Infection by West Nile Virus can lead to encephalitis, or harmful inflammation of the brain. The immune system is critical for controlling viral replication and spread early in West Nile Virus infection, but persistent immune activation causes encephalitis that can result in brain damage even after the virus has been cleared. Recent pharmacologic advances have produced drugs that modulate the body’s immune response and can control inflammation, but these drugs have not yet been tested in conditions of viral encephalitis. In order for patients to benefit from these therapies, clinicians need tools that help identify when excessive immune activity is causing encephalitis.
The key innovation of this project is the combination of noninvasive imaging with novel immune modulating drugs to improve the diagnosis and treatment of encephalitis. Our central hypothesis is that specialized immune cells known as macrophages are key drivers of encephalitis in West Nile Virus infection, and that preventing their activation will preserve memory and other cognitive functions. Our studies will explore and develop noninvasive positron emission tomography (PET) imaging as a tool for diagnosing brain inflammation. We will test our hypothesis utilizing West Nile Virus infection of mice, which captures the key elements of human disease including encephalitis. This model allows us to evaluate existing diagnostic and therapeutic tools currently used in humans for other purposes, from which we will define new clinical applications. We will thus be poised to translate our findings to human studies defining and treating viral encephalitis.
Alzheimer’s Disease (AD) is a degenerative condition that affected 5.8 million seniors in 2020 and is the sixth leading cause of death in the United States. Detecting mild cognitive impairment, often a precursor to AD, and predicting its advance to AD dementia are key clinical diagnostic problems. Early diagnosis can motivate early intervention with lifestyle changes that build cognitive reserve or reduce comorbidity and thus prolong functional independence. MRI scans and specialized tests for AD-related proteins in spinal fluid or on PET brain scans are available, but it is not known how best to deploy these expensive tests or combine the information from them. New computer-based “machine learning” software tools may provide a solution to these problems.
This project will explore the use of a machine learning technology called deep learning to diagnose the stage of AD and to predict its progression. We will use the data available from the scientifically open Alzheimer’s Disease Neuroimaging Initiative (ADNI) database, which contains MRI, PET, risk genes, cerebrospinal fluid and other data. We will develop a deep learning model that performs its predictions using MRI data alone, and can also augment the MRI data with the other datatypes for improved performance at some expense. Our modern machine-learning methods are designed to be rationally factored in with other individualized clinical information to aid clinicians in these vital diagnostic decisions.
Cognitive therapies help patients by providing ways to modify habitual but unproductive thought patterns, known as maladaptive thinking styles. Cognitive therapies are effective in treating depression, amongst other conditions, and are increasingly delivered remotely as text-based interventions. This trend toward digital delivery has accelerated on account of physical isolation and psychological stressors during the global pandemic. While this means cognitive therapy can potentially reach more patients, the effectiveness of this therapy depends on the ability of a skilled practitioner to recognize types of maladaptive thinking, and there is a critical shortage of mental health practitioners with this expertise.
In radiology, computer-aided diagnosis systems driven by artificial intelligence are used to help physicians detect signs of illness they may otherwise miss. In this project, we will develop a computer-aided detection system to support text-based cognitive therapy. To do so, we will identify indicators of maladaptive thinking styles within a set of text messages exchanged between clients and their therapists, and train neural networks to detect these indicators automatically. The resulting tools will provide a basis for an artificial intelligence-based decision support system to help clinicians recognize and manage maladaptive thinking styles that will enhance the quality and effectiveness of text-based cognitive therapy.
Schizophrenia is a debilitating mental health condition with high societal and personal costs, due largely to chronic difficulties with social and occupational functioning. While classical symptoms of schizophrenia – such as hearing voices – are often responsive to medication, people with schizophrenia also experience difficulties in social cognition, or understanding and interpreting the intentions and emotions of others. Social cognition affects the ability to function in society, and is a key determinant of real-world outcomes in schizophrenia.
