Automatic detection of pulmonary nodules on volumetric computed tomography images using a local density maximum algorithm

• US 6,728,334 B1
• Filed: 10/23/2002
• Issued: 04/27/2004
• Est. Priority Date: 10/24/2001
• Status: Active Grant

• Current Assignee: Cornell Research Foundation Incorporated (Cornell University)
• Original Assignee: Cornell Research Foundation Incorporated (Cornell University)
• Inventors: Zhao, Binsheng
• Primary Examiner(s): BRUCE, DAVID VERNON
• Application Number:
  US10/279,592
• Time in Patent Office: 552 Days
• Field of Search: 378/4, 378/8, 378/15, 378/62, 378/901, 382/131
• US Class Current: 378/62
• CPC Class Codes: G01N 2223/419 computed tomograph
• CPC Class Codes: G01N 2223/612 biological material
• CPC Class Codes: G01N 23/046 using tomography, e.g. computed tomography [CT]
• CPC Class Codes: Y10S 378/901 Computer tomography program or processor

Abstract

A three dimensional mask of the lungs can be automatically created by thresholding, labeling connected components, selecting the dominant object, and alternately employing dilation and erosion operations. With this mask the lungs can be separated from the other anatomic structures on volumetric CT images. Local density maxima in the lungs are then determined by sequentially decreasing thresholds. As the threshold declines, more and more objects (a 3D object is a group of connected voxels with density values larger than the threshold) become apparent. Geometrically overlapped objects at the subsequent threshold levels are either merged into one object or identified as local density maximum (maxima) and plateau. This process terminates if the threshold reaches a predefined density value. Other information about small lung nodules such as compact shape and size are combined into the algorithm to further remove those detected local density maxima that are not likely to be nodules.

Claims

1. A method for analyzing volumetric chest computed tomography images for lung nodules, said method comprising the steps of:

obtaining the volumetric chest computed tomography images for lung nodules from an image acquisition device;

separating the lungs from other structures on the volumetric chest computed tomography images to form lung images;

detecting nodule candidates in the lung images with a local density maximum algorithm; and

reducing false-positives among the detected nodule candidates based upon by an application of parameters concerning lung nodules.

2. A method as defined in claim 1, wherein:

the parameters include for each nodule candidate are;

a number of voxels associated with the nodule candidate;

ratio R1 defined as a volume of the nodule candidate over a volume of a modified bounding box of the nodule candidate;

ratio R2 defined as a maximal projection length of the nodule candidate along an axis of z over a maximal projection length of the nodule candidate, the maximal projection length being the larger of a projection length along the axis of x and a projection length along axis of y; and

ratio R3 defined as the maximal projection length of the nodule candidate over the minimal projection length, the minimal projection length being the smaller is smaller the projection length along the axis of x and the projection length along axis of y; and

deleting the nodule candidate from further consideration when one of the following criteria is established;

R1 is less than about 0.3;

R2 is greater than about 5.0;

R2 is less than about 0.2;

R3 is greater than about 1.5;

((the projection length along the axis of x is less than about 2.0 mm) AND (the projection length along axis of y is less than about 2.0 mm) ); and

the number of voxels associated with the nodule candidate is greater than about 800.

3. An article of manufacture for analyzing volumetric chest computed tomography images for lung nodules, said article comprising:

a machine readable medium containing one or more programs which when executed implement the steps of;

obtaining the volumetric chest computed tomography images for lung nodules from an image acquisition device;

separating the lungs from other structures on the volumetric chest computed tomography images to form lung images;

detecting nodule candidates in the lung images with a local density maximum algorithm; and

reducing false-positives among the detected nodule candidates based upon by an application of parameters concerning lung nodules.

4. An article of manufacture as defined in claim 3, wherein:

the parameters include for each nodule candidate are;

a number of voxels associated with the nodule candidate;

ratio R1 defined as a volume of the nodule candidate over a volume of a modified bounding box of the nodule candidate;

ratio R2 defined as a maximal projection length of the nodule candidate along an axis of z over a maximal projection length of the nodule candidate, the maximal projection length being the larger of a projection length along the axis of x and a projection length along axis of y; and

ratio R3 defined as the maximal projection length of the nodule candidate over the minimal projection length, the minimal projection length being the smaller is smaller the projection length along the axis of x and the projection length along axis of y; and

deleting the nodule candidate from further consideration when one of the following criteria is established;

R1 is less than about 0.3;

R2 is greater than about 5.0;

R2 is less than about 0.2;

R3 is greater than about 1.5;

((the projection length along the axis of x is less than about 2.0 mm) AND (the projection length along axis of y is less than about 2.0 mm)); and

the number of voxels associated with the nodule candidate is greater than about 800.

