Predicting cancer progression using cell run length features

1. An apparatus for predicting cancer progression, the apparatus comprising:

a processor;

a memory configured to store a digitized image of a region of tissue demonstrating cancerous pathology;

an input/output (I/O) interface;

a set of circuits comprising an image acquisition circuit, a nuclei detection and graphing circuit, a cell run-length feature (CRF) circuit, and a classification circuit; and

an interface that connects the processor, the memory, the I/O interface, and the set of circuits;

the image acquisition circuit configured to access an image of a region of tissue demonstrating cancerous pathology, where the image includes a plurality of pixels, a pixel having an intensity;

the nuclei detection and graphing circuit configured to:

detect a plurality of cellular nuclei represented in the image; and

construct a nuclear sub-graph based on the plurality of cellular nuclei, where a node of the sub-graph is a nuclear centroid of a member of the plurality of cellular nuclei;

the CRF circuit configured to:

compute a cell run-length feature (CRF) vector based on the sub-graph;

compute a set of CRF features based on the CRF vector and the sub-graph; and

generate a CRF signature based, at least in part, on the set of CRF features; and

the classification circuit configured to:

compute a probability that the region of tissue will experience cancer progression, based, at least in part, on the CRF signature; and

generate a classification of the region of tissue as a progressor or non-progressor based, at least in part, on the probability.

2. The apparatus of claim 1 , where the image is a tissue micro array (TMA) of a hematoxylin and eosin (H&E) stained pathology slide of a region of tissue demonstrating p16+ oropharyngeal squamous cell cancer scanned at 40× magnification with a resolution of 0.25 μm per pixel.

3. The apparatus of claim 1 , where the nuclei detection and graphing circuit is configured to detect the plurality of cellular nuclei represented in the image using a generalized Laplacian of Gaussian (gLoG) detection approach.

4. The apparatus of claim 1 , where the nuclei detection and graphing circuit is configured to construct the nuclear sub-graph based on a probabilistic decaying function of the Euclidean distance between a first nuclear centroid of a first member of the plurality of cellular nuclei, and a second, different nuclear centroid of a second, different member of the plurality of cellular nuclei.

5. The apparatus of claim 1 where:

the CRF vector is defined as the number of cell runs associated with different length cell runs in the nuclear sub-graph, where a cell run is defined as a single traverse from a first end point to a second, different end point of the nuclear sub-graph, and where a length is defined as the total number of nodes traversed in the cell run.

6. The apparatus of claim 1 , where the set of CRF features includes a short run emphasis feature, a long run emphasis feature, a run-length non-uniformity feature, and a run percentage feature.

7. The apparatus of claim 1 , where the CRF circuit is configured to generate the CRF signature by computing, for the image, a set of first order statistics based on the set of CRF features, where the set of first order statistics include a mean, a standard deviation, a kurtosis, a skewness, and a range.

8. The apparatus of claim 1 , where the classification circuit is configured as a machine learning classifier, where the machine learning classifier is a linear discriminant analysis (LDA) classifier, a quadratic discriminant analysis (QDA) classifier, a support vector machine (SVM) classifier, or a random forest (RF) classifier.

9. The apparatus of claim 1 , the set of circuits further comprising a cancer treatment plan circuit configured to generate a cancer treatment plan based, at least in part, on the classification, and at least one of the probability, the CRF signature, or the image.

10. A non-transitory computer-readable storage device storing computer-executable instructions that when executed by a computer control the computer to perform a method for predicting cancer progression, the method comprising:

accessing an image of a region of tissue demonstrating cancerous pathology, where the image includes a plurality of pixels, a pixel having an intensity;

detecting a plurality of cellular nuclei represented in the image;

constructing a nuclear sub-graph based on the plurality of cellular nuclei, where a node of the sub-graph is a nuclear centroid of a member of the plurality of cellular nuclei;

computing a cell run-length feature (CRF) vector based on the sub-graph;

computing a set of CRF features based on the CRF vector and the sub-graph;

generating a CRF signature based, at least in part, on the set of CRF features;

providing the CRF signature to a machine learning classifier;

receiving, from the machine learning classifier, a probability that the region of tissue will experience cancer progression, where the machine learning classifier computes the probability based, at least in part, on the CRF signature;

generating a classification by classifying the region of tissue as a progressor or non-progressor based, at least in part, on the probability; and

generating a cancer treatment plan based, at least in part, on the classification, and at least one of the probability, the CRF signature, or the image.

