Predicting overall survival in early stage lung cancer with feature driven local cell graphs (FeDeG)

1. A non-transitory computer-readable storage device storing computer-executable instructions that when executed control a processor to perform operations, the operations comprising:

accessing an image of a region of tissue demonstrating a disease, and containing a plurality of cellular structures, at least some of the plurality of cellular structures being of different types;

for each cellular structure of a subset of the plurality of cellular structures, extracting one or more cell structure-level features from that cellular structure, at least some of the one or more cell structure-level features being indicative of the different types;

constructing one or more graphs, wherein each of the one or more graphs is based on both (a) the one or more cell structure-level features, and (b) one or more spatial relationships between at least one cellular structure of the subset and one or more other cellular structures of the subset, the one or more cell structure-level features and the one or more spatial relationships being independent of one another, such that said constructing differentiates between the different types of the plurality of cellular structures;

computing one or more graph-level features based on the one or more graphs; and

generating a medical prediction related to the disease based, at least in part, on the one or more graph-level features.

2. The non-transitory computer-readable storage device of claim 1 , where the image is a digitized image of a hematoxylin and eosin (H&E) stained tissue micro array (TMA) image of a region of tissue demonstrating non-small cell lung cancer (NSCLC).

3. The non-transitory computer-readable storage device of claim 2 , where the image is scanned at 20× magnification.

4. The non-transitory computer-readable storage device of claim 1 , wherein the one or more graph-level features describe a topology or a heterogeneity of constituent cell structures of the one or more graphs.

5. The non-transitory computer-readable storage device of claim 1 , wherein the plurality of cellular structures comprise entire cells, cellular nuclei, or both.

6. The non-transitory computer-readable storage device of claim 1 , the operations further comprising segmenting one or more cellular structures of the plurality of cellular structures using a local optimal thresholding approach, where the local optimal thresholding approach is based on a cellular shape of that cellular structure, and a cellular area of that cellular structure.

7. The non-transitory computer-readable storage device of claim 1 , where the one or more cell structure-level features describe a shape of that cellular structure, a size of that cellular structure, and a texture of that cellular structure.

8. The non-transitory computer-readable storage device of claim 7 , where the one or more cell structure-level features include six nuclear morphology features.

9. The non-transitory computer-readable storage device of claim 1 , where constructing the one or more graphs comprises using a mean-shift clustering approach that includes:

estimating a mode of an underlying density function of a member of the one or more cell structure-level features; and

grouping a cellular structure of the plurality of cellular structures into a sub-graph based on the mode.

10. The non-transitory computer-readable storage device of claim 1 , where the one or more graph-level features include four groups of features, the four groups of features including:

an intersection between different graph-level feature group;

a size of graph-level feature group;

a disorder of nuclear morphology feature group; and

an architectural measure of graph-level feature group.

11. The non-transitory computer-readable storage device of claim 10 , where the one or more graph-level features include at least one feature selected from each of the four groups of features, respectively.

12. The non-transitory computer-readable storage device of claim 1 , the operations further comprising generating the medical prediction using a machine learning classifier, where the machine learning classifier is a linear discriminant analysis (LDA) classifier.

13. The non-transitory computer-readable storage device of claim 12 , the operations further comprising training the machine learning classifier.

14. An apparatus comprising one or more processors configured to:

receive a digitized image of a region of tissue (ROT) demonstrating a disease and containing a plurality of cellular structures, at least some of the plurality of cellular structures being of different types;

for each cellular structure of a subset of the plurality of cellular structures, compute one or more cell structure-level features based on that cellular structure, at least some of the one or more cell structure-level features being indicative of the different types;

construct one or more graphs, wherein each of the one or more graphs is based on both (a) the one or more cell structure-level features, and (b) one or more spatial relationships between at least one cellular structure of the subset and one or more other cellular structures of the subset, the one or more cell structure-level features and the one or more spatial relationships being independent of one another, such that said constructing differentiates between the different types of the plurality of cellular structures;

compute one or more graph-level features based on the one or more graphs; and

generate a medical prediction related to the disease based, at least in part, on the one or more graph-level features.

15. The apparatus of claim 14 , where the digitized image is a digitized hematoxylin and eosin (H&E) stained tissue micro array (TMA) image of a region of tissue demonstrating non-small cell lung cancer (NSCLC) scanned at 20× magnification.

16. The apparatus of claim 14 , wherein the one or more processors are further configured to detect the plurality of cellular structures of the digitized image using a multiple-pass adaptive voting approach.

17. The apparatus of claim 14 , wherein the one or more processors are further configured to, for each cellular structure of the plurality of cellular structures, segment that cellular structure using a local optimal thresholding approach.

18. The apparatus of claim 14 , where the one or more graph-level features include:

at least one intersection between different graph-level feature;

at least one size of graph-level feature;

at least one disorder of nuclear morphology feature; and

at least one architectural measure of graph-level feature.

19. The apparatus of claim 14 , wherein the one or more processors are further configured to generate the medical prediction using a linear discriminant analysis (LDA) classifier.

20. A method comprising:

accessing an image of a region of tissue demonstrating a disease, and containing a plurality of cellular structures, at least some of the plurality of cellular structures being of different types;

for each cellular structure of a subset of the plurality of cellular structures, extracting one or more cell structure-level features from that cellular structure, at least some of the one or more cell structure-level features being indicative of the different types;

constructing one or more graphs, wherein each of the one or more graphs is based on both (a) the one or more cell structure-level features, and (b) one or more spatial relationships between at least one cellular structure of the subset and one or more other cellular structures of the subset, the one or more cell structure-level features and the one or more spatial relationships being independent of one another, such that said constructing differentiates between the different types of the plurality of cellular structures;

computing one or more graph-level features based on the one or more graphs; and

generating a medical prediction related to the disease based, at least in part, on the one or more graph-level features.