Simultaneous feature extraction and dictionary learning using deep learning architectures for characterization of images of heterogeneous tissue samples

A method for simultaneous feature extraction and dictionary learning comprising:

• training an autoencoder to jointly learn a set of descriptive features and a dictionary of representative atoms;
• receiving an image at an input to the autoencoder;
• and
• processing the image by the autoencoder to assign at least one representative atom of the dictionary to the image,
• wherein processing the image by the autoencoder comprises tiling a patch of the image in subpatches and keeping only a highest activation value per subpatch, without changing a size of an output of an encoding portion of the autoencoder.

A method comprising:

• receiving a first image, wherein the first image comprises a patch tiled in subpatches;
• training an autoencoder with the first image to jointly learn a set of descriptive features and a dictionary of representative atoms;
• receiving a second image;
• and
• processing the second image by the autoencoder to assign at least one representative atom of the dictionary to the second image,
• wherein training the autoencoder comprises keeping only a highest activation value per subpatch, without changing a size of an output of an encoding portion of the autoencoder.