Computer-aided diagnosis system for thoracic computer tomography images

• US 20030035507A1
• Filed: 08/08/2002
• Published: 02/20/2003
• Est. Priority Date: 08/17/2001
• Status: Active Grant

• Current Assignee: Cad Investments, LLC, Riverain Technologies LLC
• Original Assignee: Riverain Medical Group, LLC
• Inventors: Hsu, Li-Yueh, Yen, H.-Y. Michael, Lure, Fleming Y.-M., Li, Ruiping, Xu, Xin-Wei, Lin, Jyh-Shyan, Martello, Edward A.
• Granted Patent: US 6,795,521 B2
• Time in Patent Office: Days
• US Class Current: 378/4
• CPC Class Codes: G06T 2207/10081 Computed x-ray tomography [CT]
• CPC Class Codes: G06T 2207/20156 Automatic seed setting
• CPC Class Codes: G06T 2207/30064 Lung nodule
• CPC Class Codes: G06T 7/0012 Biomedical image inspection
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/136 involving thresholding
• CPC Class Codes: Y10S 378/901 Computer tomography program or processor

Abstract

A method of detecting and analyzing abnormalities, like lung nodules, in thoracic computer tomography (CT) images uses digital image processing techniques and adaptive computing methods. The techniques include an automatic detection process to detect candidate abnormalities, an image matching process to match CT slices from two different CT scans, and a measurement process that determines parameters of the candidate abnormalities. Final results and processed CT images are displayed on a user interface.

Claims

1. A method of detecting and analyzing suspected abnormalities in thoracic computer tomography (CT) images, comprising the steps of:

performing object processing to segment different anatomical structures in the CT images in at least one of two-dimensional CT slices and three-dimensional CT scans, resulting in at least one segmented image;

processing said at least one segmented image, said processing including at least one of image enhancement, object analysis, object detection, and image matching; and

outputting at least one result of said processing step.

2. The method according to claim 1, wherein said step of performing object processing receives at least first and second CT images as inputs, the method further comprising the step of:

performing volume reformatting on the second CT image using information about the first CT image, prior to the step of performing object processing.

3. The method according to claim 2, wherein the step of performing volume reformatting comprises at least one of volume subsampling and interpolation.

4. The method according to claim 1, wherein the step of performing object processing comprises the steps of:

performing image thresholding on at least one of said CT images;

extracting at least one lung image from the at least one of said CT images based on the results of the image thresholding; and

extracting at least one lung contour from said at least one lung image.

5. The method according to claim 4, wherein the step of image thresholding comprises the steps of:

constructing a gray-level histogram based on the output of the step of performing object processing;

performing least squares fitting of a Gaussian curve to a low-intensity range of said histogram and selecting a background threshold based thereon;

using the background threshold to eliminate low-intensity background noise from the output of the step of performing object processing;

performing voxel threshold determination, comprising the steps of;

generating a plurality of thresholded images using a plurality of possible thresholds; and

performing a statistical correlation between each of the plurality of thresholded images and at least one original image to thereby determine a voxel threshold; and

generating a voxel thresholded image by applying thresholding using the voxel threshold, thereby generating a binary image.

6. The method according to claim 4, wherein the step of extracting at least one lung image comprises the step of:

performing connected component labeling on an output of said step of image thresholding.

7. The method according to claim 4, further comprising the step of:

performing top- and bottom-most slices padding on the output of said step of image thresholding, prior to said step of extracting at least one lung image.

8. The method according to claim 4, further comprising the step of:

removing small objects from the at least one lung image, prior to the step of extracting at least one lung contour.

9. The method according to claim 8, wherein the step of removing small objects comprises the step of:

performing a step of reversed connected component labeling.

10. The method according to claim 4, further comprising the step of:

smoothing the at least one lung contour using an active contour smoothing algorithm.

11. The method according to claim 10, wherein the active contour smoothing algorithm is a snake algorithm.

12. The method according to claim 1, wherein the step of processing comprises a step of image enhancement, and wherein the at least one result output in the outputting step includes an enhanced image.

13. The method according to claim 12, wherein the step of image enhancement comprises at least one step of histogram window leveling performed on the at least one segmented image.

