Image classification apparatus, method, and program

1. An image classification apparatus for classifying a three-dimensional image into a plurality of classes, comprising:

processing circuitry configured to:

detect at least one feature point included in the three-dimensional image;

set a reference axis in the three-dimensional image based on at least the one feature point;

generate a two-dimensional image by projecting a target object, which is included in the three-dimensional image, in a specific projection direction using the reference axis as a reference; and

classify each pixel of the target object into the plurality of classes based on the two-dimensional image.

2. The image classification apparatus according to claim 1 , wherein the processing circuitry is further configured to extract the target object from the three-dimensional image.

3. The image classification apparatus according to claim 1 ,

wherein the processing circuitry is further configured to execute a neural network in which learning is performed by using teacher data for the two-dimensional image in order to classify the target object into the plurality of classes.

4. The image classification apparatus according to claim 3 ,

wherein, in the neural network, a plurality of processing layers are hierarchically connected to each other.

5. The image classification apparatus according to claim 3 ,

wherein, in the neural network, learning is performed by using the two-dimensional image of the entire target object as teacher data, and

the processing circuitry is further configured to generate the two-dimensional image of the entire target object.

6. The image classification apparatus according to claim 3 ,

wherein, in the neural network, learning is performed for each local region of the target object by using the two-dimensional image as teacher data, and

the processing circuitry is further configured to generate the two-dimensional image for each local region of the target object.

7. The image classification apparatus according to claim 1 ,

wherein the two-dimensional image is a spin image.

8. The image classification apparatus according to claim 1 ,

wherein, in a case where the target object is blood vessels of a heart, the processing circuitry is further configured to detect at least one of an aortic valve, a mitral valve, or an apical portion as the feature point, and

the processing circuitry is further configured to classify the blood vessels of the heart into a coronary artery and a coronary vein.

9. The image classification apparatus according to claim 1 ,

wherein, in a case where the target object is branch blood vessels from an aorta of a human body, the processing circuitry is further configured to detect at least one of an aortic arch, a branch position from an abdominal aorta to an iliac artery, or a spine as the feature point, and

the processing circuitry is further configured to classify the branch blood vessels from the aorta into at least one of a carotid artery, a subclavian artery, a celiac artery, a renal artery, a mesenteric artery, or an iliac artery.

10. The image classification apparatus according to claim 1 ,

wherein the processing circuitry is further configured to generate the two-dimensional image by generating a tree structure, in which the target object is continuous, and project the generated tree structure in the specific projection direction.

11. A learning apparatus for performing learning of a neural network by using teacher data for classifying a three-dimensional image into a plurality of classes, comprising:

processing circuitry configured to:

detect at least one feature point included in the three-dimensional image serving as the teacher data;

set a reference axis in the three-dimensional image serving as the teacher data based on at least the one feature point;

generate a two-dimensional image serving as the teacher data by projecting a target object, which is included in the three-dimensional image serving as the teacher data, in a specific projection direction using the reference axis as a reference; and

learn the neural network so as to output a result of the classification with the two-dimensional image serving as the teacher data being an input.

12. The learning apparatus according to claim 11 ,

wherein, in the neural network, a plurality of processing layers are hierarchically connected to each other.

13. An image classification method for classifying a three-dimensional image into a plurality of classes, comprising:

detecting at least one feature point included in the three-dimensional image;

setting a reference axis in the three-dimensional image based on at least the one feature point;

generating a two-dimensional image by projecting a target object, which is included in the three-dimensional image, in a specific projection direction using the reference axis as a reference; and

classifying each pixel of the target object into the plurality of classes based on the two-dimensional image.

14. A learning method for performing learning of a neural network by using teacher data for classifying a three-dimensional image into a plurality of classes, comprising:

detecting at least one feature point included in the three-dimensional image serving as the teacher data;

setting a reference axis in the three-dimensional image serving as the teacher data based on at least the one feature point;

generating a two-dimensional image serving as the teacher data by projecting a target object, which is included in the three-dimensional image serving as the teacher data, in a specific projection direction using the reference axis as a reference; and

learning the neural network so as to output a result of the classification with the two-dimensional image serving as the teacher data being an input.

15. A non-transitory computer-readable recording medium having stored therein an image classification program causing a computer to execute an image classification method for classifying a three-dimensional image into a plurality of classes, the program causing the computer to execute:

a step of detecting at least one feature point included in the three-dimensional image;

a step of setting a reference axis in the three-dimensional image based on at least the one feature point;

a step of generating a two-dimensional image by projecting a target object, which is included in the three-dimensional image, in a specific projection direction using the reference axis as a reference; and

a step of classifying each pixel of the target object into the plurality of classes based on the two-dimensional image.

16. A non-transitory computer-readable recording medium having stored therein a learning program causing a computer to execute a learning method for performing learning of a neural network by using teacher data for classifying a three-dimensional image into a plurality of classes, the program causing the computer to execute:

a step of detecting at least one feature point included in the three-dimensional image serving as the teacher data;

a step of setting a reference axis in the three-dimensional image serving as the teacher data based on at least the one feature point;

a step of generating a two-dimensional image serving as the teacher data by projecting a target object, which is included in the three-dimensional image serving as the teacher data, in a specific projection direction using the reference axis as a reference; and

a step of learning the neural network so as to output a result of the classification with the two-dimensional image serving as the teacher data being an input.