Paper 13329-40
Paper 13329-40
Learning model QPI classification of cell-type in stop-and-go microfluidic chip: Comparing phase and amplitude imaging for lab-on-chip applications
27 January 2025 • 9:10 AM - 9:30 AM PST | Moscone South, Room 311 (Level 3)
Abstract
We developed a microfluidic chip with stop-and-go technology, enabling precise cell halting for quantitative phase imaging (QPI) and classification using machine learning models. This system can image and classify various cell types label-free, with potential lab-on-chip applications like RNA sequencing. We captured QPI images of mouse NIH 3T3, human HEK 293, and HEK 293 cells with lipid deposits, extracting amplitude and phase information. Using convolutional neural networks (CNN), we achieved over 87% accuracy and an F1 score above 95% for HEK 293 with lipids. Phase imaging provided up to 15% higher accuracy than amplitude-only images, demonstrating QPI's superiority.
Presenter
EPFL (Switzerland)
I moved from Venezuela to the USA when I was 17 years-old to complete my BS in Biomedical Engineering at Georgia Tech. After that, I worked in Boston at Shire Pharmaceuticals (currently Takeda) as a process engineer to improve the production of biological drugs to treat rare lysosomal storage disorders such as Fabry disease for 3 years. Then, I had the opportunity to work as a medical device engineer at the biomedical optics laboratory of Prof. Guillermo Tearney, at the Wellman Center for Photomedicine, for 3 years also. At the Tearney Lab, I developed label-free imaging technologies to make disease diagnostics such as the OCT capsule. Also, I built a hand-held spectrally encoded confocal microscope to detect skin cancer. Most recently, I have been conducting my PhD at the EPFL, in Switzerland, under the supervision of Prof. Bart Deplancke and Prof. Demetri Psaltis, working to apply label-free QPI to single-cell imaging for lab-on-chip applications such as RNA sequencing.