Search
Presentation Details
| Poster #227 Extracting Smell Disorders as Early Indicators of Neurodegenerative Diseases Using Natural Language Processing |
| Evan Guerra & Halil Kilicoglu University of Illinois Urbana-Champaign, Champaign, IL, United States |
Smell loss is an important early symptom of neurodegenerative diseases (NDs) and varies across conditions, offering potential as an early indicator of disease progression. Despite ~1,800 PubMed articles on NDs and olfaction, natural language processing (NLP) tools are underutilized for extracting and analyzing ND-olfactory relationships from vast scientific literature. By utilizing NLP methods, we extract olfactory information such as perceived intensity, detection threshold, and identification, analyze diagnosed smell disorders, and assess current NLP limitations in this domain. Using PubMed abstracts on five NDs (Alzheimer’s, Parkinson’s, multiple sclerosis, dementia, progressive supranuclear palsy), we annotated 497 abstracts for five entities: olfactory dysfunction, disease, smell test, odorant, and perceiver, as well as relationships (positive correlation, negative correlation, and association). Our NLP system utilized PubMedBERT and achieved an F1 score of 0.86, Precision of 0.91, and Recall of 0.81 on extracting olfactory-disease relationships. Alzheimer’s was the most mentioned ND, with anosmia as the most frequent olfactory entity. Strong connections were identified between anosmia and dementia, while Alzheimer’s and Parkinson’s were linked to broader terms like "olfactory dysfunction." These findings reveal gaps in current knowledge sources for olfactory terms and highlight the promise of NLP in advancing understanding of smell-related ND research. More targeted NLP efforts are needed to overcome limitations and deepen insights into ND-olfaction relationships. |