Quantitatively characterizing disease morphology with co-occurring gland tensors in localized subgraphs

• US 9,177,105 B2
• Filed: 03/26/2014
• Issued: 11/03/2015
• Est. Priority Date: 03/29/2013
• Status: Active Grant

• Current Assignee: Case Western Reserve University
• Original Assignee: Case Western Reserve University
• Inventors: Madabhushi, Anant, Lee, George, Ali, Sahirzeeshan, Sparks, Rachel
• Primary Examiner(s): Cunningham, Gregory F
• Application Number:
  US14/226,083
• Publication Number:
  US 20140294264A1
• Time in Patent Office: 587 Days
• Field of Search: 382/128
• US Class Current: 1/1
• CPC Class Codes: G01R 33/5608 Data processing and visualization specially adapted for MR, e.g. for feature analysis and pattern recognition on the basis of measured MR data, segmentation of measured MR data, edge contour detection on the basis of measured MR data, for enhancing measured MR data in terms of signal-to-noise ratio by means of noise filtering or apodization, for enhancing measured MR data in terms of resolution by means for deblurring, windowing, zero filling, or generation of gray-scaled images, colour-coded images or images displaying vectors instead of pixels image data processing or generation, in general G06T
• CPC Class Codes: G06F 18/214 Generating training patterns; Bootstrap methods, e.g. bagging or boosting
• CPC Class Codes: G06T 2207/30081 Prostate
• CPC Class Codes: G06T 7/0014 using an image reference approach
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling
• CPC Class Codes: G06V 20/695 Preprocessing, e.g. image segmentation
• CPC Class Codes: G06V 20/698 Matching; Classification
• CPC Class Codes: G16H 50/30 for calculating health indices; for individual health risk assessment

Abstract

Apparatus, methods, and other embodiments associated with objectively predicting biochemical recurrence with co-occurring gland tensors in localized subgraphs are described. One example apparatus includes a set of logics that associate directional disorder with a risk of failure in a material. A first logic detects a fundamental unit of composition in the material, segments boundaries of the fundamental unit, and calculates a directional tensor for the fundamental unit. A second logic constructs a localized sparsified subgraph whose nodes represent centroids of the fundamental units, defines pairwise spatial relationships between the fundamental units, and constructs a directional co-occurrence matrix based on the spatial relationships. A third logic derives second order statistical features from the co-occurrence matrix, and produces a risk failure score as a function of the second order statistical features. The second order statistical features include the entropy of the directional organization of the fundamental units.

Claims

1. A non-transitory computer-readable storage medium storing computer executable instructions that when executed by a computer cause the computer to perform a method of associating gland orientation disorder with malignancy and risk of post-surgical biochemical recurrence (BCR) in a prostate cancer (CaP) patient, the method comprising:

accessing a digitized image of a section of a prostate demonstrating pathology associated with CaP in the patient;

detecting a gland in a region of interest of the digitized image;

segmenting an individual gland boundary in the region of interest in the digitized image into a set of gland boundary points;

producing a gland tensor by associating a tensor with the gland, where the gland tensor indicates the dominant orientation of the gland, and where the gland tensor is based on the major axis of the gland;

constructing a subgraph of a localized gland network within the region of interest, where constructing the subgraph comprises linking individual glands located proximal to each other into the localized gland network, where the nodes of the subgraph represent individual gland centroids, and where the edges of the subgraph are defined between pairs of glands by a probabilistic decaying function;

constructing a tensor co-occurrence matrix, where elements of the tensor co-occurrence matrix comprise gland tensor pairs, where the gland tensor pairs are defined by the subgraph, and where the tensor co-occurrence matrix aggregates co-occurring gland tensors based, at least in part, on the frequency with which orientations of two individual glands located proximal to each other co-occur;

deriving second-order statistics of gland orientations in the localized gland networks in the digitized image;

selectively differentiating a cancerous tissue region from a non-cancerous tissue region in the image based, at least in part, on the second-order statistics; and

establishing a BCR score for the patient based, at least in part, on the second-order statistics.

2. The non-transitory computer-readable storage medium of claim 1, where detecting a gland and segmenting an individual gland boundary comprises controlling the computer to:

access a color space luminance channel image of the region of interest, where glands in the color space luminance channel image appear as contiguous, high intensity pixel regions;

produce a convolved luminance image by convolving the color space luminance channel image with a Gaussian kernel;

detect a seed pixel corresponding to a peak intensity in the convolved luminance image, where the seed pixel represents a current region (CR);

initialize a bounding box around the seed pixel;

denote eight connected pixels surrounding the seed pixel as a current boundary (CB);

produce a removed CB pixel by removing the pixel in the CB with the highest intensity from the CB, and

upon determining that the removed CB pixel is not outside the bounding box;

controlling the computer to incorporate the removed CB pixel into the CR as a new CR pixel;

incorporate the eight connected pixels surrounding the new CR pixel not already in the CR into the CB;

define an internal boundary (IB) as all CR pixels adjacent to the CB, and

define a boundary strength as the mean intensity of the pixels in the IB minus the mean intensity of the pixels in the CB.

3. The non-transitory computer-readable storage medium of claim 2, the method comprising defining an optimal region as the CR where the maximum boundary strength is achieved.

4. The non-transitory computer-readable storage medium of claim 1, the method comprising using a probabilistic pair-wise Markov random field classifier to remove a non-gland boundary in the region of interest.

