Autosegmentation/autocontouring system and method

• US 6,249,594 B1
• Filed: 05/29/1998
• Issued: 06/19/2001
• Est. Priority Date: 03/07/1997
• Status: Expired due to Term

• Current Assignee: IMPAC Medical Systems, Inc. (Elekta AB)
• Original Assignee: Computerized Medical Systems, Inc. (Elekta AB)
• Inventors: Hibbard, Lyn
• Primary Examiner(s): Johns, Andrew W.
• Assistant Examiner(s): NAKHJAVAN, SHERVIN K
• Application Number:
  US09/087,254
• Time in Patent Office: 1,117 Days
• Field of Search: 382/128, 382/130, 382/199, 382/173, 382/180, 382/155, 382/159, 382/256, 128/922
• US Class Current: 382/128
• CPC Class Codes: A61N 5/103 Treatment planning systems
• CPC Class Codes: G06T 2207/10081 Computed x-ray tomography [CT]
• CPC Class Codes: G06T 2207/20056 Discrete and fast Fourier transform, [DFT, FFT]
• CPC Class Codes: G06T 2207/30004 Biomedical image processing
• CPC Class Codes: G06T 2207/30061 Lung
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/12 Edge-based segmentation
• CPC Class Codes: G06T 7/143 involving probabilistic approaches, e.g. Markov random field [MRF] modelling

Abstract

A system and method is disclosed for automatically computing contours representing the boundaries of objects in three-dimensional tomographic images that may be formed by computed tomography (“CT”), magnetic resonance imaging (“MRI”), positron emission tomography (“PET”), single proton emission computed tomography (“SPECT”), or other appropriate methods. The system and method begin with a sample region of the object'"'"'s interior and the single region is expanded in a step-wise fashion. At each step, a contour maximally matching the region'"'"'s current edge, local gray-level gradient maxima, and prior contour shapes is determined. Upon completion of region expansion, the object contour is set to that step-contour having the maximum value of an objective function summing contributions from region edges, gradient edges, and prior shapes. Both the region expansion and the boundary contour determination are formulated such that there is a guaranteed average minimum error in the determination of the contours. This contour is represented as a parametric curve in which the contour size and shape are specified by the values of the parameters. These parameters are independent variables of the objective function. The parameters also are considered to be random variables capable of encoding a distribution of contour shapes, and by assuming a particular distribution, the contribution of shape constraints to the object function can be computed. The resulting contour corresponds to the set of parameters for which the objective function is a maximum.

Claims

1. A method for autocontouring a two-dimensional representation of a predetermined bounded object that may be part of an image created by a predetermined modality, with the image being defined by a plurality of pixels having property values representative of what is on the image, comprising the steps of:

(a) generating within the object an initial region of interest (ROI) with a boundary defined by a first supervised classifier based on properties of pixels inside the object compared with properties of pixels outside the object;

(b) generating a second supervised classifier based on the properties of pixels inside the ROI;

(c) expanding the ROI boundary using the second supervised classifier by evaluating the properties of a layer of pixels adjacent to a current exterior boundary of the ROI based on comparing pixel properties of each pixel in the current exterior layer to the properties of the pixels within the ROI, reclassifying the pixels in the current exterior layer as pixels to be included as part of the pixels within the ROI if the properties of discrete pixels in the current exterior layer substantially match in a predetermined manner the properties of the pixels within the ROI, and re-numerating the boundary of the ROI based on any reclassified pixels;

(d) generating a parametric representation of a contour based at least in part on two-dimensional coordinates of a predetermined number of points on the ROI boundary determined in step (c);

(e) generating an optimal objective function and a corresponding parametric contour;

(f) repeating steps (c)-(e) one exterior pixel layer at a time until a maximum match is captured;

(g) selecting a largest-valued optimal objective function that has been generated according to steps (e) and (f); and

(h) creating a contour based of the objective function selected at step (g).

2. The method as recited in claim 1, wherein the predetermined modality generates contours in tomographic images.

3. The method as recited in claim 2, wherein the predetermined modality includes X-ray computed tomography (CT).

4. The method as recited in claim 2, wherein the predetermined modality includes magnetic resonance imaging (MRI).

5. The method as recited in claim 1, wherein the parametric representation is in a form of a finite Fourier series of two-dimensional coordinates that define the boundary of the ROI.

6. The method as recited in claim 5, wherein the finite Fourier series includes the use of a predetermined number of Fourier harmonics to minimize error between points the contour and points on the boundary of the ROI.

