Computer assisted diagnosis (CAD) of cancer using multi-functional, multi-modal in-vivo magnetic resonance spectroscopy (MRS) and imaging (MRI)

1. A method of constructing a classifier to detect cancer sub-clusters in an organ using a dataset containing in vivo high resolution magnetic resonance imaging (MRI) data, comprising the steps of:

a. preprocessing the MRI data by correcting bias field inhomogeneity and non-linear MRI artifacts to create a corrected MRI scene;

b. rigidly or non-rigidly registering determined correspondences of whole-mount histological sections (WMHS) and MRI data via a Combined Feature Ensemble Mutual Information (COFEMI) technique, thereby obtaining cancerous and non-cancerous regions in the preprocessed MRI data;

c. extracting features on a per-voxel basis from the MRI scene;

d. embedding the extracted feature into a low dimensional Eigen space using manifold learning techniques on a per-voxel basis, thereby non-linearly reducing the dimensionality of the extracted image features;

e. training a classifier on the reduced dimensional Eigen feature to discriminate a voxel from an MRI scene as cancerous or normal; and

f. classifying each voxel in the MRI scene based on its reduced dimensional Eigen feature as cancerous or normal, thereby detecting cancer.

2. The method of claim 1 , wherein the dataset comprises magnetic resonance spectroscopy (MRS).

3. The method of claim 1 , wherein the dataset comprises T2-weighted (T2w) MRI.

4. The method of claim 1 , wherein the dataset comprises diffusion weighted (DWI) MRI.

5. The method of claim 1 , wherein the dataset comprises dynamic contrast-enhanced (DCE) MRI.

6. The method of claim 1 wherein each of the extracted image features is selected from a group consisting of a metabolic feature, a statistical feature, a Heralick co-occurrence feature, a Gabor feature, a local binary pattern feature, an apparent diffusion coefficient map feature, an inherent kinetic feature, or their combination.

7. The method of claim 1 , wherein the step of embedding the extracted feature is by locally linear embedding, graph embedding, isometric mapping, or their combination via consensus embedding.

8. The method of claim 1 , wherein the steps of training and classifying the reduced dimensional Eigen feature are by decision trees, probabilistic boosting trees, support vector machines, hierarchical clustering, k-means clustering, mean shift clustering, or a combination thereof.

9. The method of claim 1 , wherein the organ is a prostate.

10. A method of combining multiple magnetic resonance imaging protocols to detect cancer sub-clusters in an organ using a dataset containing in vivo high resolution MRI data, comprising the steps of:

a. preprocessing the MRI data by correcting bias field inhomogeneity and non-linear MR artifacts for each MRI protocol to create corrected multi-protocol MRI scenes;

b. rigidly or non-rigidly registering determined correspondences of whole-mount histological sections (WMHS) and multi-protocol MRI scenes via a Combined Feature Ensemble Mutual Information (COFEMI) technique, thereby obtaining cancerous and non-cancerous regions on multi-protocol MRI data;

c. extracting features on a per-voxel basis from each multi-protocol MRI scene;

d. embedding each extracted multi-protocol MRI scene feature into a low dimensional Eigen space using manifold learning techniques on a per-voxel basis, thereby non-linearly reducing the dimensionality of the extracted multi-protocol MRI scene feature;

e. concatenating low dimensional Eigen vectors from each MRI protocol scene on a per-voxel basis, thereby combining features of the reduced dimensional multi-protocol MRI scene;

f. training a classifier on the combined reduced dimensional Eigen feature to discriminate a voxel from a multi-protocol MRI scene as cancerous or normal; and

g. classifying each voxel in the multi-protocol MRI scene based on its reduced dimensional Eigen feature as cancerous or normal, thereby detecting cancer.

11. The method of claim 10 , wherein the dataset comprises MRS, DCE, DWI, and T2w MRI data.

12. The method of claim 10 , wherein each of the extracted image features is selected from a group consisting of a metabolic feature, a statistical feature, a Haralick co-occurrence feature, a Gabor feature, a local binary pattern feature, an apparent diffusion coefficient map feature, an inherent kinetic feature, or their combination.

13. The method of claim 10 , wherein the step of embedding the extracted feature is by locally linear embedding, graph embedding, isometric mapping, or their combination via consensus embedding.

14. The method of claim 10 , wherein the steps of training and classifying the reduced dimensional Eigen feature are by decision trees, probabilistic boosting trees, support vector machines, hierarchical clustering, k-means clustering, mean shift clustering, or a combination thereof.

15. The method of claim 10 , wherein the organ is a prostate.

16. A method of combining multiple magnetic resonance imaging protocols to identify high grade cancer sub-clusters in an organ using a dataset containing in vivo high resolution MRI data, comprising the steps of:

a. preprocessing the MRI data by correcting bias field inhomogeneity and non-linear MRI artifacts to create a corrected MRI scene;

b. rigidly or non-rigidly registering determined correspondences of whole-mount histological sections (WMHS) and MRI data via a Combined Feature Ensemble Mutual Information (COFEMI) technique, thereby obtaining cancerous and non-cancerous regions in the MRI data;

c. extracting features on a per-voxel basis from the MRI scene;

d. embedding the extracted feature into a low dimensional Eigen space using manifold learning techniques on a per-voxel basis, thereby non-linearly reducing the dimensionality of the extracted image features;

e. concatenating low dimensional Eigen vectors from each MRI protocol scene on a per-voxel basis, thereby combining features of the reduced dimensional multi-protocol MRI scene;

f. training a classifier on the combined reduced dimensional Eigen feature to discriminate a voxel from an MRI scene as high- or low-grade cancer; and

g. classifying each voxel in the MRI scene based on its reduced dimensional Eigen feature as high- or low-grade, thereby identifying high-grade cancer.

17. The method of claim 16 , wherein the dataset comprises MRS, DCE, DWI, and T2w MRI data.

18. The method of claim 16 , wherein each of the extracted image features is selected from a group consisting of a metabolic feature, a statistical feature, a Haralick co-occurrence feature, a Gabor feature, a local binary pattern feature, an apparent diffusion coefficient map feature, an inherent kinetic feature, or their combination.

19. The method of claim 16 , wherein the step of embedding the extracted feature is by locally linear embedding, graph embedding, isometric mapping, or their combination via consensus embedding.

20. The method of claim 16 , wherein the steps of training and classifying the reduced dimensional Eigen feature are by decision trees, probabilistic boosting trees, support vector machines, hierarchical clustering, k-means clustering, mean shift clustering, or a combination thereof.

21. The method of claim 16 , wherein the organ is a prostate.