Despite its importance, we lack objective and easy-to-deploy instruments to assess social cognition. This measurement gap presents a critical stumbling block for development of interventions to improve social cognition, because the effects of potential treatments cannot be assessed efficiently and at high resolution. Better measurements are also needed to identify individuals likely to benefit from such treatments and monitor treatment effects over time.
This project will develop innovative automated methods to measure a key component of social cognition – the ability to recognize the intentions and emotions of others. The underlying idea is to present a participant with a cue – such as a short video clip intended to be amusing – and then apply computational methods to their spoken response to see if it aligns with the intention behind the cue. The result will be a set of validated measurement tools to facilitate objective, repeatable, and scalable assessment of social cognition. These tools will accelerate our ability to rigorously test new treatments targeting these key deficits impacting people living with schizophrenia.
UW Medicine has amassed detailed patient treatment and business data in its electronic medical record (EMR). This information is a treasure trove that is not used to its full potential for two reasons: 1) For each clinical encounter, only a fraction of the information in the EMR is relevant, and virtually all of the information a clinician engages remains in a format that obscures patterns and trends; and 2) In groups of patients with the same illness, data from the EMR could be used to discern larger trends in the course of the disease or evaluate the effect of practice patterns on patient outcomes. The EMR currently does not provide a way to access this information in an agile way.
We have developed innovative software, “Leaf,” that allows medical providers to access population-based EMR data in real time. Leaf is now used at several academic medical centers nationally. In this project, we will collaborate with the UW Memory and Brain Wellness Center to design and evaluate “dashboards” that visualize how a patient’s history and trajectory compare to other, similar patients. For instance, daily function and cognitive testing data for a person with Alzheimer’s disease, already gathered over the course of several years, could be graphed and compared to the same information from all UW patients with Alzheimer’s disease. We will pilot these dashboards in Leaf and collect patient and provider feedback. We intend to publish our results and make code available as part of the open Leaf platform for rapid dissemination.
Following severe cases of traumatic brain injury (TBI), the brain can swell, leading to elevations in intracranial pressure (ICP). Patients who develop high ICP following severe TBI are more likely to have poor neurologic recovery from their injury, and control of ICP likely contributes to improved outcomes. ICP detection and management is typically guided by invasive monitors placed through the skull and into the injured brain. These devices are highly accurate and reliable, but they are also expensive and expose the patient to rare but potentially serious risks. This is problematic because as few as one-third of patients are found to have elevated ICP, even when the best available evidence is used to guide their placement.
Using ultrasound to measure optic nerve sheath diameter (ONSD) could be an inexpensive, noninvasive and reliable means of monitoring ICP. Located behind the eye, the optic nerve sheath surrounds the nerve carrying visual signals to the brain. Increases in intracranial pressure are transmitted into this conduit, causing it to dilate. Ultrasound-measured ONSD has been shown to correlate with ICP in many neurologic conditions, including TBI, but it has not been systematically evaluated as a screening or a monitoring tool.
This study will routinely measure ONSD in patients undergoing invasive ICP monitoring for severe traumatic brain injury at Harborview Medical Center. The goal is to determine whether ONSD measurement with ultrasound can be combined with readily available clinical data to improve the prediction of elevated ICP, and to assess whether it can be used to monitor ICP during a patient’s hospital stay. If successful, ONSD measurement could have a significant impact on TBI care in both high and low resource settings.
Positron emission tomography (PET) is an imaging technique that uses radioactive substances to visualize and assess the brain function. Apart from its heavy use in clinical oncology, PET is widely used in a variety of other conditions such as various neurological, psychiatric, neuropsychological, and cognitive disorders and is the gold standard for assessing neurodegeneration. In particular, PET is clinically used to distinguish Alzheimer’s disease from other dementias and assess the disease progression. Despite its clinical importance, PET imaging encounters barriers because of limited availability, expense and radiation exposure.
