Image diagnostic system, and methods of operating thereof
AssigneeHuron Technologies International Inc.
InventorsMahdi S. Hosseini, Konstantinos N. Plataniotis, Lyndon Chan, Jasper Hayes, Savvas Damaskinos
Life of this patent
Estimating the term…
Abstract
Various image diagnostic systems, and methods of operating thereof, are disclosed herein. Example embodiments relate to operating the image diagnostic system to identify one or more tissue types within an image patch according to a hierarchical histological taxonomy, identifying an image patch associated with normal tissue, generating a pixel-level segmented image patch for an image patch, generating an encoded image patch for an image patch of at least one tissue, searching for one or more histopathological images, and assigning an image patch to one or more pathological cases.
Claims36
A system for generating a pixel-level segmented image patch for an image patch, the system comprises a processor operable to:
- apply a developed feature extractor to the image patch to identify one or more tissue segments within the image patch;
- apply a developed feature classifier to the one or more tissue segments to identify a respective tissue type and generate a corresponding confidence score representing a degree of similarity according to at least one prediction model generated based on a set of training image patches of a normal tissue of that tissue type, the tissue type being identified according to a hierarchical histology taxonomy;
- generate a class activation map for each identified tissue type based on the associated confidence score;
- and
- apply pixel-level segmentation to the class activation map to define a contour of each tissue segment of that tissue type to generate the pixel-level segmented image patch.
A method for generating a pixel-level segmented image patch for an image patch, the method comprises operating a processor to:
- apply a developed feature extractor to the image patch to identify one or more tissue segments within the image patch;
- apply a developed feature classifier to the one or more tissue segments to identify a respective tissue type and generate a corresponding confidence score representing a degree of similarity according to at least one prediction model generated based on a set of training image patches of a normal tissue of that tissue type, the tissue type being identified according to a hierarchical histology taxonomy;
- generate a class activation map for each identified tissue type based on the associated confidence score;
- and
- apply pixel-level segmentation to the class activation map to define a contour of each tissue segment of that tissue type to generate the pixel-level segmented image patch.
A system for generating an encoded image patch for an image patch of at least one tissue, the system comprises a processor operable to:
- apply a developed feature extractor to the image patch to identify one or more tissue segments within the image patch;
- apply a developed feature classifier to the one or more tissue segments to identify a respective tissue type and generate a corresponding confidence score representing a degree of similarity according to at least one prediction model generated based on a set of training image patches of a normal tissue of that tissue type, the tissue type being identified according to a hierarchical histology taxonomy;
- generate a class activation map for each identified tissue type based on the associated confidence score;
- apply pixel-level segmentation to the class activation map to define a contour of each tissue segment of that tissue type to generate the pixel-level segmented image patch;
- generate one or more binary masks for the image patch, each binary mask being associated with an identified tissue type and corresponding to a pixel-level representation of one or more tissue segments of that tissue type;
- for each binary mask, identify one or more features related to a structure of each tissue segment;
- generate one or more feature correspondence values relating each feature identified in each binary mask and each tissue type identified for the image patch;
- and
- generate the encoded image patch based on the one or more feature correspondence values.
The system of 25, wherein the processor is operable to generate a unique encoded image patch for each hierarchy level of the hierarchical histological taxonomy based on which the one or more tissue types are identified.
A method for generating an encoded image patch for an image patch of at least one tissue, the method comprises operating a processor to:
- apply a developed feature extractor to the image patch to identify one or more tissue segments within the image patch;
- apply a developed feature classifier to the one or more tissue segments to identify a respective tissue type and generate a corresponding confidence score representing a degree of similarity according to at least one prediction model generated based on a set of training image patches of a normal tissue of that tissue type, the tissue type being identified according to a hierarchical histology taxonomy;
- generate a class activation map for each identified tissue type based on the associated confidence score;
- apply pixel-level segmentation to the class activation map to define a contour of each tissue segment of that tissue type to generate the pixel-level segmented image patch;
- generate one or more binary masks for the image patch, each binary mask being associated with an identified tissue type and corresponding to a pixel-level representation of one or more tissue segments of that tissue type;
- for each binary mask, identify one or more features related to a structure of each tissue segment;
- generate one or more feature correspondence values relating each feature identified in each binary mask and each tissue type identified for the image patch;
- and
- generate the encoded image patch based on the one or more feature correspondence values.
Dependencies are inferred from the claim text.