SYSTEMS AND METHODS FOR AUTOMATIC GENERATION OF TRAINING SETS FOR MACHINE INTERPRETATION OF IMAGES

• US 20190080146A1
• Filed: 09/13/2017
• Published: 03/14/2019
• Est. Priority Date: 09/13/2017
• Status: Active Grant

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Santamaria-Pang, Alberto, Li, Qing, Sui, Yunxia, Dylov, Dmitry Vladimirovich, Sevinsky, Christopher James, Marino, Michael E., Gerdes, Michael J., Meyer, Daniel Eugene, Ginty, Fiona, Sood, Anup
• Granted Patent: US 10,423,820 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06T 2207/10024 Color image
• CPC Class Codes: G06T 2207/10056 Microscopic image
• CPC Class Codes: G06T 2207/20104 Interactive definition of region of interest [ROI]
• CPC Class Codes: G06T 2207/30024 Cell structures in vitro; Tissue sections in vitro
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/33 using feature-based methods
• CPC Class Codes: G06V 10/774 Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06V 20/698 Matching; Classification
• CPC Class Codes: G06V 40/172 Classification, e.g. identification

Abstract

The subject matter of the present disclosure generally relates to techniques for image analysis. In certain embodiments, various morphological or intensity-based features as well as different thresholding approaches may be used to segment the subpopulation of interest and classify object in the images.

Claims

1. A system, comprising:

a processor;

a memory storing instructions configured to be executed by the processor, the instructions comprising instructions to;

receive image data;

generate an object identifier image from the image data;

extract a feature vector that contains intensity and shape information related to the object identifier image from the image data;

identify individual objects within the object identifier image;

associate a portion of the image with each individual object;

generate a feature vector for each individual object based on intensity values and shape descriptors associated with the portion of the image, such that the individual objects are associated with respective feature vectors;

identify a set of object groups within a distribution of the feature vectors;

identify a set of threshold values separating the object groups; and

assign a subset of individual objects to only one of the object groups using the set of threshold values and the feature vectors.

2. The system of claim 1, comprising instructions to:

receive a second set of images of a second sample; and

assign each individual object in the second sample to only one of the plurality of groups using the feature vectors.

3. The system of claim 1, where in the individual objects are cells.

4. The system of claim 3, wherein a cell class is determined from a plurality of individual images with pixel values corresponding to the intensity in different images.

5. The system of claim 1, wherein the image data are from medical imaging techniques (e.g., MR, CT, US, PET, SPECT)

6. The system in claim 1, wherein the image class population distribution is determined using a clustering algorithm.

7. The system of claim 6, wherein the clustering or association algorithm is fit to two probability distribution functions.

8. The system of claim 7, wherein the two probability distribution functions represent a probability distinguishing functions for an object being positive or negative with respect to the channel.

9. The system of claim 1, wherein the image data comprises pixels.

10. A method comprising:

acquiring a set of images, wherein the set of images comprises a plurality of parameters;

identifying individual objects within the set of images using the plurality of parameters;

associating a subset of pixels with each individual object;

integrating one or more intensity and shape feature vectors of the subset for each individual object in each of the one or more images to generate one or more multi-parametric feature vectors for each individual object;

identifying a first group and a second group within a distribution of each of the one or more multi-parametric feature vectors;

identifying a probability threshold separating the first group and the second group;

assigning each individual object to only one of the plurality of groups using the one or more feature vectors and the probability threshold.

11. The method of claim 10, wherein the set of images comprises camera images and wherein the plurality of parameters comprises facial recognition features.

12. The method of claim 10, wherein the plurality of parameters comprises a plurality of intensity-based images.

13. The method in claim 10, wherein the distribution is determined using a clustering algorithm.

14. The method of claim 13, wherein the clustering or association algorithm is fit to two probability distribution functions.

15. The method of claim 14, wherein the two probability distribution functions represent a probability for an object being positive or negative with respect to the individual parameter.

16. The method of claim 10, wherein the subset of pixels has a maximum or minimum size defined by a user.

17. A method comprising:

receiving a set of multiplexed images of a sample comprising a plurality of cells, wherein the set of multiplexed images comprises;

cell identifier image data of a cell identifier image, the cell identifier image data representative of binding of a cell identifier signal generator to a cell identifier target in the sample; and

cell marker image data of a cell marker image, the cell marker intensity data representative of binding of a cell marker signal generator to a cell marker target in the sample, wherein the cell marker image data comprises a cell marker intensity value for each pixel of the cell marker image;

identifying individual cells within the sample using the cell identifier image data;

associating a subset of pixels with each individual cell;

integrating the marker intensity value across the pixels of the subset for each individual cell to generate an integrated marker intensity value for each individual cell;

identifying a first group and a second group within a distribution of the integrated marker intensity values of the cells;

identifying a threshold integrated marker intensity value separating the first group and the second group; and

assigning each individual cell to only one of the plurality of groups using the integrated cell marker intensity value.

18. The method of claim 17, comprising instructions to:

receive a second set of multiplexed images of a second sample; and

assign each individual cell in the second sample to only one of the plurality of groups using the integrated cell marker intensity value.

19. The method of claim 17, wherein a marker cell class is determined from a plurality of individual images with pixel values corresponding to the intensity of different markers.

20. The method in claim 17, wherein the marker class population distribution is determined using a clustering algorithm.

21. The method of claim 20, wherein the clustering or association algorithm is fit to two probability distribution functions.

22. The method of claim 21, wherein the two probability distribution functions represent a probability for a cell being positive or negative with respect to the marker.