This project seeks to address this barrier to brain health using artificial intelligence to predict PET brain images from magnetic resonance imaging (MRI) data. Such a method would be extremely beneficial in clinical settings because unlike PET, MRI is widely available, non-invasive and relatively inexpensive. The approach essentially turns an MRI scanner into a PET scanner, opening up this technology to sites and applications in which PET is either unavailable or infeasible. Doing so would give millions of people access to initial screens for Alzheimer’s disease, assessment of disease progression and an easy way to monitor treatment.
Lewy body dementias (LBD), a term referring to both dementia with Lewy bodies and Parkinson’s disease dementia, are the second most common type of degenerative dementia in older adults. These are complex disorders in which patients may exhibit disruptive behaviors that make caregiving challenging. Compared to other types of dementias, caregivers of people with LBD report higher stress and more severe depressive symptoms. The ongoing COVID-19 pandemic has multiplied the challenges that caregivers of persons with dementia face in providing care for their loved ones. As such, support interventions for caregivers of persons with LBD are urgently needed.
In this study, we will adapt our online intervention for older adults with frailty to target the unique needs of caregivers of people with LBD. We will conduct participatory design sessions with potential users to determine their needs and priorities specific to LBD and deploy the re-designed intervention in a pilot study focused on usability and efficacy. Through this newly tailored support system, we aim to bolster the health of caregivers as well as their ability to assist care partners living with LBD.
This intervention could potentially be used in conjunction with usual care and/or as a stand-alone module in emergent circumstances, such as the current pandemic, when routine professional interventions may not be readily available. By fostering the development of a community-driven online support system, this project will begin to lay the groundwork for promoting resilience within families affected by the behavioral challenges of dementia.
Traumatic brain injury (TBI) is common in the United States with 2.87 million emergency department visits related to TBI per year. Chronic pain is a frequent complaint following TBI, with more than half of patients reporting pain. Individuals with TBI are often prescribed opioids for pain following their injury, but unfortunately may be especially vulnerable to post-injury alcohol and drug use problems.
Despite increased opioid prescriptions and risk factors for this population, there are no clinical practice guidelines for opioid prescription following TBI and limited published research. The project seeks to address this knowledge gap by using routinely collected clinical data from several different data sources to examine when and how opioids are prescribed following TBI in a community-based population.
This complete picture of opioid prescription following TBI may reveal trends of higher opioid prescription for specific subpopulations or areas of healthcare. Through understanding the trajectory of opioid prescription following TBI, we will be able to identify the scope of the problem and the most appropriate time points for intervention. Ultimately this project will provide the foundation for new approaches to reduce opioid prescription in the clinical management of TBI.
Millions of people lack access to mental health treatment due to barriers such as limited therapist availability, long wait times, high cost, and stigma. The COVID-19 pandemic has problematically increased demand for treatment while decreasing access. Because the internet is widely available, many people first turn to the internet for mental health support, giving rise to massive online psychotherapy, counseling and peer-to-peer support platforms such as Ginger and Talklife. However, not all conversations lead to improvement and may miss opportunities to help or even make things worse as platforms struggle to keep up with the increasing demands and lack methods for evaluating and promoting high-quality conversations.
This project seeks to improve the quality and scalability of online mental health support through real-time, evidence-based conversation feedback. We will leverage and analyze datasets of support interactions and associated outcomes across millions of individuals that use the partnering online mental health platforms at Ginger and Talklife. Our goal is to develop and pilot-test artificial intelligence methods that provide supporters on these platforms with practical just-in-time feedback and training. If successful, at least three benefits will follow our work. First, millions of help seekers using partnering mental health platforms Ginger and Talklife will receive higher quality responses through, for example, an expression of higher empathy. Second, those providing help will gain expertise faster and with less distress. Third, platforms and researchers will discover conversational best practices which can then be used to improve helper training and quality evaluation.
