Method for the detection and automatic characterization of nodules in a tomographic image and a system of medical imaging by tomodensimetry

• US 7,295,870 B2
• Filed: 11/22/2002
• Issued: 11/13/2007
• Est. Priority Date: 11/23/2001
• Status: Expired due to Fees

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Knoplioch, Jerome Francois, Ferrant, Matthieu, Hourdin, Guy-Michel, Allain, Pascal Raymond
• Primary Examiner(s): Mercader; Eleni Mantis
• Assistant Examiner(s): Roy; Baisakhi
• Application Number:
  US10/302,135
• Publication Number:
  US 20030167001A1
• Time in Patent Office: 1,817 Days
• Field of Search: 600/407, 600/425, 600/476, 382/128, 382/131
• US Class Current: 600/425
• CPC Class Codes: G06T 7/0012 Biomedical image inspection

Abstract

For the automatic detection and characterization of nodules in a tomographic image of an anatomical zone of a patient, the image is segmented for identifying therein a region of interest, and the segmented image is processed for identifying the nodule.

During the processing step, an ellipsoid inscribed in the region of interest is modeled for deciding whether the image elements inscribed in this ellipsoid correspond to a nodule, and, for each zone of the region of interest extending beyond the ellipsoid, the image elements are identified which do not belong to the module according to mathematical morphologic criteria.

Claims

1. Method of automatic detection and characterization of nodules (N) in a tomographic image of an anatomical zone of a patient, comprising a segmentation of the image for identifying therein a region of interest (N, V) by determining a contour in three dimensions using water separation lines of luminance gradient data of the image to determine a discontinuity of luminance in the data, and the processing of the segmented image for identifying the nodule (N), wherein during the step of processing, an ellipsoid inscribed within the region of interest is modeled, it is decided that the image elements inscribed within this ellipsoid correspond to a nodule and, for each zone of the region of interest extending outside the ellipsoid, the image elements are identified which do not belong to the nodule, according to criteria of mathematical morphology.

2. Method according to claim 1, wherein a Euclidean distance field is produced with respect to the perimeter of the segmented region of interest, the image elements locally having the greatest distance value are determined, and the center of the ellipsoid is positioned on an image element having the greatest distance value.

3. Method according to claim 2, wherein the said ellipsoid is constituted by a sphere delimited by the greatest spherical volume having the said center and inscribed within the element of interest.

4. Method according to claim 2, wherein a geodesic distance transformation is performed so as to create, starting from the said center, geodesic surfaces (S) respectively extending at increasing distances with respect thereto and each delimited by a set of image elements inscribed within the element of interest and equidistant from the center.

5. Method according to claim 4, wherein, during the step of identification of image elements not belonging to the nodule, a separate processing is performed of each geodesic surface (S) situated outside the said sphere so as to determine whether the image elements which constitute it correspond at least partially to a nodule.

6. Method according to claim 5, wherein, during the processing of each geodesic surface, a discriminant analysis is performed from at least one morphologic criterion chosen from among the distance separating it from the center of the sphere, the connectivity of the image elements of the geodesic surface with the preceding ones, and the maximum Euclidean distance on this image element.

7. Method according to claim 1, wherein, preceding the segmentation step, a preliminary filtering is performed during which a volume (V) is produced in which is inscribed a portion of the image likely to include a nodule, the said step of segmentation being implemented on the image elements of the said volume.

8. Method according to claim 7, wherein the step of filtering is implemented by a thresholding of numerical values associated with the image elements.

9. Method according to claim 7, wherein the said step of thresholding is implemented by morphologic segmentation, in particular by extraction of water separation lines (WSL).

10. Method according to claim 7, wherein, starting from a selected point (X), a histogram is produced of Boolean identification variables of the said portion of the image likely to include a nodule (N), the Boolean variables being the output of filtering of the numerical values respectively associated with image elements extending from the selected point along a set of directions, the said volume (V) being produced from the said Boolean variables.

11. Method according to claim 1, wherein, preceding the segmentation step, furthermore a step is performed of detection of the wall (P) of an organ of the anatomical zone examined, a mask is produced from the detected wall, and the mask is applied to the tomographic image.

12. Method according to claim 11, wherein, during the step of detection of the wall of the organ, a volume is modeled with a contour applied against the internal wall of the organ within the organ, and the contour of the said volume is then filtered for extracting therefrom image elements not belonging to the said wall.

13. Use of a method according to claim 1 for the detection and the characterization of pulmonary nodules.

14. A method in accordance with claim 1 wherein the region of interest is a volumetric region, said method further comprising determining a contour in three dimensions using water separation lines of luminance gradient data of the image to determine a discontinuity of luminance in the volumetric data.

