COMPUTER AIDED DIAGNOSTIC SYSTEM INCORPORATING APPEARANCE ANALYSIS FOR DIAGNOSING MALIGNANT LUNG NODULES

• US 20120201445A1
• Filed: 02/08/2012
• Published: 08/09/2012
• Est. Priority Date: 02/08/2011
• Status: Active Grant

• Current Assignee: University of Louisville
• Original Assignee: University of Louisville
• Inventors: El-Baz, Ayman S.
• Granted Patent: US 9,014,456 B2
• Time in Patent Office: Days
• US Class Current: 382/133
• CPC Class Codes: G06F 18/295 Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
• CPC Class Codes: G06T 2207/10081 Computed x-ray tomography [CT]
• CPC Class Codes: G06T 2207/20081 Training; Learning
• CPC Class Codes: G06T 2207/30064 Lung nodule
• CPC Class Codes: G06T 7/0014 using an image reference approach
• CPC Class Codes: G06V 10/7557 based on appearance, e.g. active appearance models [AAM]
• CPC Class Codes: G06V 10/85 Markov-related models; Markov random fields
• CPC Class Codes: G06V 2201/031 of internal organs

Abstract

A computer aided diagnostic system and automated method diagnose lung cancer through modeling and analyzing the visual appearance of pulmonary nodules. A learned appearance model used in such analysis describes the appearance of pulmonary nodules in terms of voxel-wise conditional Gibbs energies for a generic rotation and translation invariant second-order Markov-Gibbs random field (MGRF) model of malignant nodules with analytically estimated characteristic voxel neighborhoods and potentials.

Claims

1. A method of classifying a pulmonary nodule, the method comprising:

receiving image data associated with a chest scan;

segmenting image data associated with lung tissue from the image data associated with the chest scan;

equalizing the segmented image data;

segmenting image data associated with a pulmonary nodule from the equalized and segmented image data; and

classifying the pulmonary nodule as benign or malignant by applying a learned appearance model to the segmented image data associated with the pulmonary nodule, wherein the learned appearance model is based upon visual appearances of a plurality of known pulmonary nodules.

2. The method of claim 1, wherein the learned appearance model comprises a generic rotation and translation invariant second-order Markov Gibbs Random Field (MGRF) model of malignant nodules with analytically estimated characteristic voxel neighborhoods and potentials.

3. The method of claim 2, wherein the learned appearance model models voxel-wise conditional Gibbs energies.

4. The method of claim 1, wherein classifying the pulmonary nodule as benign or malignant by applying a learned appearance model to the segmented image data associated with the pulmonary nodule includes determining an intensity variation between pairs of voxels of a plurality of voxels included in the segmented image data associated with the pulmonary nodule, and wherein classifying the pulmonary nodule as benign or malignant is based at least in part on the determined intensity variations.

5. The method of claim 4, wherein the intensity variations in the neighborhood of voxels is based at least in part on a gray level associated with each voxel of the neighborhood of voxels, and wherein the gray level associated with each voxel corresponds to a Hounsfield value associated with the voxel.

6. The method of claim 1, wherein the segmented image data associated with the pulmonary nodule includes data indicating a Hounsfield value for each voxel of a plurality of voxels associated with the pulmonary nodule, and wherein classifying the pulmonary nodule as benign or malignant is based at least in part on variances in Hounsfield values between voxel pairs of the plurality of voxels associated with the pulmonary nodule.

7. The method of claim 6, wherein classifying the pulmonary nodule as benign or malignant by applying a learned appearance model to the segmented image data associated with the pulmonary nodule includes determining a Gibbs energy associated with the pulmonary nodule based on the variance in Hounsfield values between voxels of the voxel pairs associated with the pulmonary nodule.

8. An apparatus, comprising:

at least one processor; and

program code configured to be executed by the at least one processor to classify a pulmonary nodule as benign or malignant by applying a learned appearance model to the segmented image data associated with the pulmonary nodule, wherein the learned appearance model is based upon visual appearances of a plurality of known pulmonary nodules.

9. A program product, comprising:

a computer readable storage medium; and

program code stored on the computer readable storage medium and configured upon execution to classify a pulmonary nodule as benign or malignant by applying a learned appearance model to the segmented image data associated with the pulmonary nodule, wherein the learned appearance model is based upon visual appearances of a plurality of known pulmonary nodules.

10. A method of generating a learned appearance model for classifying pulmonary nodules as benign or malignant, comprising:

normalizing image data associated with a plurality of known pulmonary nodules; and

processing the normalized image data to learn a 3D appearance of the plurality of known pulmonary nodules to generate the learned appearance model.

11. The method of claim 10, wherein processing the normalized image data includes learning a spatial distribution of Hounsfield values of the known pulmonary nodules.

12. The method of claim 11, wherein processing the normalized image data includes learning a distribution of Hounsfield values of known malignant and benign pulmonary nodules.

13. The method of claim 10, wherein the learned appearance model comprises a generic rotation and translation invariant second-order Markov Gibbs Random Field (MGRF) model of malignant nodules with analytically estimated characteristic voxel neighborhoods and potentials.

14. The method of claim 13, wherein the learned appearance model models voxel-wise conditional Gibbs energies.

15. The method of claim 10, wherein the normalized image data includes data indicating a plurality of voxels associated with each known pulmonary nodule, wherein each voxel includes data indicating an associated intensity value and wherein processing the normalized image data to learn a 3D appearance of the plurality of known pulmonary nodules to generate the learned appearance model includes analyzing the plurality of voxels associated with each known pulmonary nodule to determine an intensity difference between at least a portion of the voxels associated with each known pulmonary nodule.

16. The method of claim 15, wherein processing the normalized image data to learn a 3D appearance of the plurality of known pulmonary nodules to generate the learned appearance model includes determining a Gibbs energy value associated with each known pulmonary nodule based at least in part on the intensity values of voxels associated with each known pulmonary nodule.