Do computers dream of electric glomeruli?

By training a convolutional neural network on more than 20,000 histological images of rat glomeruli, a computer can be taught to recognize them. After training the network using images with positive examples of glomeruli and negative examples of renal fields with no glomeruli present, new images of renal histology can be passed through the network to determine whether they contain glomeruli. Passing the image forward through the network predicts what objects may be present within the image. Passing images in reverse can be used to ascertain what features the network has been trained to find.
Source: Kidney International - Category: Urology & Nephrology Authors: Tags: Nephrology Image Source Type: research