15. Computer program loaded into the internal memory of a calculator, wherein said computer program includes at least one code segment configured to instruct the calculator to identify a region of interest (N,V) in a segmentation of a tomographic image of an anatomical zone of a patient by:

determining a contour in three dimensions using water separation lines of luminance gradient data of the image to determine a discontinuity of luminance in the data;

modeling an ellipsoid inscribed within the region of interest, wherein image elements inscribed within this ellipsoid correspond to a nodule; and

identifying for each zone of the region of interest extending outside the ellipsoid, the image elements which do not belong to the nodule, according to criteria of mathematical morphology.

16. Tomodensimetric medical imaging system, comprising an x-ray emitter (16), corresponding detection means (18) capable of measuring differences of absorption density of x-rays by the tissues of an anatomical zone examined, and a calculator (20) for processing signals output from the detection means (18) for producing an image of the said anatomical zone, the calculator (20) comprising image segmentation means (28) for identifying a region of interest in the said image by determining a contour in three dimensions using water separation lines of luminance gradient data of the image to determine a discontinuity of luminance in the data, and means (30, 32, 34, 36) for processing the segmented image for the identification of nodules present therein, wherein the calculator (20) furthermore comprises means (30) for modeling an ellipsoid inscribed within the region of interest for deciding that the image elements inscribed in this ellipsoid correspond to a nodule, and means (32, 34) for analysis by mathematical morphology for identifying, in each zone of the region of interest extending outside the ellipsoid, the image elements not belonging to the nodule.

17. A method of automatic analysis of nodules in a tomographic image of an object comprising:

identifying a suspect zone that includes a nodule;

extracting a sub volume of the 3D image that includes the nodule;

identifying a region of interest in the sub volume by determining a contour in three dimensions using an extraction of water separation lines of luminance gradient data of the image to determine a discontinuity of luminance in the data;

identifying the nodule type wherein the type includes at least one of a well-circumscribed nodule, a vascularized nodule, and a pleural nodule;

segmenting the nodule based on its type; and

displaying the segmented nodule.

18. A method in accordance with claim 17 wherein if the nodule is identified as a pleural nodule, segmenting the nodule based on its type comprises:

determining a wall of the object from which the nodules extend;

generating a mask from the wall; and

applying the mask to the image such that the object wall is extracted from the image.

19. A method in accordance with claim 17 wherein if the nodule is identified as a pleural nodule, segmenting the nodule based on its type comprises:

growing a volume with a closed contour in the object until the contour is applied against the wall of the object;

simplifying the shape of the closed contour using a morphologic tool such that a vessel leading from the nodule is omitted.

20. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises:

filtering relative density values of the image voxels in each direction from a point selected by the operator;

determining an associated Boolean variable for each voxel based on whether the corresponding voxel is included within the portion of the image likely to include a suspect volume;

generating a histogram that regroups the voxels and the Boolean variables; and

filtering the histogram using hysteresis to determine a suspect volume, wherein the hysteresis facilitates reducing an appearance of errors of discontinuity or of segmentation.

21. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises determining a sub volume that includes at least a portion of the nodule and at least one of background tissue, a vessel, and a portion of a wall of a lung.

22. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises generating a Euclidean distance field with respect to the perimeter of the region of interest.

23. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises:

associating a null numerical value with the voxels located on the perimeter of the region of interest; and

associating a high numerical value with the voxels situated at the greatest distance from the perimeter.

24. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises determining the voxel having the greatest distance value as the center of the ellipsoid.

25. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises modeling a sphere.

26. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises determining the largest sphere centered at the center of the ellipsoid and being inscribed within the region of interest.

27. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises determining the geodesic surfaces from the center of the ellipsoid and extending to an increasing distance with respect thereto.

28. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises assigning to each voxel of each geodesic surface the same numerical value of a geodesic distance.

29. A method in accordance with claim 17 wherein segmenting the nodule based on its type comprises determining whether each geodesic surface located outside the sphere are included in a nodule by detecting a sudden increase of morphologic criteria, including at least one of the connectivity of the voxels and the distance separating them from the center of the sphere.

30. A method in accordance with claim 17 further comprising:

calculating a volume of the nodule; and

displaying the nodule volume.

31. A method in accordance with claim 17 wherein identifying a suspect zone that includes a nodule comprises at least one of selecting a suspect zone by a user and automatically determining a suspect zone.