7. The method as recited in claim 6, wherein the number of Fourier harmonics relates to a length of the boundary.

8. A method for autocontouring a two-dimensional representation of a predetermined bounded object that may be part of an image created by a predetermined modality, with the image being defined by a plurality of pixels having property values representative of what is on the images, comprising the steps of:

(a) generating within the object an initial region of interest (ROI) with a boundary;

(b) generating a supervised classifier based on predetermined properties of pixels inside of the ROI and properties of pixels outside of the ROI for evaluating if pixels in a layer adjacent to a current exterior boundary of the ROI should be included as part of an interior of the ROI;

(c) expanding the boundary of the ROI using the supervised classifier by iteratively evaluating exterior layers of pixels adjacent to the boundary of the ROI until there is maximum matching of the ROI boundary and the object boundary; and

(d) generating a contour for the object according to the maximum expanded ROI boundary based on a maximization of an objective function.

9. The method as recited in claim 8, wherein the predetermined modality generates contours in tomographic images.

10. The method as recited in claim 9, wherein predetermined modality includes X-ray computed tomography (CT).

11. The method as recited in claim 9, wherein predetermined modality includes magnetic resonance imaging (MRI).

22. The method as recited in claim 9, wherein the predetermined modality includes magnetic resonance imaging (MRI).

12. A method for autocontouring a three-dimensional representation of a predetermined bounded object that may be part of an image created a series of two-dimensional images formed by a predetermined modality, with each of the two-dimensional images being defined by a plurality of pixels having property values representative of what is on the image, comprising the steps of:

(a) generating within an object on a single image an initial region of interest (ROI) with a boundary defined by a first supervised classifier based on properties of pixels inside the object on the single image compared with properties of pixels of with the same object in an adjacent image;

(b) generating a second supervised classifier based on the properties of pixels inside the ROI;

(c) expanding the ROI boundary using the second supervised classifier by evaluating the properties of a layer of pixels adjacent to a current exterior boundary of the ROI based on comparing pixel properties of each pixel of the current exterior layer to the properties of the pixels within the ROI, reclassifying the pixels in the current exterior layer as pixels to be included as part of the pixels within the ROI if the properties of discrete pixels in the current exterior layer substantially match in a predetermined manner the properties of the pixels within the ROI, and re-numerating the boundary of the ROI based on any reclassified pixels;

(d) generating a parametric representation of a contour based at least in part on two-dimensional coordinates of a predetermined number of points on the ROI boundary determined in step (c);

(e) generating an optimal objective function and a corresponding parametric contour,

(f) repeating steps (c)-(e) one exterior pixel layer at a time until a maximum match is captured;

(g) selecting a largest-valued optimal objective function that has been generated according to steps (e) and (f);

(h) saving the objective function selected at step (g);

(i) repeating steps (a)-(h) for each two-dimensional image defining the three-dimensional representation of the object; and

(j) creating a three-dimensional contour based of the objective functions selected and saved at steps (g) and (h).

13. The method as recited in claim 12, wherein the predetermined modality generates contours in tomographic images.

14. The method as recited in claim 13, wherein the predetermined modality includes X-ray computed tomography (CT).

15. The method as recited in claim 13, wherein the predetermined modality includes magnetic resonance imaging (MRI).

16. The method as recited in claim 12, wherein the parametric representation is in a form of a finite Fourier series of two-dimensional coordinates that define the boundary of the ROI.

17. The method as recited in claim 16, wherein the finite Fourier series includes the use of a predetermined number of Fourier harmonics to minimize error between points the contour and points on the boundary of the ROI.

18. The method as recited in claim 17, wherein the number of Fourier harmonics relates to a length of the boundary.

19. A method for autocontouring a three-dimensional representation of a predetermined bounded object that may be part of an image created a series of two-dimensional images formed by a predetermined modality, with each of the two-dimensional images being defined by a plurality of pixels having property values representative of what is on the image, comprising the steps of:

(a) generating within the object on a single image an initial region of interest (ROI) with a boundary;

(b) generating a supervised classifier based on predetermined properties of pixels inside of the ROI and properties of pixels outside of the ROI for evaluating if pixels in a layer adjacent to a current exterior boundary of the ROI should be included as part of an interior of the ROI;

(c) expanding the boundary of the ROI using the supervised classifier by iteratively evaluating exterior layers of pixels adjacent to the boundary of the ROI until there is maximum matching of the ROI boundary and the object boundary;

(d) generating an optimal objective function for the maximum matching ROI boundary;

(e) saving the optimal objective function generated at step (d);

(f) repeating steps (c)-(e) for each two-dimensional image defining the three-dimensional representation of the object; and

(j) generating a three-dimensional set of contours based of the objective functions generated and saved at steps (d) and (e).

20. The method as recited in claim 19, wherein the predetermined modality generates contours in tomographic images.

21. The method as recited in claim 20, wherein the predetermined modality includes X-ray computed tomography (CT).