14. The method according to claim 13, wherein said at least one step of histogram window leveling comprises the steps of:

lung area histogram window leveling; and

mediastinum area histogram window leveling;

and wherein the results of the steps of lung area histogram window leveling and mediastinum area window leveling are combined to produce an output image.

15. The method according to claim 1, wherein the step of processing comprises a step of object analysis performed on the at least one segmented image, and wherein the at least one result output in the outputting step comprises at least one of an image, a feature, and a measurement.

16. The method according to claim 15, wherein the step of object analysis includes analysis of at least one of sphericity, size, HU value, and a vessel/nodule tree in the at least one segmented image.

17. The method according to claim 15, wherein the step of object analysis comprises the steps of:

computing a seed position for a consecutive measurement process;

performing gray-level thresholding on the at least one segmented image;

performing morphological shape processing on the output of said gray-level thresholding;

performing connected component labeling on the output of said morphological shape processing;

adjusting contours obtained in said connected component labeling;

reconstructing at least one three-dimensional object based on the adjusted contours; and

performing measurements on the at least one three-dimensional object.

18. The method according to claim 17, wherein the step of computing a seed position comprises the step of:

determining a center of gravity from a predefined local image window.

19. The method according to claim 17, wherein the step of adjusting contours comprises the step of:

performing neighbor profile tracing.

20. The method according to claim 1, wherein the step of processing comprises a step of object detection, and wherein the at least one result output in the outputting step comprises at least one of an image, a feature, and a measurement.

21. The method according to claim 20, wherein the step of object detection includes at least one of image enhancement, initial selection, three-dimensional object grouping, feature extraction, and classification.

22. The method according to claim 20, wherein the step of object detection comprises the steps of:

filtering an input image;

performing gray-level thresholding on the filtered input image;

making an initial nodule selection based on the output of the gray-level thresholding;

extracting features based on the output of the initial nodule selection; and

classifying objects determined in the step of initial nodule selection using the features extracted in the step of extracting features.

23. The method according to claim 22, wherein the step of filtering comprises the step of:

performing enhancement filtering for matching a nodule'"'"'s intensity profile.

24. The method according to claim 22, wherein the step of classifying objects comprises the step of:

analyzing relationships between features arising at different threshold levels with respect to at least one of size, shape, and location.

25. The method according to claim 1, wherein the step of processing comprises the steps of:

object analysis performed on the output of said object processing; and

measuring doubling time based on results of said object analysis;

and wherein the at least one result output in the outputting step comprises a measurement of doubling time.

26. The method according to claim 25, wherein the step of measuring doubling time comprises the step of:

calculating doubling time based on at least two time-differenced CT scans.

27. The method according to claim 1, wherein the step of processing comprises a step of image matching based on the output of said object processing, and wherein the at least one result output in the outputting step comprises at least one of an image and a measurement.

28. The method according to claim 27, wherein the step of image matching includes at least one of slice matching, volume matching, and slice warping, and wherein the step of image matching involves results obtained from at least two CT images.

29. The method according to claim 28, wherein the step of processing comprises a step of slice matching, and wherein the step of slice matching comprises the steps of:

generating a curve from two-dimensional features in each CT image;

computing a gradient for each curve;

correlating gradients of the curves; and

determining a shift distance in one CT image with respect to the other CT image.

30. The method according to claim 27, wherein the step of image matching comprises the steps of:

inputting the outputs of said object processing corresponding to two CT images obtained at different times;

computing a lung area curve for each of the CT images;

fitting the two lung area curves thus computed to each other;

registering lung surface volume based on the results of the step of fitting; and

warping one of the CT images to the other CT image.

31. The method according to claim 30, wherein the step of fitting the two lung area curves comprises the step of:

comparing the two lung area curves to minimize a sum of squared differences.

32. The method according to claim 30, wherein the step of registering lung surface volume comprises the step of:

using an iterative closest point algorithm to determine a transformation by which to transform one CT image onto the other CT image.

33. The method according to claim 32, wherein said step of warping comprises the step of:

transforming the one CT image onto the other CT image using the transformation determined using the iterative closest point algorithm.