CPC G01N 15/1463 (2013.01) [B01L 3/502761 (2013.01); B07C 5/3425 (2013.01); G01N 15/1404 (2013.01); G01N 15/1425 (2013.01); G01N 15/1434 (2013.01); G01N 15/1459 (2013.01); G01N 15/1484 (2013.01); B01L 2200/0636 (2013.01); B01L 2200/0652 (2013.01); B01L 2300/0816 (2013.01); B01L 2300/0864 (2013.01); B01L 2400/0454 (2013.01); G01N 2015/1006 (2013.01); G01N 2015/1081 (2013.01); G01N 2015/149 (2013.01); G01N 2015/1413 (2013.01)] | 26 Claims |
1. A self-aware particle manipulation system which learns to identify target particles, comprising:
a microfabricated particle sorting device, wherein the microfabricated particle sorting device moves in a plane, and is disposed in a microfluidic channel network having a sort channel and a waste channel, through which a sample stream flows, wherein the microfluidic channel network is also formed in the plane, and wherein the sample stream includes target particles and non-target material;
an optical interrogation device upstream of the microfabricated particle sorting device which identifies target particles, whereby the target particles are diverted into the sort channel by the microfabricated particle sorting device based on sort parameters;
an optical confirmation device downstream of the microfabricated particle sorting device, wherein the optical confirmation device uses a camera to generate new images of the sort channel in the microfluidic channel network;
at least one controller, wherein the at least one controller is configured to adjust the sort parameters and control the microfabricated particle sorting device and the at least one controller further comprises:
a classifier algorithm; and
an analyzer algorithm,
wherein the at least one controller is programmed to:
control the classifier algorithm to identify target particles in the new images using an automated image-based particle detection algorithm, wherein the algorithm is based on a plurality of pre-existing images; and
control the analyzer algorithm to determine the current sorting performance from the new images and the output of the classifier and determine how t improve the sorting performance.
|