Kernel sparse models for automated tumor segmentation

Patent metadata

Field	Value
Publication number	US-2016005183-A1
Application number	US-201514853617-A
Country	US
Kind code	A1
Filing date	Sep 14, 2015
Priority date	Mar 14, 2013
Publication date	Jan 7, 2016
Grant date	—

Abstract

A robust method to automatically segment and identify tumor regions in medical images is extremely valuable for clinical diagnosis and disease modeling. In various embodiments, an efficient algorithm uses sparse models in feature spaces to identify pixels belonging to tumorous regions. By fusing both intensity and spatial location information of the pixels, this technique can automatically localize tumor regions without user intervention. Using a few expert-segmented training images, a sparse coding-based classifier is learned. For a new test image, the sparse code obtained from every pixel is tested with the classifier to determine if it belongs to a tumor region. Particular embodiments also provide a highly accurate, low-complexity procedure for cases when the user can provide an initial estimate of the tumor in a test image.

First claim

What is claimed is:

1 . A method of segmenting a tumor region in an image, the method being implemented via execution of computer instructions configured to run at one or more processing modules and configured to be stored at one or more non-transitory memory storage modules, the method comprising:
computing a kernel sparse code for each pixel of at least a portion of the image; and identifying, using a classifier, each pixel belonging to the tumor region.

2 . The method of claim 1 , wherein:
the image is a T1-weighted contrast-enhanced MRI scan.

3 . The method of claims 1 , wherein:
the tumor region represents at least a portion of a brain tumor.

4 . The method of claims 1 , wherein:
the tumor region represents at least a portion of a GBM tumor.

5 . The method of claims 1 , further comprising:
displaying at least a portion of the tumor region on a screen.

6 . The method of claim 1 , further comprising:
computing one or more learned dictionaries using kernel K-lines clustering, wherein:
computing the kernel sparse code for each pixel comprises computing the kernel sparse code for each pixel using the one or more learned dictionaries.

7 . The method of claim 6 , wherein:
the one or more learned dictionaries comprise a tumor kernel dictionary and a non-tumor kernel dictionary.

8 . The method of claim 7 , wherein
the tumor kernel dictionary and the non-tumor kernel dictionary are each based at least in part on intensity and not on spatial location.

9 . The method of claim 6 , wherein:
the one or more learned dictionaries are based at least in part on both intensity and spatial location.

10 . The method of claims 1 , wherein:
the classifier is a 2-class linear SVM.

11 . The method of claim 1 , further comprising:
computing a first intensity kernel matrix using expert-segmented training images; computing a spatial location kernel matrix using the expert-segmented training images; and fusing the first intensity kernel matrix and the spatial location kernel matrix to obtain a first ensemble kernel matrix.

12 . The method of claim 1 , further comprising:
computing a second ensemble kernel matrix using the image.

13 . The method of claim 1 , further comprising:
computing a first intensity kernel matrix using expert-segmented training images; computing a spatial location kernel matrix using the expert-segmented training images; fusing the first intensity kernel matrix and the spatial location kernel matrix to obtain a first ensemble kernel matrix; computing a kernel dictionary based at least in part on the first ensemble kernel matrix using kernel K-lines clustering; and computing a second ensemble kernel matrix using the image, wherein:
computing the kernel sparse code for each pixel comprises computing the kernel sparse code for each pixel based at least in part on the second ensemble kernel matrix and the kernel dictionary; and
the classifier is a 2-class linear SVM.

14 . The method of claim 1 , wherein:
the classifier is a reconstruction error-based classifier.

15 . The method of claim 1 , further comprising:
computing a first intensity kernel matrix using segmented training images.

16 . The method of claim 1 , further comprising:
computing a second intensity kernel matrix using the image.

17 . The method of claim 1 , further comprising:
computing a first intensity kernel matrix using expert-segmented training images; computing a tumor kernel dictionary and a non-tumor kernel dictionary, each based at least in part on intensity and not on spatial location, using kernel K-lines clustering; receiving an initialization of the tumor region in the image; and computing a second intensity kernel matrix using the tumor region in the image, wherein:
computing the kernel sparse code for each pixel comprises computing the kernel sparse code for each pixel in the tumor region in the image based at least in part on the second intensity kernel matrix and at least one of the tumor kernel dictionary or the non-tumor kernel dictionary; and
the classifier is a reconstruction error-based classifier.

18 . A system for segmenting a tumor region in an image, the system comprising:
one or more processing modules; and one or more non-transitory memory storage modules storing computing instructions configured to run on the one or more processing modules and perform the acts of:
computing a kernel sparse code for each pixel of at least a portion of the image; and
identifying, using a classifier, each pixel belonging to the tumor region.

19 . The system of claim 18 , wherein the computing instructions are further configured to perform the acts of:
computing a first intensity kernel matrix using expert-segmented training images; computing a spatial location kernel matrix using the expert-segmented training images; fusing the first intensity kernel matrix and the spatial location kernel matrix to obtain a first ensemble kernel matrix; computing a kernel dictionary based at least in part on the first ensemble kernel matrix using kernel K-lines clustering; and computing a second ensemble kernel matrix using the image, wherein:
computing the kernel sparse code for each pixel comprises computing the kernel sparse code for each pixel based at least in part on the second ensemble kernel matrix and the kernel dictionary; and
the classifier is a 2-class linear SVM.

20 . The system of claim 18 , wherein the computing instructions are further configured to perform the acts of:
computing a first intensity kernel matrix using expert-segmented training images; computing a tumor kernel dictionary and a non-tumor kernel dictionary, each based at least in part on intensity and not on spatial location, using kernel K-lines clustering; receiving an initialization of the tumor region in the image; and computing a second intensity kernel matrix using the tumor region in the image, wherein:
computing the kernel sparse code for each pixel comprises computing the kernel sparse code for each pixel in the tumor region in the image based at least in part on the second intensity kernel matrix and at least one of the tumor kernel dictionary and the non-tumor kernel dictionary; and
the classifier is a reconstruction error-based classifier.

Assignees

• Thiagarajan Jayaraman Jayaraman
• Ramamurthy Karthikeyan
• Spanias Andreas
• Frakes David
• Univ Arizona
• Univ Arizona State

Inventors

• Thiagarajan Jayaraman Jayaraman
• Ramamurthy Karthikeyan
• Spanias Andreas
• Frakes David

Classifications

• G06T2207/10088

  Magnetic resonance imaging [MRI] · CPC title

• G16H30/40

  for processing medical images, e.g. editing · CPC title

• G06T7/11

  Region-based segmentation · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• A61B2576/026

  for the brain · CPC title