Date of Award
Winter 2024
Document Type
Thesis (Undergraduate)
Department
Computer Science
First Advisor
Saeed Hassanpour
Abstract
We introduce a new method for the classification of Gram-stained WSIs. As a test for the diagnosis of blood infections, Gram stains are highly relevant to informing patient treatment. Rapid analysis of Gram stains has been shown to be positively associated with better clinical outcomes, indicating the need for better tools to aid in automatic Gram stain analysis. To date, this area of research has been underexplored, with previous studies relying on the manual patch-level annotation of WSIs to generate training data. This is the first application of a transformer-based model to Gram-stain WSI classification, an approach that is far more scalable to large datasets since it avoids the need for manual annotations. Two new Gram stain datasets have been collected for evaluation, and the use of attention maps to extract regions of high diagnostic relevance is explored.
Recommended Citation
McMahon, Jack, "Advancing Clinical Bacterial Diagnosis: Gram-Stained Whole-Slide Image Classification with Attention-Based Deep Learning" (2024). Computer Science Senior Theses. 34.
https://digitalcommons.dartmouth.edu/cs_senior_theses/34
Included in
Bacterial Infections and Mycoses Commons, Biomedical Informatics Commons, Computer Sciences Commons, Medical Microbiology Commons