5. The non-transitory computer-readable storage medium of claim 1, the method comprising obtaining the dominant orientation of the gland through principal component analysis (PCA) of the set of gland boundary points, where the first principal component describes the directionality of the gland in the form of the major axis, where the greatest variance in the gland boundary occurs along the major axis.

6. The non-transitory computer-readable storage medium of claim 5, the method comprising converting the major axis to an angle counterclockwise from a positive horizontal axis.

7. The non-transitory computer-readable storage medium of claim 1, where the subgraph is a sparsified graph, where the sparsified graph defines a pairwise spatial relationship between glands, and where the contributions of stromal regions in the region of interest are minimized by preventing edges in the sparsified graph from traversing heterogeneous tissues including stroma and epithelium.

8. The non-transitory computer-readable storage medium of claim 1, where the probabilistic decaying function is a probabilistic decaying function of the Euclidean distance between gland centroids.

9. The non-transitory computer-readable storage medium of claim 1, where the density of the subgraph is controlled by a randomly generated value V from 0.0 through infinity, where 0 represents a high probability of connecting nodes, and infinity represents a low probability of connecting nodes, V being a real number.

10. The non-transitory computer-readable storage medium of claim 1, the method comprising discretizing gland orientations in angular bins following conversion of the major axis to an angle counterclockwise from a positive horizontal axis.

11. The non-transitory computer-readable storage medium of claim 1, where the second-order statistics comprise at least one of entropy, homogeneity, energy, correlation, contrast energy, contrast inverse moment, contrast average, contrast variance, contrast entropy, intensity average, intensity variance, and intensity entropy.

12. The non-transitory computer-readable storage medium of claim 1, the method comprising differentiating the cancerous tissue region and the non-cancerous tissue region as a function of the disorganization of the gland orientation.

13. The non-transitory computer-readable storage medium of claim 2, where the color space luminance channel image is a Commission internationale de l'"'"'é

clairage L*a*b* (CIELAB) image.

14. A method for controlling a computer to compute a biochemical recurrence (BCR) score for a cancer patient, comprising:

detecting a gland in a region of interest of a digitized histopathology section associated with the cancer patient;

segmenting gland boundaries of the gland into a set of gland boundaries;

calculating a directional gland tensor for the gland;

constructing a localized subgraph of a gland network in the region of interest;

defining pairwise spatial relationships between two or more glands in the subgraph;

constructing a directional gland tensor co-occurrence matrix based, at least in part, on the pairwise spatial relationships;

deriving second order statistical features from the directional gland tensor co-occurrence matrix; and

controlling the computer to produce a BCR recurrence score based, at least in part, on the second order statistical features, where the BCR recurrence score predicts BCR with at least 66% accuracy.

15. The method of claim 14, comprising:

detecting the gland using automatic region growing; and

segmenting the gland boundaries using automatic region growing.

16. The method of claim 14, comprising calculating the directional gland tensor from a principal component analysis (PCA) of the set of gland boundaries.

17. The method of claim 14, where the localized subgraph is a cluster graph, and where the nodes of the localized subgraph represent gland centroids.

18. The method of claim 14, where the edges of the localized subgraph between at least one pair of glands are defined by a probabilistic decaying function, where the probabilistic decaying function minimizes the contributions of stromal regions in the region of interest by preventing edges of the subgraph from traversing heterogeneous tissues including stroma and epithelium.

19. The method of claim 14, where the second order statistical features include at least one of entropy, homogeneity, energy, and correlation.

20. The method of claim 19, where entropy is defined as Σ

<sub>θ

1,θ

2</sub>−

C(θ

₁,θ

₂, )log(C(θ

₁,θ

₂))) where θ

₁ is a directional gland tensor for a first gland in a pair of glands, where the pair of glands is in the localized subgraph, where θ

₂ is a directional gland tensor for a second gland in the pair of glands, and where C is the directional gland tensor co-occurrence matrix, where the directional gland tensor co-occurrence matrix is a b×

b co-occurrence matrix.

21. The method of claim 19, where homogeneity is defined as

22. The method of claim 19, where energy is defined as Σ

<sub>θ

1,θ

2</sub>C(θ

₁,θ

₂)² where θ

₁ is a directional gland tensor for a first gland in a pair of glands, where the pair of glands is in the localized subgraph, where θ

₂ is a directional gland tensor for a second gland in the pair of glands, and where C is the directional gland tensor co-occurrence matrix, where the directional gland tensor co-occurrence matrix is a b×

b co-occurrence matrix.

23. The method of claim 19, where correlation is defined as

24. An apparatus, comprising:

a processor;

a memory;

an input/output interface;

a set of logics that associate directional disorder with a risk of failure in a region of interest in a material, and

an interface to connect the processor, the memory, the input/output interface and the set of logics,

the set of logics comprising;

a first logic that detects a fundamental unit of composition in the region of interest, segments boundaries of the fundamental unit, and calculates a tensor for the fundamental unit, where the tensor indicates the dominant orientation of the fundamental unit;

a second logic that constructs a sparsified localized subgraph from the detected fundamental units, defines pairwise spatial relationships between the fundamental units in the subgraph, and constructs a directional tensor co-occurrence matrix based, at least in part on the pairwise spatial relationships, where the nodes of the sparsified localized subgraph represent the centroids of the fundamental unit; and

a third logic that derives second order statistical features from the directional tensor co-occurrence matrix, and produces a risk of failure score based, at least in part, on the second order statistical features.