11. The non-transitory computer-readable storage device of claim 10 , where the image is a tissue micro array (TMA) of a hematoxylin and eosin (H&E) stained pathology slide of a region of tissue demonstrating p16+ oropharyngeal squamous cell cancer scanned at 40× magnification with a resolution of 0.25 μm per pixel.

12. The non-transitory computer-readable storage device of claim 10 , where the image is a digitized whole slide image (WSI) of a region of tissue demonstrating cancerous pathology.

13. The non-transitory computer-readable storage device of claim 10 , where detecting at least one cellular nucleus represented in the image includes detecting at least one cellular nucleus using a generalized Laplacian of Gaussian (gLoG) detection approach.

14. The non-transitory computer-readable storage device of claim 13 , where the gLoG detection approach comprises:

generating a bank of gLoG kernels, where a first member of the bank of gLoG kernels has a first scale and a first orientation, and where a second member of the bank of gLoG kernels has a second, different scale or a second, different orientation;

generating a response map by convolving the bank of gLoG kernels with the image;

identifying a local maximum of the response map;

upon determining that a first local maximum is proximal to a second, different local maximum:

merging the first local maximum with the second, different local maximum;

determining a maximum response within a group of maxima that are proximal; and

defining the maximum response as the centroid a nucleus.

15. The non-transitory computer-readable storage device of claim 10 , where the nuclear sub-graph is constructed based on a probabilistic decaying function of the Euclidean distance between a first nuclear centroid of a first member of the plurality of cellular nuclei, and a second, different nuclear centroid of a second, different member of the plurality of cellular nuclei.

16. The non-transitory computer-readable storage device of claim 10 ,

where the CRF vector is defined as the number of cell runs associated with different length cell runs in the nuclear sub-graph,

where a cell run is defined as a single traverse from a first end point to a second, different end point of the nuclear sub-graph, and

where a length is defined as the total number of nodes traversed in the cell run.

17. The non-transitory computer-readable storage device of claim 10 , where the set of CRF features includes a short run emphasis feature, a long run emphasis feature, a run-length non-uniformity feature, and a run percentage feature.

18. The non-transitory computer-readable storage device of claim 10 , where generating the CRF signature comprises computing, for the image, a set of first order statistics based on the set of CRF features, where the set of first order statistics include a mean, a standard deviation, a kurtosis, a skewness, and a range.

19. The non-transitory computer-readable storage device of claim 10 , where the machine learning classifier is a linear discriminant analysis (LDA) classifier, a quadratic discriminant analysis (QDA) classifier, a support vector machine (SVM) classifier, or a random forest (RF) classifier.

20. A non-transitory computer-readable storage device storing computer-executable instructions that, in response to execution, cause a cancer progression prediction system to perform operations comprising:

accessing an image of a region of tissue demonstrating cancerous pathology, where the image includes a plurality of pixels, a pixel having an intensity;

detecting a plurality of cellular nuclei represented in the image;

constructing a nuclear sub-graph based on the plurality of cellular nuclei, where a node of the sub-graph is a nuclear centroid of a member of the plurality of cellular nuclei;

computing a cell run-length feature (CRF) vector based on the sub-graph;

computing a set of CRF features based on the CRF vector and the sub-graph;

generating a CRF signature based, at least in part, on the set of CRF features;

computing a probability that the region of tissue will experience cancer progression based, at least in part, on the CRF signature;

generating a classification by classifying the region of tissue as a progressor or non-progressor based, at least in part, on the probability;

generating a cancer treatment plan based, at least in part, on the classification, and at least one of the probability, the CRF signature, or the image; and

displaying, the cancer treatment plan and at least one of the classification, the probability, the CRF signature, or the image on a computer monitor.