Developing a training set for a deep learning system configured to determine a centerline in a three dimensional image

1. A system for generating a training example to train artificial intelligence software to automatically determine a centerline of an elongated structure of three dimensional images, the system comprising

an electronic processor configured to

receive a plurality of reference points for a subset of a plurality of slices of a first three dimensional image, wherein at least one of the plurality of reference points is received every predetermined number of slices of the plurality of slices and each of the plurality of reference points marks a centerline of the elongated structure within one of the subset of the plurality of slices;

receive a selection of a starting reference point within the plurality of reference points;

determine an order of the plurality of reference points;

fit a spline curve to the plurality of reference points based on the order of the reference points to define a marked centerline of the three dimensional image to create the training example;

add the training example to a training set; and

use the training set to train the artificial intelligence software to automatically determine a centerline of an elongated structure in a second three dimensional medical image.

2. The system according to claim 1 , wherein the electronic processor is configured to determine the order of the reference points by

(a) determining a starting reference point within the plurality of reference points;

(b) setting the starting reference point as the current reference point in an order of the plurality of reference points;

(c) determining a three dimensional distance between the current reference point and each of the plurality of reference points not yet included in the order of the plurality of reference points;

(d) weighting each three dimensional distance based on a known shape of the centerline;

(e) selecting a reference point not yet included in the order of the plurality of reference points having a shortest weighted three dimensional distance from the current reference point as the next reference point in the order of plurality of reference points;

(f) setting the next reference point as the current reference point and repeating (c) through (e) until each of the plurality of reference points are included in the order of the plurality of reference points.

3. The system according to claim 2 , wherein weighting each three dimensional distance based on a known shape of the centerline includes weighting each three dimensional distance based on a position of the current reference point along the known shape of the centerline and an expected position of the next reference point along the known shape of the centerline.

4. The system according to claim 1 , wherein the elongated structure is one selected from the group consisting of an artery, a colon, or an esophagus.

5. The system according to claim 1 , wherein the electronic processor is further configured to:

determine a two dimensional cross section of the second three dimensional medical image based on the centerline determined of the elongated structure in the second three dimensional medical image via the artificial intelligence software;

convert the two dimensional cross section to polar coordinates;

fit a line to the elongated structure in the two dimensional cross section converted to polar coordinates; and

reconvert the two dimensional cross section to Cartesian coordinates.

6. The system according to claim 5 , wherein the electronic processor is further configured to:

determine a diameter of the elongated structure in the two dimensional cross section; and

when the diameter exceeds a predetermined threshold and the elongated structure is an artery, determine the elongated structure is aneurysmal.

7. The system according to claim 5 , wherein the line fitted to the elongated structure in the two dimensional cross section indicates a circumference of the elongated structure.

8. The system according to claim 1 , wherein the training set includes a plurality of different training examples.

9. A method for generating a training example to train artificial intelligence software to automatically determine a centerline of an elongated structure of three dimensional images, the method comprising

receiving a plurality of a reference points for a subset of a plurality of slices of a first three dimensional image, wherein at least one of the plurality of reference points is received every predetermined number of slices of the plurality of slices and each of the plurality of reference points marks a centerline of the elongated structure within one of the subset of the plurality of slices;

receiving a selection of a starting reference point within a plurality of reference points;

determining an order of the plurality of reference points;

fitting a spline curve to the plurality of reference points based on the order reference points based on the order of the reference points to define a marked centerline of the three dimensional image to create a training example;

adding the training example to a training set; and

using the training set to train the artificial intelligence software to automatically determine a centerline of an elongated structure in a second three dimensional image.

10. The method according to claim 9 , wherein determining the order of the reference points includes

(a) determining a starting reference point within the plurality of reference points;

(b) setting the starting reference point as the current reference point in an order of the plurality of reference points;

(c) determining a three dimensional distance between the current reference point and each of the plurality of reference points not yet included in the order of the plurality of reference points;

(d) weighting each three dimensional distance based on a known shape of the centerline;

(e) selecting a reference point not yet included in the order of the plurality of reference points having a shortest weighted three dimensional distance from the current reference point as the next reference point in the order of plurality of reference points;

(f) setting the next reference point as the current reference point and repeating (c) through (e) until each of the plurality of reference points are included in the order of the plurality of reference points.

11. The method according to claim 10 , wherein weighting each three dimensional distance includes weighting each three dimensional distance based on a position of the current reference point along the known shape of the centerline and an expected position of the next reference point along the known shape of the centerline.

12. The method according to claim 9 , wherein the elongated structure is one selected from the group consisting of an artery, a colon, or an esophagus.

13. The method according to claim 9 , wherein the method further includes

determining a two dimensional cross section of the second three dimensional medical image based on the centerline determined of the elongated structure in the second three dimensional medical image via the artificial intelligence software;

converting the two dimensional cross section to polar coordinates;

fitting a line to the elongated structure in the two dimensional cross section converted to polar coordinates; and

reconverting the two dimensional cross section to Cartesian coordinates.

14. The method according to claim 13 , wherein the method further includes

determining a diameter of the elongated structure in the two dimensional cross section; and

if the diameter exceeds a predetermined threshold and the elongated structure is an artery, determining the elongated structure is aneurysmal.

15. The method according to claim 13 , wherein the line fitted to the elongated structure in the two dimensional cross section indicates a circumference of the elongated structure.