Histological image analysis
AssigneeOslo Universitetssykehus
InventorsOle Johan Skrede, Tarjei Sveinsgjerd Hveem, John Robert Maddison, Havard Emil Greger Danielsen, Knut Liestol
Life of this patent
Estimating the term…
Abstract
A machine learning algorithm is trained on a number of microscopic images and a measure of outcome of each image. Each image is divided into tiles. The measure of outcome is assigned to each tile of the image. The tiles are then used to train the machine learning algorithm. The trained algorithm may then be used to evaluate images.
Claims20
A computer-implemented method of training a machine-learning algorithm to perform histological image analysis, comprising:
- obtaining, by a computer, a plurality of histological images and a measure of outcome for each of the images;
- dividing each of the plurality of histological images into a plurality of tiles;
- training a machine-learning algorithm to generate a score for an image tile using as training inputs all of the plurality of tiles of the histological images;
- combining the plurality of scores for the plurality of tiles in a histological image to generate a single predicted outcome value for the histological image;
- optimising the way in which the plurality of scores are combined in order to minimise the variance between the single predicted outcome value and the respective measure of outcome for the plurality of histological images, and wherein:
- training the machine-learning algorithm comprises training the machine-learning algorithm with images from a first image scanning equipment and modified copies of the images to jointly minimize the variance between the single predicted outcome value and the respective measure of outcome for the plurality of histological images.
A computer-implemented method of training a machine-learning algorithm to perform histological image analysis, comprising:
- obtaining, by a computer, a plurality of histological images of a same object using at least two different pieces of image scanning equipment and a measure of outcome for each of the images;
- aligning the plurality of histological images obtained using at least two different pieces of image scanning equipment such that a same object in images provided by the at least two different pieces of image scanning equipment is aligned;
- dividing each of the plurality of histological images into a plurality of tiles;
- training a machine-learning algorithm to generate a score for an image tile using as training inputs all of the plurality of tiles of the histological images;
- combining the plurality of scores for the plurality of tiles in a histological image to generate a single predicted outcome value for the histological image;
- and
- optimising the way in which the plurality of scores are combined in order to minimise the variance between the single predicted outcome value and the respective measure of outcome for the plurality of histological images, wherein:
- training the machine-learning algorithm comprises jointly minimizing the variance between the single predicted outcome value and the respective measure of outcome for the plurality of histological images.
Dependencies are inferred from the claim text.