Authors:

Jack Gauderman

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

In the field of machine learning, a significant challenge lies in acquiring sufficiently large, labeled datasets to train models. Our research addresses this limitation by developing a robust machine learning pipeline that analyzes light microscopy images for respiratory health diagnostics with minimal training requirements. We leverage Micro-Segment Anything Model (SAM), a vision-transformer-based segmentation framework derived from Meta's general purpose SAM and fine-tuned for microscopy applications. When integrated into our proposed computer vision preprocessing system, Micro-SAM enables accurate and reliable characterization of microscopy images for respiratory health diagnostics. Additionally, we have developed methods for cell characterization of light microscopy images by extracting a suite of features to describe cell texture, morphology, and biological attributes. When validated using both object-level and pixel-level ground truth annotations, the pipeline achieves high cell detection accuracy as well as precise cell outline quality. This capability streamlines the screening of light microscopy images and reduces the need for labor-intensive laboratory procedures. This is particularly advantageous in high-stress environments, such as those faced by Air Force pilots, where timely detection of cellular changes is critical for safety. This work demonstrates that, even with limited data, machine learning can deliver reliable diagnostics and enhance efficiency across diverse operational scenarios. Poster #21 Comparison of Respiration Rate Estimation Methods in the Presence of Noise Using a Multimodal Wireless Upper Armband Rachel Kurian

Source:

Texas A&M University / 2025

Topics:

No topics listed

Co-authors:

Jack Gauderman