Image diagnostic system, and methods of operating thereof

1. A system for identifying one or more tissue types within an image patch according to a hierarchical histological taxonomy, the system comprises:

a supervised digital pathology database having a set of training image patches stored thereon; and

a processor in communication with the supervised digital pathology database and operable to:

access the set of training image patches stored in the supervised digital pathology database;

develop a feature extractor to generate a training feature vector identifying one or more training tissue segments in each training image patch in the set of training image patches;

develop a feature classifier to assign a tissue type to each training tissue segment identified in the training feature vector and to update a class confidence score for that tissue type, the tissue type being assigned according to the hierarchical histological taxonomy;

apply the feature extractor to the image patch to identify one or more tissue segments within the image patch;

apply the feature classifier to assign a tissue type to the one or more tissue segments identified by the feature extractor for the image patch and to generate a confidence score for assigned tissue type within the image patch with reference to the class confidence score; and

generate a confidence score vector containing the confidence scores for each tissue type identified for the feature vector by the feature classifier.

2. The system of claim 1 , wherein the processor is operable to:

develop a convolutional neural network based on the set of training image patches.

3. The system of claim 2 , wherein an architecture of the convolutional neural network comprises three convolutional blocks, a global max pooling layer, a single fully-connected layer, and a sigmoid layer.

4. The system of claim 1 , wherein the processor is operable to:

generate a feature vector identifying the one or more tissue segments of the image patch.

5. The system of claim 1 , wherein the confidence score vector has a size corresponding to a number of tissue types identified by the feature classifier for the image patch.

6. The system of claim 1 , wherein the processor is operable to:

map one or more prediction models to each tissue segment within the image patch to identify the tissue type relative to the training feature vector and class confidence score for that tissue type.

7. The system of claim 1 , wherein the processor is operable to:

evaluate the confidence score generated by the feature classifier by comparing the confidence score with a confidence threshold; and

generate a quality indicator indicating the confidence score satisfies the confidence threshold, and otherwise, generate the quality indicator indicating the confidence score fails to satisfy the confidence threshold.

8. The system of claim 1 , wherein the set of training image patches comprises one or more image patches showing at least one normal tissue and the at least one normal tissue is labelled with a tissue type.

9. The system of claim 1 , wherein the processor is further operable to:

identify one or more regions of interest in the image patch from which to generate one or more image patches; and

identify the one or more tissue types in each image patch of the one or more image patches.

10. A method for identifying one or more tissue types within an image patch according to a hierarchical histological taxonomy, the method comprises operating a processor to:

access a set of training image patches stored in a supervised digital pathology database;

develop a feature extractor to generate a training feature vector identifying one or more training tissue segments in each training image patch in the set of training image patches;

develop a feature classifier to assign a tissue type to each training tissue segment identified in the training feature vector and to update a class confidence score for that tissue type, the tissue type being assigned according to the hierarchical histological taxonomy;

apply the feature extractor to the image patch to identify one or more tissue segments within the image patch;

apply the feature classifier to assign a tissue type to the one or more tissue segments identified by the feature extractor for the image patch and to generate a confidence score for assigned tissue type within the image patch with reference to the class confidence score; and

generate a confidence score vector containing the confidence scores for each tissue type identified for the feature vector by the feature classifier.

11. The method of claim 10 comprises operating the processor to:

develop a convolutional neural network based on the set of training image patches.

12. The method of claim 11 , wherein an architecture of the convolutional neural network comprises three convolutional blocks, a global max pooling layer, a single fully-connected layer, and a sigmoid layer.

13. The method of claim 10 comprises operating the processor to:

generate a feature vector identifying the one or more tissue segments of the image patch.

14. The method of claim 10 , wherein the confidence score vector has a size corresponding to a number of tissue types identified by the feature classifier for the image patch.

15. The method of claim 10 comprises operating the processor to:

map one or more prediction models to each tissue segment within the image patch to identify the tissue type relative to the training feature vector and class confidence score for that tissue type.

16. The method of claim 10 comprises operating the processor to:

evaluate the confidence score generated by the feature classifier by comparing the confidence score with a confidence threshold; and

generate a quality indicator indicating the confidence score satisfies the confidence threshold, and otherwise, generate the quality indicator indicating the confidence score fails to satisfy the confidence threshold.

17. The method of claim 10 comprises operating the processor to:

identify one or more regions of interest in the image patch from which to generate one or more image patches; and

identify the one or more tissue types in each image patch of the one or more image patches.

18. A non-transitory computer-readable medium comprising instructions executable on a processor for implementing a method for identifying one or more tissue types within an image patch according to a hierarchical histological taxonomy, the method comprising operating the processor to:

access a set of training image patches stored in a supervised digital pathology database;

develop a feature extractor to generate a training feature vector identifying one or more training tissue segments in each training image patch in the set of training image patches;

develop a feature classifier to assign a tissue type to each training tissue segment identified in the training feature vector and to update a class confidence score for that tissue type, the tissue type being assigned according to the hierarchical histological taxonomy;

apply the feature extractor to the image patch to identify one or more tissue segments within the image patch;

apply the feature classifier to assign a tissue type to the one or more tissue segments identified by the feature extractor for the image patch and to generate a confidence score for assigned tissue type within the image patch with reference to the class confidence score; and

generate a confidence score vector containing the confidence scores for each tissue type identified for the feature vector by the feature classifier.

19. The method of claim 10 , wherein the set of training image patches comprises one or more image patches showing at least one normal tissue and the at least one normal tissue is labelled with a tissue type.

20. The method of claim 19 , wherein the at least one normal tissue is labelled according to the hierarchical histological taxonomy.