5. An apparatus for analyzing volumetric chest computed tomography images for lung nodules, said apparatus comprising:

an image analyzing unit configured to;

obtain the volumetric chest computed tomography images for lung nodules from an image acquisition device;

separate the lungs from other structures on the volumetric chest computed tomography images to form lung images;

detect nodule candidates in the lung images with a local density maximum algorithm; and

reduce false-positives among the detected nodule candidates based upon by an application of parameters concerning lung nodules.

6. A method as defined in claim 5, wherein:

the parameters include for each nodule candidate are;

a number of voxels associated with the nodule candidate;

ratio R1 defined as a volume of the nodule candidate over a volume of a modified bounding box of the nodule candidate;

ratio R2 defined as a maximal projection length of the nodule candidate along an axis of z over a maximal projection length of the nodule candidate, the maximal projection length being the larger of a projection length along the axis of x and a projection length along axis of y; and

ratio R3 defined as the maximal projection length of the nodule candidate over the minimal projection length, the minimal projection length being the smaller is smaller the projection length along the axis of x and the projection length along axis of y; and

the image analyzing unit is configured to delete the nodule candidate from further consideration when one of the following criteria is established;

R1 is less than about 0.3;

R2 is greater than about 5.0;

R2 is less than about 0.2;

R3 is greater than about 1.5;

((the projection length along the axis of x is less than about 2.0 mm) AND (the projection length along axis of y is less than about 2.0 mm)); and

the number of voxels associated with the nodule candidate is greater than about 800.

7. A method for detecting lung nodule candidates in extracted lung images, said method comprising the steps of:

(a) selecting processing parameters which define;

a threshold step value;

a stop value;

an initial threshold level equal to the maximum voxel density value of the extracted lung images minus the threshold step value;

a minimal size of local maximum; and

a minimal density peak height of local maximum;

(b) segmenting the extracted lung images at the threshold level;

(c) identifying three-dimensional (3D) objects at the threshold level;

(d) determining whether a previous object has been detected that overlaps with a current object and proceeding to step (f) when no previous object has been detected;

(e) determining local maxima at the threshold level based on an analysis of the previous object and current objects;

(f) saving the objects on a previous object array;

(g) reducing the threshold level by the threshold step value;

(h) repeating steps (b) through (g) provided that the threshold value is greater than the stop value.

8. A method as defined in claim 7, wherein step (c) comprises the substeps of:

(a) finding a plurality of segments of connected pixels having a value equal to 1 in a row of a plane;

(b) assigning a separate object number to each of said plurality of segments to define an object list;

(c) unifying the object number of each of said plurality of segments when it overlaps another of said plurality of segments in at least one of another row and plane;

(d) updating the object list;

(e) repeating substeps (a) through (d) for each row in the plane; and

(f) repeating substep (e) for each plane.

9. A method as defined in claim 7, wherein step (e) comprises the substeps of:

(a) determining for each current object whether the current object has more than one previous object on the current object;

(b) proceeding to substep (j) when the current object does not have more than one previous object on the current object;

(c) determining for each current object whether local maxima have already been found on the previous object associated with the current object;

(d) proceeding to substep (h) when local maxima have already been found on the previous object;

(e) determining whether the previous object is a local maximum;

(f) proceeding to substep (i) when the previous objects are not a local maximum;

(g) marking the previous object as local maximum;

(h) marking the current object as a plateau;

(i) returning to step (f);

(j) determining whether there is only one previous object on the current object;

(k) proceeding to substep (i) when there is no previous object on the current object;

(l) determining whether the ratio of the object-to-box ratio of the previous object to that of the current object is larger than a specified value;

(m) proceeding to substep (i) when the object-to-box ratio is less than the specified value; and

(n) proceeding to substep (c).

