Cellavision Ab

Adam Morell

Martin Almers

Sven Hedlund

Kent Stråhlén

The disclosure relates to a method for training a neural network classifier () to classify a digital image depicting one or more objects of a biological sample into a specific class, which class belongs to a predefined set of classes (C-C), the method comprising: providing a training set of digital images (-) originating 5 from a plurality of biological samples, each digital image () of the training set being labeled with a specific class (C) of the one or more objects of the digital image (), each digital image () of the training set being associated with global data () pertaining to the respective sample, training the neural network () using pixel data of each digital image () from the training set of digital 10 images (-) and the global data () associated with said digital image () as input, and using the specific class (C) of the label of said digital image () as a correct output from the neural network (). The disclosure further relates to an analyzing apparatus ().

Method and apparatus for training a neural network classifier to classify an image depicting one or more objects of a biological sample

The present invention relates to a method for changing the appearance of an original image comprising N&#x3e;1 classes of image elements, the method comprising the step of: for each pixel and for at least one subset of the original image: calculating N probability values, each probability value defining a probability for the pixel of belonging to a corresponding one of the N classes of image elements, transforming a color of the pixel by using predetermined color transforms for each of the N classes of image elements and the N probability values for the pixel.

Segmentation based image transform

A method is disclosed for determining a sought object contour in a digital microscope image, which includes a plurality of image elements and reproduces a biological material. The method includes assigning edge values to at least a first subset of the image elements in the image; assigning values of a first gradient vector component whose values each includes a first linear combination of edge values of some surrounding image elements to at least a second subset of the image elements in the image; assigning values of a second gradient vector component whose values each include a second linear combination of edge values of some surrounding image elements to at least a third subset of the image elements in the image; and calculating an estimate of the sought object contour based upon values of the first and second gradient vector components.

Growing community of inventors

Lund, Sweden

Adam Morell