Method and system for hierarchical tissue analysis and classification
First Claim
1. A method for hierarchical segmentation and classification of image data from a tissue specimen, comprising:
- receiving one or more images of a tissue specimen;
segmenting the one or more images at a first resolution level and at least a second resolution level utilizing a segmentation algorithm, wherein one or more parameters of the segmentation algorithm are trainable via user classification feedback, wherein the image at the first resolution level is segmented into a first plurality of segment primitives and the image at the at least a second resolution level is segmented into a second plurality of segment primitives;
extracting one or more features from some of the first plurality of segment primitives and some of the at least a second plurality of segment primitives utilizing a segment feature extraction algorithm, wherein one or more parameters of the segment feature extraction algorithm are trainable via user classification feedback;
building a first segmentation hierarchy by generating one or more clusters of some of the first plurality of segment primitives utilizing a clustering algorithm, wherein one or more of the clustering algorithm parameters are trainable via user classification feedback;
building at least a second segmentation hierarchy by generating one or more clusters of some of the at least a second plurality of segment primitives utilizing the clustering algorithm;
extracting one or more features from the first segmentation hierarchy and the at least a second segmentation hierarchy utilizing a hierarchy feature extraction algorithm, wherein one or more hierarchy feature extraction algorithm parameters are trainable via feedback from user classification feedback;
determining an inter-level relationship between one or more clusters generated for the first resolution level and one or more clusters generated for the at least a second level by comparing one or more characteristics of the one or more clusters of the first resolution level to one or more characteristics of the one or more clusters of the at least a second resolution level; and
autoclassifying one or more tissue elements of the tissue specimen via a user-trained classification algorithm utilizing at least one of the extracted features from some of the first plurality of segment primitives, the extracted features from some of the at least a second plurality of segment primitives, the extracted features from the first segmentation hierarchy, the extracted features from the at least a second segmentation hierarchy, the determined inter-level relationship between clusters of the first resolution level and the clusters of the at least a second resolution level, or user classification feedback.
1 Assignment
0 Petitions
Accused Products
Abstract
The present invention may include segmenting an image at a first resolution level and a second resolution level, wherein one or more parameters of a segmentation algorithm are trainable via user classification feedback, extracting features from a first plurality of segment primitives and a second plurality of segment primitives, wherein one or more parameters of a segment feature extraction algorithm are trainable via user classification feedback, building a first and second segmentation hierarchy by generating one or more clusters of the first plurality of segment primitives and the second plurality of segment primitives, extracting one or more features from the first segmentation hierarchy and the second segmentation hierarchy utilizing a hierarchy feature extraction algorithm, determining an inter-level relationship between the clusters generated for the first resolution level and the second level, and automatically classifying one or more tissue elements of the tissue specimen via a user-trained classification algorithm.
37 Citations
25 Claims
-
1. A method for hierarchical segmentation and classification of image data from a tissue specimen, comprising:
-
receiving one or more images of a tissue specimen; segmenting the one or more images at a first resolution level and at least a second resolution level utilizing a segmentation algorithm, wherein one or more parameters of the segmentation algorithm are trainable via user classification feedback, wherein the image at the first resolution level is segmented into a first plurality of segment primitives and the image at the at least a second resolution level is segmented into a second plurality of segment primitives; extracting one or more features from some of the first plurality of segment primitives and some of the at least a second plurality of segment primitives utilizing a segment feature extraction algorithm, wherein one or more parameters of the segment feature extraction algorithm are trainable via user classification feedback; building a first segmentation hierarchy by generating one or more clusters of some of the first plurality of segment primitives utilizing a clustering algorithm, wherein one or more of the clustering algorithm parameters are trainable via user classification feedback; building at least a second segmentation hierarchy by generating one or more clusters of some of the at least a second plurality of segment primitives utilizing the clustering algorithm; extracting one or more features from the first segmentation hierarchy and the at least a second segmentation hierarchy utilizing a hierarchy feature extraction algorithm, wherein one or more hierarchy feature extraction algorithm parameters are trainable via feedback from user classification feedback; determining an inter-level relationship between one or more clusters generated for the first resolution level and one or more clusters generated for the at least a second level by comparing one or more characteristics of the one or more clusters of the first resolution level to one or more characteristics of the one or more clusters of the at least a second resolution level; and autoclassifying one or more tissue elements of the tissue specimen via a user-trained classification algorithm utilizing at least one of the extracted features from some of the first plurality of segment primitives, the extracted features from some of the at least a second plurality of segment primitives, the extracted features from the first segmentation hierarchy, the extracted features from the at least a second segmentation hierarchy, the determined inter-level relationship between clusters of the first resolution level and the clusters of the at least a second resolution level, or user classification feedback. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19)
-
-
20. An apparatus for hierarchical segmentation and classification of image data from a tissue specimen, comprising:
-
one or more image sources; one or more user interfaces; one or more display devices; and one or more computer systems configured to; receive one or more images of a tissue specimen; segment the one or more images at a first resolution level and at least a second resolution level utilizing a segmentation algorithm, wherein one or more parameters of the segmentation algorithm are trainable via user classification feedback, wherein the image at the first resolution level is segmented into a first plurality of segment primitives and the image at the at least a second resolution level is segmented into a second plurality of segment primitives; extract one or more features from some of the first plurality of segment primitives and some of the at least a second plurality of segment primitives utilizing a segment feature extraction algorithm, wherein one or more parameters of the segment feature extraction algorithm are trainable via user classification feedback; build a first segmentation hierarchy by generating one or more clusters of some of the first plurality of segment primitives utilizing a clustering algorithm, wherein one or more of the clustering algorithm parameters are trainable via user classification feedback; build at least a second segmentation hierarchy by generating one or more clusters of some of the at least a second plurality of segment primitives utilizing the clustering algorithm; extract one or more features from the first segmentation hierarchy and the at least a second segmentation hierarchy utilizing a hierarchy feature extraction algorithm, wherein one or more hierarchy feature extraction algorithm parameters are trainable via feedback from user classification feedback; determine an inter-level relationship between one or more clusters generated for the first resolution level and one or more clusters generated for the at least a second level by comparing one or more characteristics of the one or more clusters of the first resolution level to one or more characteristics of the one or more clusters of the at least a second resolution level; and autoclassify one or more tissue elements of the tissue specimen utilizing at least one of the extracted features from some of the first plurality of segment primitives, the extracted features from some of the at least a second plurality of segment primitives, the extracted features from the first segmentation hierarchy, the extracted features from the at least a second segmentation hierarchy, the determined inter-level relationship between clusters of the first resolution level and the clusters of the at least a second resolution level, or user classification feedback. - View Dependent Claims (21, 22, 23, 24, 25)
-
Specification