Kolmogorov Embedding Kernel for Spatiotemporal Pattern Metrics in 5-D Live Cell & Tissue Microscopy

Wednesday, October 9, 2024

2:30 PM-4:00 PM

BIOMED Seminar

Title:
A Kolmogorov Embedding Kernel for Spatiotemporal Pattern Metrics in 5-D Live Cell & Tissue Microscopy
    
Speaker:
Andrew Cohen, PhD
Associate Professor
Dept. of Electrical and Computer Engineering
College of Engineering
Drexel University

Details:
We show a new semi-supervised learning framework for microscopy movies as a tool for discovery of patterns in live cell & tissue. The normalized compression distance (NCD) provides a metric distance between 5-D (𝑥,𝑦,𝑧,𝑡𝑖𝑚𝑒,𝑐ℎ𝑎𝑛𝑛𝑒𝑙) image pairs. The theory is Kolmogorov complexity, an absolute or universal probability based on statistics of lossless file compression. Multi-scale 3-D image filters enhance blob, plate, tube -like structures based on a priori structure knowledge. The compression is a 3-D free lossless algorithm (FLIF) that uses decistion tree encoding to identify patterns of spatio-temporal structure. This NCD is a reproducing kernel for a Hilbert space (RKHS). An embedding is any representation of the N input movies as points in an N-dimensional space.

The key advantage of the RKHS over all other embeddings is that the RKHS preserves optimally the extrinsic geometry, the space outside of the embedded points, as defined by the metric distance. This allows the use of simplified learning machines with improved generalization capabilities in the RKHS. One application uses pairs of retinal OCT images to measure patterns of progression of, for example, glaucoma and towards the broader field of oculomics. Another application shows the improved visualization and RKHS embedding of patterns of bio-mechanical signaling dynamics in human breast epithelial cells in the context of cancer-associated mutations and / or optogenetic perturbation with markers including ERK, calcium, etc. We’ll talk briefly about open-source software, hardware requirements and finish with a call for collaboration.

Biosketch:
Andrew R. Cohen, PhD, is an associate professor in the department of Electrical & Computer Engineering at Drexel University in Philadelphia, PA, USA. Prior to joining Drexel, he was on the faculty of Electrical Engineering and Computer Science at the University of Wisconsin, Milwaukee. Dr. Cohen received his PhD and MS in computer engineering, and his BS in electrical engineering from the Rensselaer Polytechnic Institute in Troy, NY, USA.

Dr. Cohen’s research comprises using computational analysis of time-lapse microscopy images to understand dynamic behaviors in disease and development. His work integrates approaches in image processing, distributed software architecture, algorithmic information theory and machine learning to formulate insights into how living systems undergo change. Dr. Cohen was previously employed as a microprocessor product engineer at Intel Corp., and as a software design engineer in the operating systems group at Microsoft. He is a senior member of the IEEE.

Contact Information

Carolyn Riley
cr63@drexel.edu