10. An article of manufacture for detecting lung nodule candidates in extracted lung images, said article comprising:

a machine readable medium containing one or more programs which when executed implement the steps of;

(a) selecting processing parameters which define;

a threshold step value;

a stop value;

an initial threshold level equal to the maximum voxel density value of the extracted lung images minus the threshold step value;

a minimal size of local maximum; and

a minimal density peak height of local maximum;

(b) segmenting the extracted lung images at the threshold level;

(c) identifying three-dimensional (3D) objects at the threshold level;

(d) determining whether a previous object has been detected that overlaps with a current object and proceeding to step (f) when no previous object has been detected;

(e) determining local maxima at the threshold level based on an analysis of the previous object and current objects;

(f) saving the objects on a previous object array;

(g) reducing the threshold level by the threshold step value;

(h) repeating steps (b) through (g) provided that the threshold value is greater than the stop value.

11. An article of manufacture as defined in claim 10, wherein step (c) comprises the substeps of:

(a) finding a plurality of segments of connected pixels having a value equal to 1 in a row of a plane;

(b) assigning a separate object number to each of said plurality of segments to define an object list;

(c) unifying the object number of each of said plurality of segments when it overlaps another of said plurality of segments in at least one of another row and plane;

(d) updating the object list;

(e) repeating substeps (a) through (d) for each row in the plane; and

(f) repeating substep (e) for each plane.

12. An article of manufacture as defined in claim 10, wherein step (e) comprises the substeps of:

(a) determining for each current object whether the current object has more than one previous object on the current object;

(b) proceeding to substep (j) when the current object does not have more than one previous object on the current object;

(c) determining for each current object whether local maxima have already been found on the previous object associated with the current object;

(d) proceeding to substep (h) when local maxima have already been found on the previous object;

(e) determining whether the previous object is a local maximum;

(f) proceeding to substep (i) when the previous objects are not a local maximum;

(g) marking the previous object as local maximum;

(h) marking the current object as a plateau;

(i) returning to step (f);

(j) determining whether there is only one previous object on the current object;

(k) proceeding to substep (i) when there is no previous object on the current object;

(l) determining whether the ratio of the object-to-box ratio of the previous object to that of the current object is larger than a specified value;

(m) proceeding to substep (i) when the object-to-box ratio is less than the specified value; and

(n) proceeding to substep (c).

13. An apparatus for detecting lung nodule candidates in extracted lung images, said apparatus comprising:

a nodule detecting unit configured to;

(a) select processing parameters which define;

a threshold step value;

a stop value;

an initial threshold level equal to the maximum voxel density value of the extracted lung images minus the threshold step value;

a minimal size of local maximum; and

a minimal density peak height of local maximum;

(b) segment the extracted lung images at the threshold level;

(c) identify three-dimensional (3D) objects at the threshold level;

(d) determine whether a previous object has been detected that overlaps with a current object and proceeding to step (f) when no previous object has been detected;

(e) determine local maxima at the threshold level based on an analysis of the previous object and current objects;

(f) save the objects on a previous object array;

(g) reducing the threshold level by the threshold step value;

(h) repeat steps (b) through (g) provided that the threshold value is greater than the stop value.

14. An apparatus for detecting lung nodule candidates as defined in claim 13, wherein the nodule detecting unit identifies three-dimensional (3D) objects at the threshold level by being configured to:

(a) find a plurality of segments of connected pixels having a value equal to 1 in a row of a plane;

(b) assign a separate object number to each of said plurality of segments to define an object list;

(c) unify the object number of each of said plurality of segments when it overlaps another of said plurality of segments in at least one of another row and plane;

(d) update the object list;

(e) repeat (a) through (d) for each row in the plane; and

(f) repeat (e) for each plane.

15. An apparatus for detecting lung nodule candidates as defined in claim 13, wherein the nodule detecting unit analyses of the previous object and current objects to determine the local maxima at the threshold level by being configured to:

(a) determine for each current object whether the current object has more than one previous object on the current object;

(b) proceed to substep (j) when the current object does not have more than one previous object on the current object;

(c) determine for each current object whether local maxima have already been found on the previous object associated with current object;

(d) proceed to substep (h) when local maxima have already been found on the previous object;

(e) determine whether the previous object is a local maximum;

(f) proceed to substep (i) when the previous objects are not a local maximum;

(g) mark the previous object as local maximum;

(h) mark the current object as a plateau;

(i) return to step (f);

(j) determine whether there is only one previous object on the current object;

(k) proceed to substep (i) when there is no previous object on the current object;

(l) determine whether the ratio of the object-to-box ratio of the previous object to that of the current object is larger than a specified value;

(m) proceed to substep (i) when the object-to-box ratio is less than the specified value; and

(n) proceed to substep (c).