Predicting prostate cancer recurrence in pre-treatment prostate magnetic resonance imaging (MRI) with combined tumor induced organ distension and tumor radiomics

1. An apparatus for predicting prostate cancer (PCa) recurrence, the apparatus comprising:

a processor;

a memory;

an input/output (I/O) interface;

a set of circuits including an image acquisition circuit, a segmentation circuit, a registration circuit, a distension feature circuit, a radiomics circuit, a combined tumor induced organ distension with tumor radiomics (COnTRa) circuit, and a display circuit; and

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

where the memory is configured to store a first pre-treatment image of a region of tissue demonstrating PCa, a second pre-treatment image of the region of tissue, a recurrence-negative (C−) median template, and a surface of interest (SOI) mask, where the region of tissue includes a prostate capsule, the first pre-treatment image having a plurality of voxels, and the second pre-treatment image having a plurality of voxels, a voxel having an intensity;

where the image acquisition circuit is configured to access the first pre-treatment image and the second pre-treatment image;

where the segmentation circuit is configured to:

generate a first segmented prostate by segmenting the prostate capsule represented in the first pre-treatment image, and;

generate a second segmented prostate by segmenting the prostate capsule represented in the second pre-treatment image;

where the registration circuit is configured to:

generate a registered prostate by registering the first segmented prostate with the SOI mask;

generate a patient-specific SOI mask from the registered prostate and the SOI mask; and

generate a patient-specific SOI mesh from the patient-specific SOI mask;

where the distension feature circuit is configured to:

extract a set of distension features from the patient-specific SOI mesh; and

compute a first probability of PCa recurrence based on the set of distension features;

where the radiomics circuit is configured to:

extract a set of radiomics features from the second pre-treatment image; and

compute a second probability of PCa recurrence based on the set of radiomics feature;

where the COnTRa circuit is configured to:

compute a joint probability that the region of tissue will experience PCa recurrence based on the first probability and the second probability; and

where the display circuit is configured to display the joint probability.

2. The apparatus of claim 1 , where the SOI mask is a spatially contextual surface of interest that defines a region of differential distension between recurrence-positive (C+) and C− regions of tissue.

3. The apparatus of claim 1 , where the first pre-treatment image is a T2w magnetic resonance imaging (MRI) image of a region of tissue demonstrating PCa.

4. The apparatus of claim 1 , where the second pre-treatment image is a T2w apparent diffusion coefficient (ADC) dynamic contrast enhanced (DCE) MRI image of the region of tissue.

5. The apparatus of claim 1 , where the registration circuit is configured to register the SOI mask with the first pre-treatment image using an affine registration technique and a B-spline registration technique.

6. The apparatus of claim 1 , where the patient-specific SOI mesh includes a plurality of vertices.

7. The apparatus of claim 1 , where the set of distension features includes a Gaussian curvature (θ) feature, and a surface normal orientation (Φ) feature represented in a spherical coordinate system.

8. The apparatus of claim 7 , where the set of distension features further includes a θ kurtosis feature, a Φ skewness feature, a Φ standard deviation feature, and a Φ mean feature computed from the θ feature and the Φ feature.

9. The apparatus of claim 1 , where the distension feature circuit further comprises a machine learning component configured to compute the first probability based on the set of distension features.

10. The apparatus of claim 9 , where the machine learning component is configured as a random forest (RF) classifier having a depth of two and 1000 trees.

11. The apparatus of claim 1 , where the set of radiomics features includes a subset of first order statistical features, a subset of Haralick features, and a subset of Gabor features.

12. The apparatus of claim 1 , where the radiomics circuit further comprises a machine learning component configured to compute the second probability based on the set of radiomics features.

13. The apparatus of claim 12 , where the machine learning component is configured as a random forest (RF) classifier having a depth of two and 1000 trees.

14. The apparatus of claim 1 , the set of circuits further comprising an atlas circuit configured to:

generate a recurrence-positive (C+) atlas;

generate a C− atlas;

generate a registered atlas by registering the C+ atlas with the C− atlas; and

generate the SOI mask from the registered atlas.

