Machine learning classification and training for digital microscopy images
AssigneeTechcyte, Inc.
InventorRichard Boyd Smith
Life of this patent
Estimating the term…
Abstract
Systems, methods, and devices for training models or algorithms for classifying or detecting particles or materials in microscopy images are disclosed. A method includes receiving a plurality of microscopy images of a specimen and a classification for the specimen. The plurality of microscopy images includes a first image captured at a first magnification and a second image captured at the first magnification with a different focus than the first image. The method includes training a machine learning model or algorithm using the plurality of images, wherein the first image and the second image are provided with one or more labels indicating the classification.
Claims20
A method comprising:
- receiving a specimen sample disposed on a slide, the slide including a slide label indicating a classification for the specimen;
- imaging the specimen sample disposed on the slide;
- receiving a plurality of microscopy images of the specimen sample and classification for the specimen sample, the plurality of microscopy images comprising a first image captured at a first magnification and a second image captured at the first magnification with a different focal depth than the first image, wherein the focal depth for the second image overlaps with a focal depth of the first image;
- training a machine learning model or algorithm using the plurality of images, wherein the training further includes providing the first image and the second image with overlapping focal depths to the machine learning model or algorithm with the classification of the specimen sample;
- labeling the plurality of microscopy images with image labels indicating the classification of the sample specimen;
- and
- associating the slide label with the image labels on the plurality of microscopy images, the slide label further identifying a region of the slide from which the plurality of microscopy images were imaged.
A system for training a neural network, the system comprising:
- one or more processors;
- and
- non-transitory computer readable storage media storing instructions that, when executed by one or more processors, cause the one or more processors to:
- image a specimen sample disposed on a slide, the slide including a slide label indicating a classification for the specimen sample;
- receive a plurality of microscopy images of the specimen sample and the classification for the specimen sample, the plurality of microscopy images comprising a first image captured at a first magnification and a second image captured at the first magnification with a different focal depth than the first image, wherein the focal depth for the second image overlaps with a focal depth of the first image;
- and
- train a machine learning model or algorithm using the plurality of images, wherein the training further includes providing the first image and the second image with overlapping focal depths to the machine learning model or algorithm with the classification of the specimen sample;
- labeling the plurality of microscopy images with image labels indicating the classification of the sample specimen;
- and
- associating the slide label with the image labels on the plurality of microscopy images, the slide label further identifying a region of the slide from which the plurality of microscopy images were imaged.
Non-transitory computer readable storage media storing instructions that, when executed by one or more processors, cause the one or more processors to:
- obtain microscopy images of a specimen sample disposed on a slide having a slide label indicating a classification for the specimen sample, wherein obtaining microscopy image comprises, for one or more regions of the sample:
- capturing a first image of the specimen sample with a first focal depth at a first magnification;
- capturing one or more additional images with one or more additional focal depths at the first magnification, wherein the first focal depth overlaps at least one of the one or more additional focal depths;
- training a neural network using the first image and the one or more additional images, wherein the training further includes providing the first image and the one or more additional images with the overlapping focal depths to the neural network with the classification of the specimen sample;
- labeling the first image and the one or more additional images with image labels indicating the classification of the sample specimen;
- and
- associating the slide label with the image labels on the first image and the one or more additional images, the slide label for the first image including ground truth for the first image and the slide label for the one or more additional images identifies a same region and same specimen captured in the first image and the one or more additional images.
Dependencies are inferred from the claim text.