Highlights from AAN’s Annual Meeting
Members from the CSAND Lab recently attended The American Academy of Neurology’s (AAN) annual conference. AAN is the world’s largest association of neurologists and neuroscience professionals, with more than 40,000 members. The association is dedicated to promoting high quality patient-centered neurologic care, with its main goals being growing a diverse neurology workforce, strengthening neurological practice, promoting neuroscience research, and improving overall neurologic health.
AAN included a variety of excellent presentations and talks, but here we will focus on our lab’s presentations:
Our lab’s recent presentations at the 2024 AAN annual conference discussed more promising insights into early Alzheimer’s Disease (AD) detection and mood assessment in neurocognitive disorders. CSAND Lab director, Dr. Peter S. Pressman, and one of our lab medical students, Gordon Matthewson, shared findings that offer valuable contributions to our understanding of these conditions.
Dr. Pressman’s presentation demonstrated the use of machine learning applied to spontaneous speech for AD and Amnestic Mild Cognitive Impairment (aMCI)1 screening. Analyzing conversational samples from 153 participants, including AD patients, aMCI patients, and healthy controls, we identified speech features indicative of cognitive decline.
By employing a Decision Tree Classifier2 focused on conversation coherence3, we achieved an accuracy of approximately 95.42%4, demonstrating promising potential for rapid and cost-effective screening.
It’s not just the accuracy that matters here though, but the fact that these results were delivered outside of the “black box” often used to describe machine learning techniques. These results are more easily interpretable, and therefore potentially more meaningful, to clinicians, researchers, and ultimately patients and their loved ones. For example, it may be possible to use technology like this to quickly, cost effectively, and accurately check for memory problems in individuals just by analyzing how they talk at routine primary care visits.
Meanwhile, Gordon Matthewson’s presentation investigated mood discrepancies between patients and caregivers in neurocognitive disorders. This work analyzed data from the Standard of Care Rates (SoCRates) database, collected from over 3,000 patient visits at the University of Colorado, Anschutz Medical Campus Memory Disorders Clinic.
The study uncovered correlations between patient-caregiver mood ratings and disease progression. Notably, larger discrepancies in anxiety and depression ratings were associated with specific aspects of caregiver burden and patient functional status.
Stay tuned for further updates as we continue our efforts to push the boundaries of neurology research!
Thank you to AAN for this wonderful opportunity!
Written with assistance from ChatGPT-4.
- Amnestic disorders are a group of disorders that involve loss of memories, loss of the ability to create new memories, or loss of the ability to learn new information ↩︎
- A decision tree classifier is a way for computers to make decisions based on asking a series of questions about the data provided. Each question splits the data into groups based on certain characteristics. For example, in the case of Dr. Pressman’s study, the decision tree might start by asking, “Does the person’s speech sound coherent?” If yes, it might then ask, “Do they use a wide range of vocabulary?” If no, it might suggest a cognitive issue. These questions keep splitting the data until the computer can make a prediction, which in this example, would be about whether someone might have a memory problem or not. ↩︎
- Conversational coherence refers to how well someone’s speech flows logically and makes sense in a conversation. In Dr. Pressman’s study, they looked at features of speech that indicate how well someone can maintain a coherent conversation, including, but not limited to, things like staying on topic, using appropriate grammar, and forming sentences that follow a logical progression. ↩︎
- Another way to think of this is that by analyzing aspects of speech, the decision tree classifier could identify patterns that suggested cognitive decline with approximately 95% accuracy. ↩︎