the benefits the research could bring to those who’ve suffered a concussion or mild traumatic brain injury (TBI)?
Adam and Esther: Brain injury is a very complex and highly variable disease. Typically, it’s hard to look at an image of the brain and identify exactly where an injury has occurred, particularly when the patient has a mild concussion. The good news is that there have been many advances in brain imaging technologies in recent years; one of the most powerful is Diffusion Tensor Imaging (DTI). DTI can give us a new way to identify and diagnose mild brain injuries that would not necessarily show up on MRI or CT scan. This is the first study of DTI data from patients with mild traumatic brain injury that will be analyzed with Ayasdi’s platform. We all believe that we have an unprecedented opportunity to discover huge breakthroughs in diagnosing, treating, and potentially preventing adverse effects that can happen after a traumatic brain injury.
Pek: DTI is a very good way to get another view into brain injury, but the output is a massive amount of data that most analytical tools fail to glean insights from. DTI data calls for an analytics platform designed to understand complex and massive datasets—perfectly suited to Ayasdi. We are very excited to see what we will find.
Q2. While the proposal has been aimed primarily at TBIs suffered by the military and athletes, do you think the research conducted have benefits to TBIs associated with automobile accidents and other common accidents that impact many more people?
Adam and Esther: Yes, we do. Even though we’re focused on athletes in this study, we know that mild TBI is common. Up to 1/3 of people sustain a head injury requiring some type of medical attention such as emergency department or clinic visit or hospitalization by their mid-twenties. Approximately 80% of head injuries are mild and most of those people recover without any issues. However, around 15%-20% of them have persistent problems that can last weeks, months, or even years. Those issues can range from somatic symptoms such as headaches to cognitive impairments to emotional problems such as depression. We hope that we can identify and treat the patients that have a high likelihood to go on to have problems after mild TBI.
Pek: Just to underscore that point, if we can find those more sensitive markers, then we can predict outcomes at the time of the fall or hit. This information will radically change how doctors treat those patients—whether they are football players, kids, or seniors.
Q3. What do you think could be the biggest innovation in brain injury research/detection/treatment that you see coming within the next decade or so?
Adam and Esther: If we find what we hope to find, then it will signal a huge amount of education and counseling, particularly for patients. Patients who understand that they are at a higher risk of adverse effects, after a mild concussion, can monitor potential symptoms more closely. And, the patients who are not likely to have adverse effects will be reassured and that alone is therapeutic. From a physician’s perspective, it’s very hard to diagnose and treat mild brain injuries because many patients have pre-existing conditions that can complicate matters. We hope that in 10 years, we can arm doctors not only with information that will indicate which head injury patients will develop issues, but also identify what issues they will be based on where the brain injury occurs. Meaning that if a patient has an injury to a specific part of the brain, then they will be likely to develop a specific ongoing problem down the line. That’s the direction that we’re heading.
Pek: We also hope to define new, more granular ways to categorize brain injury. Instead of mild, moderate, and severe, we believe that there is more of a continuum of injury. With more precise definitions and diagnoses, both patients and doctors have better information to prevent and treat future complications. We hope that with more information will come more collaboration between patient and doctor and greater public awareness about mild traumatic brain injury.
Q4. The winning Ayasdi-UCSF entry used Topological Data Analysis to characterize and classify high-quality detailed MRI and CAT scans of the brain. Could you briefly explain what this kind of data analysis is and what benefits it has over other analytical techniques?
Adam and Esther: First, let us explain what DTI is and why we need a platform like Ayasdi’s to analyze this highly complex dataset. DTI characterizes the ease with which water diffuses along different directions in 3 dimensions, and can be used to assess the integrity of white matter tracts within the brain. Within intact, highly organized white matter tracts, water diffuses much more easily along the direction of the tracts than it does perpendicular to them. This preferred directionality of water diffusion is thought to be altered when the tracts are injured in traumatic brain injury. One of the difficulties in analyzing DTI data is the sheer size of the data sets — more than 100,000 voxels in white matter alone in the human brain using a typical DTI sequence. The data depict extremely complex spatial patterns that are difficult to analyze using linear analytical techniques. That’s where Ayasdi comes in.
Pek: Linear analytical techniques, like regression or singular value decomposition, require that the analyst make a lot of assumptions about the data—that may or may not be true. It usually becomes a needle in the haystack game. Ayasdi, on the other hand, uses Topological Data Analysis (TDA), combined with an ensemble of machine learning techniques, to enable data scientists, domain experts, and business people to discover insights from their data without the need to write code, queries, or ask questions.
Basically, we believe that data has shape and that shape has meaning. We find relevant and similar patterns in the data and create images that depict the shape of the DTI data. That’s how Ayasdi is able to extract novel patterns that wouldn’t be possible any other way.
Esther Yuh, M.D., Ph.D., is Assistant Professor of Radiology and attending neuroradiologist at UCSF and at San Francisco General Hospital. Dr. Yuh was UCSF Department of Radiology’s Outstanding Teaching Fellow of the Year in 2009, and was a 2010 nominee for the UCSF Exceptional Physician Award. Her major clinical and research interests are the improvement of imaging and image analysis for better characterization, classification and treatment of acute TBI patients.
Adam Ferguson, Ph.D., is Assistant Professor of Neurosurgery and Principal Investigator at UCSF’s Brain and Spinal Injury Center. He directs a multidisciplinary research team that combines expertise in anatomy, neurosurgery, physical therapy, psychology, engineering, and data science in pursuit of novel therapeutic approaches for neurotrauma and related disorders. Dr. Ferguson has published over 40 peer-reviewed papers and 86 peer-reviewed abstracts on neurotrauma.
Pek Lum, Ph.D., is Ayasdi’s Chief Data Scientist and VP of Solutions and leads Ayasdi’s products and solutions team. She provides leadership on product, data acquisition and analytics. Her team’s responsibilities include product research and design as well as development of analytical solutions for customers. Her work is widely published in scientific journals, and has contributed to discoveries in drug development and the understanding of complex diseases.