Research groups at the University of Tokyo, Osaka University, and ThinkCyte, Inc. analyze and discriminate cells that are difficult to distinguish with the human eye at ultra-high speed and with high accuracy, and then analyze and discriminate the cells at ultra-high speed (conventional microscope method). We have developed a system for sorting at a rate of XNUMX times or more.
The system, dubbed "ghost cytometry," can use the "eyes" of a machine to identify and selectively separate cells in real time at speeds of thousands to millions of cells per second.
"If you want to analyze image information without human intervention, you don't need images."From here, we proposed the concept of "making a machine learning model directly discriminate cell morphology information without creating an image (a data method for human recognition)".By machine learning the cell morphology data without imaging it, imaging processing for ultra-high-speed real-time discrimination has become possible.
Furthermore, by fusing this imaging method with microfluidic cell sorting technology, 1) high-speed cell fluorescence images are measured, 2) real-time analysis is performed by machine learning, and 3) selective sorting in microfluidic. , The world's first "machine learning-driven real-time cell separation technology" has been realized.
Until now, cell classification, separation and sorting based on morphological observation have been performed based on human experience and cognitive ability, but the speed and accuracy have been limited.With this result, it will be possible to evaluate, select, and utilize a large amount of cells in morphology, contributing to medical treatment that requires high safety and reliability, such as blood / body fluid diagnosis, regenerative medicine, and cell therapy. There is expected.
In addition, the researchers have established a venture company (ThinkCyte, Inc.) beyond the boundaries of academia and organizations, and have achieved this through joint research, which is also evaluated as a pioneering case.
Paper information:[Science] Ghost Cytometry