System and method for image-based quantification of white and brown adipose tissue at the whole-body, organ and body-region levels

1. A method of automatically detecting and quantifying white and brown adipose tissue from an imaging scan of a subject comprising:

providing the imaging scan of the subject wherein the imaging scan is created using computed tomography (CT), positron emission tomography (PET), magnetic resonance imaging (MRI), positron emission tomography/computed tomography (PET/CT), positron emission tomography/magnetic resonance imaging (PET/MRI) or contrast-enhanced ultrasound (CEUS);

automatically detecting a body region of the subject in the imaging scan using extracted convolutional neural network (CNN) features;

segmenting total adipose tissue (TAT) in the body region;

separating and segmenting subcutaneous adipose tissue (SAT) from visceral adipose tissue (VAT) in the body region in the imaging scan of the subject comprising

estimating a SAT-VAT separation boundary;

removing outliers the separation boundary; and

creating a fine SAT-VAT separating surface using three-dimensional (3D) Conditional Random Fields (CRF) using shape, anatomy and appearance cues; and

detecting and segmenting brown adipose tissue (BAT) from other tissue after TAT segmentation in the imaging scan of the subject.

2. The method of claim 1 , wherein the body region detected is selected from the group consisting of an abdominal region and a thorax region.

3. The method of claim 2 , wherein the body region is automatically detected by using a detection algorithm based on deep learning features.

4. The method of claim 1 , wherein the outliers are removed from the boundary using geometric median absolute derivation (MAD) or local outlier scores (LoOS).

5. The method of claim 1 , wherein the detecting and segmenting of BAT step further comprises:

performing automatic seed selection for BAT;

delineating potential BAT regions; and

differentiating BAT regions from non-BAT regions.

6. The method of claim 5 , wherein fixed Hounsfield unit (HU) interval filtering is used to identify TAT.

7. The method of claim 5 , wherein background and foreground seeds are identified during automatic seed selection.

8. The method of claim 5 , wherein image co-segmentation using Random Walk (RW) is used to delineate potential BAT regions.

9. The method of claim 5 , wherein a probabilistic metric based on a combination of total variation and Cramer-Von Mises distances is used to differentiate BAT regions from non-BAT regions.

10. The method of claim 1 , further comprising automatically detecting specific organs comprising:

extracting 3D convolutional neural network (CNN) features from source data;

transforming 3D CNN features from source data to target data by applying Geodesic Flow Kernal (GFK) to the 3D CNN features; and

localizing the organ in a bounding volume using Random Forest;

wherein the target data is organ detection in 3D CT scans.

11. The method of claim 1 , wherein the imaging scan is selected from the group consisting of a positron emission tomography/computed tomography (PET/CT) scan, a positron emission tomography/magnetic resonance imaging scan (PET/MRI) and a contrast-enhanced ultrasound (CEUS) scan.

12. A method of creating a risk profile of a subject by automatically detecting and quantifying white and brown adipose tissue from an imaging scan of the subject comprising:

the imaging scan of the subject wherein the imaging scan is created using computed tomography (CT), positron emission tomography (PET), magnetic resonance imaging (MRI), positron emission tomography/computed tomography (PET/CT), positron emission tomography/magnetic resonance imaging (PET/MRI) or contrast-enhanced ultrasound (CEUS);

automatically detecting a body region of the subject in the imaging scan using extracted convolutional neural network (CNN) features wherein the body region detected is an abdominal region or a thorax region;

segmenting total adipose tissue (TAT) in the body region;

separating and segmenting subcutaneous adipose tissue (SAT) from visceral adipose tissue (VAT) in the imaging scan of the subject comprising

estimating a SAT-VAT separation boundary;

removing outliers using geometric median absolute derivation (MAD) or local outlier scores (LoOS); and

creating a fine SAT-VAT separating surface using 3D Conditional Random Fields (CRF) using shape, anatomy and appearance cues;

detecting and segmenting brown adipose tissue (BAT) from other tissue after TAT segmentation in the imaging scan of the subject; and

creating a risk profile based on a quantitative amount of VAT and BAT found in the subject.

13. The method of claim 12 , wherein the detecting and segmenting brown adipose tissue (BAT) from other tissue step further comprising:

performing automatic seed selection for BAT;

performing image co-segmentation; and

differentiating BAT regions from non-BAT regions.

14. The method of claim 12 , further comprising automatically detecting specific organs comprising:

extracting 3D convolutional neural network (CNN) features from source data;

transforming 3D CNN features from source data to target data by applying Geodesic Flow Kernal (GFK) to the 3D CNN features; and

localizing the organ in a bounding volume using Random Forest;

wherein the target data is organ detection in 3D CT scans.

15. A method of automatically detecting and quantifying white and brown adipose tissue from an imaging scan of a subject comprising:

providing the imaging scan of the subject wherein the imaging scan is created using positron emission tomography/computed tomography (PET/CT);

automatically detecting a body region of the subject in the imaging scan using extracted convolutional neural network (CNN) features;

segmenting total adipose tissue (TAT) in the body region;

separating and segmenting subcutaneous adipose tissue (SAT) from visceral adipose tissue (VAT) in the imaging scan of the subject comprising:

estimating a SAT-VAT separation boundary;

removing outliers using geometric median absolute derivation (MAD) or local outlier scores (LoOS); and

creating a fine SAT-VAT separating surface using 3D Conditional Random Fields (CRF) using shape, anatomy and appearance cues;

detecting and segmenting brown adipose tissue (BAT) from other tissue after TAT segmentation in the imaging scan of the subject comprising:

performing automatic seed selection for BAT;

performing image co-segmentation; and

differentiating BAT regions from non-BAT regions.

16. The method of claim 15 , further comprising automatically detecting specific organs comprising:

extracting 3D convolutional neural network (CNN) features from source data;

transforming 3D CNN features from source data to target data by applying Geodesic Flow Kernal (GFK) to the 3D CNN features; and

localizing the organ in a bounding volume using Random Forest;

wherein the target data is organ detection in 3D CT scans.