Context driven image mining to generate image-based biomarkers

1. A method comprising:

acquiring pixel values of medical images of patients;

generating objects of a data network for each of the medical images by linking selected pixel values to selected objects according to a class network and a process hierarchy;

measuring the objects to obtain a value for each of a plurality of image features;

selecting a first subset of the image features that belong to a first image-based biomarker;

correlating the first image-based biomarker for each of the patients with a clinical endpoint observed for that patient;

displaying on a graphical user interface how the first image-based biomarker correlates with the clinical endpoint observed for each of the patients;

identifying a subgroup of the patients for which the first image-based biomarker correlates poorly with the clinical endpoints observed for the patients in the subgroup; and

selecting a second subset of the image features that belong to a second image-based biomarker that better correlates with the clinical endpoints observed for the patients in the subgroup.

2. The method of claim 1 , wherein each value for each image feature in the first subset is weighted and summed with weighted values of the remaining image features in the first subset to obtain a magnitude of the first image-based biomarker.

3. The method of claim 2 , wherein the correlating the first image-based biomarker with the clinical endpoint involves plotting the magnitude of the first image-based biomarker for each patient against the clinical endpoint observed for that patient.

4. The method of claim 3 , wherein the identifying the subgroup of patients involves performing a regression analysis on the magnitudes of the first image-based biomarker plotted against the clinical endpoints observed for the patients.

5. The method of claim 1 , wherein the identifying the subgroup is performed using a scatter plot displayed on the graphical user interface, and wherein each point on the scatter plot represents one of the patients.

6. The method of claim 1 , wherein the subgroup is expanded to include patients corresponding to points on the scatter plot that are closest to the points originally included in the subgroup.

7. The method of claim 1 , wherein the identifying the subgroup involves plotting the clinical endpoints observed for each of the patients as a Kaplan-Meier curve.

8. The method of claim 1 , wherein the correlating the first image-based biomarker with the clinical endpoint involves performing a decision tree analysis with each of the first subset of image features to generate a predicted clinical endpoint for each patient and then plotting the predicted clinical endpoint for each patient against the clinical endpoint observed for that patient.

9. The method of claim 1 , further comprising:

modifying how the value of each image feature in the second subset is weighted such that the second image-based biomarker better correlates with the clinical endpoint observed for the patients in the subgroup.

10. The method of claim 1 , further comprising:

saving the second image-based biomarker to a computer-readable medium.

11. The method of claim 1 , further comprising:

correlating the second image-based biomarker for each of the patients in the subgroup with the clinical endpoint observed for that patient.

12. The method of claim 1 , wherein the class network has a class, and wherein a membership function defines a likelihood that each of the objects of the data network belongs to the class.

13. The method of claim 12 , wherein a process step is part of the process hierarchy, and wherein the process step has a domain that designates the class.

14. The method of claim 1 , wherein the clinical endpoint is a survival time.

15. The method of claim 1 , wherein the clinical endpoint is a time for recurrence of cancer.