Predicting DCIS recurrence risk using a machine learning-based high-content image analysis approach

Patent metadata

Field	Value
Publication number	US-12175367-B2
Application number	US-201917415260-A
Country	US
Kind code	B2
Filing date	Dec 16, 2019
Priority date	Dec 17, 2018
Publication date	Dec 24, 2024
Grant date	Dec 24, 2024

Abstract

Embodiments of the present systems and methods may provide improved capability to predict the risk of recurrence of ductal carcinoma in situ (DCIS) conditions using whole slide image analysis based on machine learning techniques. For example, in an embodiment, a computer-implemented method for determining treatment of a patient may comprise receiving an image of living tissue of a patient, annotating the entire image into tissue structures, extracting texture features from the annotated image, determining a distribution of the extracted texture features relative to tissue conditions, classifying the patient into a risk group based on the distribution, and treating the patient accordingly based on the risk group.

First claim

What is claimed is:

1. A computer-implemented method for determining treatment of a patient, the method comprising:
receiving an image of living tissue of a patient, the image comprising a stained histopathology slide;
annotating, via a whole slide annotation routine, the entire image into tissue structures, wherein the whole slide annotation routine includes at least one preprocessing routing to preprocess the stained histopathology slide using whole slide color normalization and down-sampling, and at least one patch extraction routine, wherein the at least one patch extraction routine comprises using a sliding window to extract patches of the preprocessed stained histopathology slide;
extracting texture features from the annotated image;
determining a distribution of the extracted texture features relative to tissue conditions, wherein the determining step comprises quantifying the distribution of the extracted texture features within the annotated entire image and quantifying spatial relationships relative to the tissue conditions of the tissue structures;
classifying the patient into a risk group based on the determining step; and
treating the patient accordingly based on the risk group.

2. The method of claim 1 , wherein the annotating comprises:
color deconvoluting each of the plurality of patches to a plurality of stain layers;
extracting a plurality of texture features from the plurality of patches;
inputting the extracted texture features into a random forest to output a probability of each patch belonging to a category of tissue structure; and
combining the patch probabilities to form an image annotation of tissue structures.

3. The method of claim 2 , wherein the plurality of texture features comprise at least one selected texture feature, at least one convolutional neural network fully connected terminal layer features, or a combination of the two.

4. The method of claim 3 , wherein the determining the distribution comprises determining feature distributions, spatial features which compare distances between different tissue regions, and region confidence.

5. The method of claim 4 , wherein the classifying comprises:
selecting a plurality of features; and
inputting the selected features into at least one machine learning process to output a probability of a condition to be treated and a treatment recommendation.

6. The method of claim 1 , wherein the living tissue is breast tissue and the risk groups relate to risk of recurrence of breast cancer.

7. The method of claim 6 , wherein the breast cancer is ductal carcinoma in situ.

8. The method of claim 7 , wherein the categories of tissue structure comprise malignant duct, immune rich stroma, non-immune rich stroma, non-cancerous duct, and blood vessel.

9. A system for detecting malicious email messages, the system comprising a processor, memory accessible by the processor, and computer program instructions stored in the memory and executable by the processor to perform:
receiving an image of living tissue of a patient, the image comprising a stained histopathology slide;
annotating, via a whole slide annotation routine, the entire image into tissue structures, wherein the whole slide annotation routine includes at least one preprocessing routing to preprocess the stained histopathology slide using whole slide color normalization and down-sampling, and at least one patch extraction routine, wherein the at least one patch extraction routine comprises using a sliding window to extract patches of the preprocessed stained histopathology slide;
extracting texture features from the annotated image;
determining a distribution of the extracted texture features relative to tissue conditions, wherein the determining step comprises quantifying the distribution of the extracted texture features within the annotated entire image and quantifying spatial relationships relative to the tissue conditions of the tissue structures;
classifying the patient into a risk group based on the determining step; and
treating the patient accordingly based on the risk group.

10. The system of claim 9 , wherein the annotating comprises:
color deconvoluting each of the plurality of patches to a plurality of stain layers;
extracting a plurality of texture features from the plurality of patches;
inputting the extracted texture features into a random forest to output a probability of each patch belonging to a category of tissue structure; and
combining the patch probabilities to form an image annotation of tissue structures.

11. The system of claim 10 , wherein the plurality of texture features comprise at least one selected texture feature, at least one convolutional neural network fully connected terminal layer features, or a combination of the two.

12. The system of claim 11 , wherein the determining the distribution comprises determining feature distributions, spatial features which compare distances between different tissue regions, and region confidence.

13. The system of claim 12 , wherein the classifying comprises:
selecting a plurality of features; and
inputting the selected features into at least one machine learning process to output a probability of a condition to be treated and a treatment recommendation.

14. The system of claim 9 , wherein the living tissue is breast tissue and the risk groups relate to risk of recurrence of breast cancer.

15. The system of claim 14 , wherein the breast cancer is ductal carcinoma in situ.

16. The system of claim 15 , wherein the categories of tissue structure comprise malignant duct, immune rich stroma, non-immune rich stroma, non-cancerous duct, and blood vessel.

17. A computer program product for detecting malicious email messages, the computer program product comprising a non-transitory computer readable storage having program instructions embodied therewith, the program instructions executable by a computer, to cause the computer to perform a method comprising:
receiving an image of living tissue of a patient, the image comprising a stained histopathology slide;
annotating, via a whole slide annotation routine, the entire image into tissue structures, wherein the whole slide annotation routine includes at least one preprocessing routing to preprocess the stained histopathology slide using whole slide color normalization and down-sampling, and at least one patch extraction routine, wherein the at least one patch extraction routine comprises using a sliding window to extract patches of the preprocessed stained histopathology slide;
extracting texture features from the annotated image;
determining a distribution of the extracted texture features relative to tissue conditions, wherein the determining step comprises quantifying the distribution of the extracted texture features within the annotated entire image and quantifying spatial relationships relative to the tissue conditions of the tissue structures;
classifying the patient into a risk group based on the determining step; and
treating the patient accordingly based on the risk group.

18. The computer program product of claim 17 , wherein the annotating comprises:
color deconvoluting each of the plurality of patches to a plurality of stain layers;
extracting a plurality of texture features from the plurality of patches;
inputting the extracted texture features into a random forest to output a probability of each patch belonging to a category of tissue structure; and
combining the patch probabilities to form an image annotation of tissue structures.

19. The computer pro

Assignees

• Univ Georgia State Res Found

Inventors

• Klimov Sergey
• Jiang Yi
• Gertych Arkadiusz

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06V10/26

  Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion · CPC title

• G06F18/24323

  Tree-organised classifiers · CPC title

• G06T2207/30068

  Mammography; Breast · CPC title