Solid State Technology and Devices Seminar: Deep Cytometry and Other Applications of Time Stretch Instruments

Seminar | April 5 | 1-2 p.m. | Cory Hall, The Hogan Room, 521

 Professor Bahram Jalali, Department of Electrical and Computer Engineering, UCLA

 Electrical Engineering and Computer Sciences (EECS)

Deep learning has achieved spectacular performance in image and speech recognition and synthesis. It is naturally
suited for applications where large amounts of data are available. Pioneered and advanced in the last 20 years, high
throughput instruments based on the Photonic Time Stretch have established record realtime measurement speed in
spectroscopy, interferometry, OCT, and imaging flow cytometry. Time stretch instruments have led to the discovery
of new scientific phenomena in nonlinear dynamics and to qualitatively different instrumentation modalities. They
generate approximately 1 Tbit/s of measurement data and are ideal for use with deep learning. In our laboratory, we
have shown that high-throughput label-free cell classification with high accuracy can be achieved through a
combination of time stretch with microfluidics for finding cancer cells in the blood. Such a technology holds
promise for early detection of primary cancer or metastasis. In this talk, I will provide an overview of the technology
and describe a new implementation of deep learning in the context of time stretch flow cytometry which avoids data
pre-processing. The new network directly identifies the cells from the raw 1D time stretch waveforms hence
obviates the computationally costly process of image formation and feature extraction. The improvement in
computational efficiency makes it ideal for cell sorting via deep learning.

 dadevera@berkeley.edu, 510-642-3214