Organ identification using AI
A computer-implemented method of identifying a tissue type in digital histological images of human or animal tissue comprises training a convolutional neural network (CNN) to identify a particular target tissue type in a plurality of training data sets of digital histological images, inputting a test data set of digital histological images into the trained CNN, and receiving as an output result of the CNN a probability value that the inputted test data set corresponds to the target tissue type. A training procedure of the CNN comprises performing with training data sets: selecting a target tissue area of the training data set, dividing the target tissue area into different sets of tiles of constant size but having different image magnifications, and inputting the sets of tiles into a multi-headed CNN, wherein the sets of tiles having different image magnifications are processed in parallel and the features of the sets of tiles are concatenated.
1 . A computer-implemented method of identifying a tissue type of an organ in digital histological images of human or animal tissue, the method comprising:
training a convolutional neural network to identify a particular target organ tissue type in a plurality of training data sets of digital histological images of human or animal tissue;
inputting a test data set of digital histological images of human or animal tissue into the trained convolutional neural network; and
receiving as an output result of the convolutional neural network a probability value that the inputted test data set corresponds to the target organ tissue type,
wherein the training of the convolutional neural network comprises:
selecting a target organ tissue area of a training data set of the plurality of training data sets;
dividing the target organ tissue area into a first set of tiles of constant size and having a first image magnification and dividing the target organ tissue area into at least one second set of tiles of constant size and having a second image magnification different from the first image magnification, wherein dividing the target organ tissue area into the first set of tiles and the at least one second set of tiles comprises:
extracting a foreground mask of the target organ tissue area;
providing annotations classifying areas of the target organ tissue area; and
merging the annotations with the foreground mask;
inputting the first set of tiles and the at least one second set of tiles into the convolutional neural network, wherein the convolutional neural network is an at least two-headed convolutional neural network in which the first set of tiles and the at least one second set of tiles are processed in parallel whereby features of the first set of tiles and the at least one second set of tiles are concatenated; and
labelling output results of the convolutional neural network with respect to the target organ tissue type.
2 . The method of claim 1 , wherein the size of the tiles of all the sets of tiles are identical.
3 . The method of claim 1 , wherein one or more centroids of the first set of tiles and the at least one second set of tiles are identical.
4 . The method of claim 1 , wherein the training data sets and test data set of digital histological images of human or animal tissue are whole slide images.
5 . The method of claim 1 , wherein the target organ tissue type is one of identified tissue types for tissues of different organs.
6 . The method of claim 1 , wherein the first set of tiles and the at least one second set of tiles correspond to image magnification factors of 1.25, 5, and 10.
7 . The method of claim 1 , further comprising applying a binary training model for annotation of the target organ tissue type.
8 . The method of claim 1 , wherein the training of the convolutional neural network comprises random horizontal and/or vertical flips of the first set of tiles and the at least one second set of tiles.
9 . The method of claim 1 , wherein the training of the convolutional neural network comprises training the convolutional neural network with variations of one or more of a color, hue, saturation, brightness and/or contrast of at least one of the first set of tiles and the at least one second set of tiles.
10 . A non-transitory recording medium readable by a computer and having recorded thereon a computer program including instructions for:
training a convolutional neural network to identify a particular target organ tissue type in a plurality of training data sets of digital histological images of human or animal tissue;
inputting a test data set of digital histological images of human or animal tissue into the trained convolutional neural network; and
receiving as an output result of the convolutional neural network a probability value that the inputted test data set corresponds to the target organ tissue type,
wherein the training of the convolutional neural network comprises:
selecting a target organ tissue area of a training data set of the plurality of training data sets;
dividing the target organ tissue area into a first set of tiles of constant size and having a first image magnification and dividing the target organ tissue area into at least one second set of tiles of constant size and having a second image magnification different from the first image magnification, wherein dividing the target organ tissue area into the first set of tiles and the at least one second set of tiles comprises:
extracting a foreground mask of the target organ tissue area;
providing annotations classifying areas of the target organ tissue area; and
merging the annotations with the foreground mask;
inputting the first set of tiles and the at least one second set of tiles into the convolutional neural network, wherein the convolutional neural network is an at least two-headed convolutional neural network in which the first set of tiles and the at least one second set of tiles are processed in parallel whereby features of the first set of tiles and the at least one second set of tiles are concatenated; and
labelling output results of the convolutional neural network with respect to the target organ tissue type.
11 . A processing device comprising a storage unit having stored thereon instructions causing the processing device to:
train a convolutional neural network to identify a particular target organ tissue type in a plurality of training data sets of digital histological images of human or animal tissue;
input a test data set of digital histological images of human or animal tissue into the trained convolutional neural network; and
receive as an output result of the convolutional neural network a probability value that the inputted test data set corresponds to the target organ tissue type,
wherein the training of the convolutional neural network comprises:
selecting a target organ tissue area of a training data set of the plurality of training data sets;
dividing the target organ tissue area into a first set of tiles of constant size and having a first image magnification and dividing the target organ tissue area into at least one second set of tiles of constant size and having a second image magnification different from the first image magnification, wherein dividing the target organ tissue area into the first set of tiles and the at least one second set of tiles comprises:
extracting a foreground mask of the target organ tissue area;
providing annotations classifying areas of the target organ tissue area; and
merging the annotations with the foreground mask;
inputting the first set of tiles and the at least one second set of tiles into the convolutional neural network, wherein the convolutional neural network is an at least two-headed convolutional neural network in which the first set of tiles and the at least one second set of tiles are processed in parallel whereby features of the first set of tiles and the at least one second set of tiles are concatenated; and
labelling output results of the convolutional neural network with respect to the target organ tissue type.