15. 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 prostate cancer (PCa) recurrence, the method comprising:

accessing a first pre-treatment radiological image of a region of tissue demonstrating PCa;

accessing a second pre-treatment radiological image of the region of tissue;

generating a first segmented prostate by automatically segmenting a prostate capsule represented in the first pre-treatment radiological image;

generating a second segmented prostate by automatically segmenting the prostate capsule represented in the first pre-treatment radiological image;

generating a registered segmented prostate by registering the first segmented prostate with a surface of interest (SOI) mask;

generating a patient-specific SOI mesh from registered segmented prostate;

extracting a set of distension features from the patient-specific SOI mesh;

providing a first machine learning classifier the set of distension features;

extracting a set of radiomic features from the second segmented prostate;

providing a second machine learning classifier the set of radiomic features;

receiving, from the first machine learning classifier, a first probability that the region of tissue will experience PCa recurrence based, at least in part, on the set of distension features;

receiving, from the second machine learning classifier, a second probability that the region of tissue will experience PCa recurrence based, at least in part, on the set of radiomic features;

computing a combined probability that the region of tissue will experience PCa recurrence based on the first probability and the second probability;

upon detecting that the combined probability is greater than a threshold probability:

classifying the region of tissue as likely to experience PCa recurrence;

upon detecting that the combined probability is less than or equal to the threshold probability:

classifying the region of tissue as unlikely to experience PCa recurrence;

displaying the classification and at least one of the combined probability, the first probability, the second probability, the set of radiomics features, the set of distension features, the first pre-treatment radiological image, or the second pre-treatment radiological image.

16. The non-transitory computer-readable storage device of claim 15 , where the first pre-treatment radiological image is a T2w apparent diffusion coefficient (ADC) dynamic contrast enhanced (DCE) magnetic resonance imaging (MRI) image of the region of tissue, and where the second pre-treatment radiological image is a T2w MRI image.

17. The non-transitory computer-readable storage device of claim 15 , where the set of radiomics features includes a subset of first order statistical features, a subset of Haralick features, and a subset of Gabor features.

18. The non-transitory computer-readable storage device of claim 15 , where the set of distension features includes a Gaussian curvature (θ) feature, a surface normal orientation (Φ) feature represented in a spherical coordinate system, and a θ kurtosis feature, a Φ skewness feature, a Φ standard deviation feature, and a Φ mean feature computed from the θ feature and the Φ feature.

19. The non-transitory computer-readable storage device of claim 15 , where the first machine learning classifier is a random forest (RF) classifier having a depth of two and 1000 trees, and where the second machine learning classifier is a random forest (RF) classifier having a depth of two and 1000 trees.

20. A non-transitory computer-readable storage device storing instructions that when executed by a processor control the processor to perform operations, the operations including:

accessing a first pre-treatment image of a region of tissue demonstrating prostate cancer (PCa), where the first pre-treatment image includes a segmented prostate capsule;

accessing a second pre-treatment image of the region of tissue, where the second pre-treatment image includes the segmented prostate capsule;

generating a registered segmented prostate capsule by registering the segmented prostate capsule represented in the first pre-treatment image to a recurrence negative (C−) median template;

registering a surface of interest (SOI) mask to the registered segmented prostate capsule;

generating a patient-specific SOI mask from the registered SOI mask;

generating a patient-specific SOI mesh from the patient-specific SOI mask;

extracting a set of organ distension features from the patient-specific mesh, where the set of organ distension features a Gaussian curvature (θ) feature, a surface normal orientation (Φ) feature represented in a spherical coordinate system, and a θ kurtosis feature, a Φ skewness feature, a Φ standard deviation feature, and a Φ mean feature computed from the θ feature and the Φ feature;

computing a first probability that the region of tissue will experience PCa recurrence based, at least in part, on the set of organ distension features;

extracting a set of radiomic features from the segmented prostate capsule represented in the second pre-treatment image, where the set of radiomic features includes a subset of first order statistical features, a subset of Haralick features, and a subset of Gabor features;

computing a second probability that the region of tissue will experience PCa recurrence based, at least in part, on the set of radiomics features;

computing a joint-probability that the region of tissue will experience PCa recurrence based on the first probability and the second probability;

classifying the region of tissue as likely to experience biochemical recurrence (BCR) or unlikely to experience BCR based, at least in part, on the joint probability; and

displaying the classification, the joint probability, the set of organ distension features, the set of radiomic features, the patient-specific SOI mesh, the patient-specific SOI mask, the first pre-treatment image, or the second pre-treatment image.