Direct computation of image-derived biomarkers

1. A method for computing image-derived biomarkers, the method comprising:

receiving image data defining a three-dimensional image volume representative of an anatomical region of interest and comprising a plurality of voxels;

applying an intensity model on a per-voxel basis across the three-dimensional image volume to assign a weighting value for each voxel indicating a likelihood that the voxel includes anatomical features related to a biomarker of interest, wherein the intensity model is a machine learning model trained using intensity features of previously acquired image data; and

deriving information related to the biomarker of interest without prior creation of a segmented mesh from the image data by performing one or more modeling computations directly on the image data, wherein the weighting values are used to limit the modeling computations to voxels comprising the anatomical features related to the biomarker of interest.

2. The method of claim 1 , further comprising:

generating an approximate segmentation of the anatomical region of interest based on results of performing the one or more modeling computations.

3. The method of claim 2 , wherein the approximate segmentation is a binary mask.

4. The method of claim 2 , wherein the approximate segmentation is a gray-level mask representative of segmentation accuracy for the anatomical region of interest.

5. The method of claim 2 , wherein the approximate segmentation comprises a probability of particular pixels in the three-dimensional image volume belonging to the anatomical region of interest.

6. The method of claim 2 , further comprising:

using the approximate segmentation to iteratively perform the one or more modeling computations directly on the image data to derive information related to the biomarker of interest.

7. The method of claim 6 , further comprising:

refining the approximate segmentation during each iteration based on results of the one or more modeling computations.

8. The method of claim 1 , wherein the anatomical region of interest comprises cardiac vascular structure and the one or more modeling computations provide a measure of fluid dynamics within the cardiac vascular structure.

9. The method of claim 8 , wherein the measure of fluid dynamics is pressure difference within the cardiac vascular structure.

10. The method of claim 8 , wherein the measure of fluid dynamics is fractional flow reserve within the cardiac vascular structure.

11. A method for computing image-derived biomarkers, the method comprising:

receiving image data comprising a plurality of voxels defining a three-dimensional image volume representative of an anatomical region of interest and comprising a plurality of voxels;

determining an anatomical structure related to a biomarker of interest;

selecting an intensity model based on the anatomical structure related to the biomarker of interest, wherein the intensity model is a machine learning model trained using intensity features of previously acquired image data;

applying the intensity model on a per-voxel basis across the three-dimensional image volume to determine a weighting value for each voxel indicating a likelihood that the voxel includes a portion of the anatomical structure; and

deriving information related to the biomarker of interest without prior creation of a segmented mesh from the image data by performing one or more modeling computations directly on the image data, wherein the weighting values are used to limit the modeling computations to voxels comprising the anatomical features related to the biomarker of interest.

12. The method of claim 11 , wherein determining the anatomical structure related to the biomarker of interest comprises:

receiving user input defining a medical condition; and

selecting the anatomical structure and the biomarker of interest based on the medical condition.

13. The method of claim 11 , wherein the intensity model is selected based on a database of past history from other patients and annotated models for how different intensity values correspond to presence of the anatomical structure.

14. The method of claim 11 , further comprising:

determining one or more study-specific input values defining seed points for the intensity model,

wherein the intensity model uses the one or more study-specific input values in determining the plurality of weighting values.

15. The method of claim 14 , wherein the one or more study-specific input values are automatically selected based on previous studies related to the anatomical structure.

16. The method of claim 14 , wherein the one or more study-specific input values are determined based on user input via a graphical user interface (GUI).

17. The method of claim 11 , further comprising:

generating an approximate segmentation of the anatomical structure based on results of applying the computational model to the image data.

18. The method of claim 17 , wherein the approximate segmentation is a binary mask.

19. The method of claim 17 , wherein the approximate segmentation is a gray-level mask representative of segmentation accuracy for the anatomical region of interest.

20. The method of claim 17 , further comprising:

using the approximate segmentation to iteratively apply the computational model to the image data; and

refining the approximate segmentation during each iteration based on results of applying the computational model to the image data.

21. A system for computing image-derived biomarkers, the system comprising:

a scanner device configured to acquire image data defining a three-dimensional image volume representative of an anatomical region of interest and comprising a plurality of voxels; and

a computer operably coupled to the scanner device and configured to:

apply an intensity model to assign a weighting value for each voxel indicating a likelihood that the voxel includes anatomical features related to a biomarker of interest, wherein the intensity model is a machine learning model trained using intensity features of previously acquired image data, and

derive information related to the biomarker of interest without prior creation of a segmented mesh from the image data by performing one or more modeling computations directly on the image data, wherein the weighting values are used to limit the modeling computations to voxels comprising the anatomical features related to the biomarker of interest.