Image processing device, learning device, radiography system, image processing method, learning method, image processing program, and learning program
An image processing device acquires a plurality of projection images, inputs the acquired plurality of projection images to a tomographic image estimation model, which is a trained model generated by performing machine learning on a machine learning model using learning data composed of a set of correct answer data that is three-dimensional data indicating a three-dimensional structure and of a plurality of virtual projection images, onto which the three-dimensional structure has been projected by performing pseudo-projection on the three-dimensional structure with radiation at a plurality of virtual irradiation positions using the three-dimensional data, and which receives the plurality of projection images as an input and outputs an estimated tomographic image group, and acquires the estimated tomographic image group output from the tomographic image estimation model.
1. An image processing device that processes a plurality of projection images obtained by sequentially irradiating an object with radiation at each of a plurality of irradiation positions being included in a first irradiation angle range and having different irradiation angles, the image processing device comprising:
at least one processor,
wherein the processor acquires the plurality of projection images, resulting in an acquired plurality of projection images, inputs the acquired plurality of projection images to a tomographic image estimation model, which is a trained model generated by performing machine learning on a machine learning model using learning data composed of a set of correct answer data that is three-dimensional data indicating a three-dimensional structure and of a plurality of virtual projection images, onto which the three-dimensional structure has been projected by performing pseudo-projection on the three-dimensional structure with the radiation at a plurality of virtual irradiation positions using the three-dimensional data, and which receives the plurality of projection images as an input and outputs an estimated tomographic image group, and acquires the estimated tomographic image group output from the tomographic image estimation model,
wherein the correct answer data is a plurality of correct answer tomographic images, each corresponding to a different tomographic plane, and each of the correct answer tomographic images is obtained by reconstructing sinograms derived from radiography performed over a second irradiation angle range that is wider than the first irradiation angle range.
2. The image processing device according to claim 1 ,
wherein the object is a breast.
3. A radiography system comprising:
a radiation source that generates radiation;
a radiography apparatus that performs tomosynthesis imaging which irradiates an object with the radiation at each of a plurality of irradiation positions having different irradiation angles to capture projection images of the object at each of the irradiation positions; and
the image processing device according to claim 1 .
4. A learning device comprising:
at least one processor,
wherein the processor performs pseudo-projection on a three-dimensional structure with radiation at a plurality of virtual irradiation positions, using three-dimensional data indicating the three-dimensional structure, to generate a plurality of virtual projection images onto which the three-dimensional structure has been projected, and performs machine learning on a machine learning model, using learning data composed of a set of correct answer data which is the three-dimensional data and of the plurality of virtual projection images, to generate a tomographic image estimation model that receives a plurality of projection images as an input and outputs an estimated tomographic image group,
wherein the plurality of projection images is obtained by sequentially irradiating the object with radiation from each of a plurality of irradiation positions being included in a first irradiation angle range and having different irradiation angles, and
wherein the correct answer data is a plurality of correct answer tomographic images, each corresponding to a different tomographic plane, obtained by reconstructing sinograms derived from radiography performed over a second irradiation angle range that is wider than the first irradiation angle range.
5. The learning device according to claim 4 ,
wherein the processor performs the machine learning on the machine learning model using the learning data in which partial projection regions, which correspond to partial tomographic regions of the correct answer tomographic images, in the plurality of virtual projection images are associated with each of the partial tomographic regions.
6. The learning device according to claim 4 ,
wherein the processor simulates attenuation of the radiation according to an absorption coefficient to generate the plurality of virtual projection images.
7. The learning device according to claim 4 ,
wherein the processor gives a noise component corresponding to an arrival dose to a dose of the radiation assumed to reach a radiation detector, which generates the projection images, to generate the plurality of virtual projection images.
8. The learning device according to claim 4 ,
wherein the plurality of virtual irradiation positions are positions that simulate irradiation positions of the radiation in tomosynthesis imaging.
9. The learning device according to claim 4 ,
wherein the three-dimensional structure is a structure indicating a breast, and
the plurality of projection images are projection images obtained by imaging the breast as an object.
10. An image processing method that is executed by a computer and that processes a plurality of projection images obtained by sequentially irradiating an object with radiation at each of a plurality of irradiation positions being included in a first irradiation angle range and having different irradiation angles, the image processing method comprising:
acquiring the plurality of projection images, resulting in an acquired plurality of projection images;
inputting the acquired plurality of projection images to a tomographic image estimation model, which is a trained model generated by performing machine learning on a machine learning model using learning data composed of a set of correct answer data that is three-dimensional data indicating a three-dimensional structure and of a plurality of virtual projection images, onto which the three-dimensional structure has been projected by performing pseudo-projection on the three-dimensional structure with the radiation at a plurality of virtual irradiation positions using the three-dimensional data, and which receives the plurality of projection images as an input and outputs an estimated tomographic image group; and
acquiring the estimated tomographic image group output from the tomographic image estimation model,
wherein the correct answer data is a plurality of correct answer tomographic images, each corresponding to a different tomographic plane, and each of the correct answer tomographic images is obtained by reconstructing sinograms derived from radiography performed over a second irradiation angle range that is wider than the first irradiation angle range.
11. A learning method that is executed by a computer, the learning method comprising:
performing pseudo-projection on a three-dimensional structure with radiation at a plurality of virtual irradiation positions, using three-dimensional data indicating the three-dimensional structure, to generate a plurality of virtual projection images onto which the three-dimensional structure has been projected; and
performing machine learning on a machine learning model, using learning data composed of a set of correct answer data which is the three-dimensional data and of the plurality of virtual projection images, to generate a tomographic image estimation model that receives a plurality of projection images as an input and outputs an estimated tomographic image group,
wherein the plurality of projection images is obtained by sequentially irradiating the object with radiation from each of a plurality of irradiation positions being included in a first irradiation angle range and having different irradiation angles, and
wherein the correct answer data is a plurality of correct answer tomographic images, each corresponding to a different tomographic plane, obtained by reconstructing sinograms derived from radiography performed over a second irradiation angle range that is wider than the first irradiation angle range.
12. A non-transitory computer-readable storage medium storing an image processing program that processes a plurality of projection images obtained by sequentially irradiating an object with radiation at each of a plurality of irradiation positions being included in a first irradiation angle range and having different irradiation angles, the image processing program causing a computer to perform a process comprising:
acquiring the plurality of projection images, resulting in an acquired plurality of projection images;
inputting the acquired plurality of projection images to a tomographic image estimation model, which is a trained model generated by performing machine learning on a machine learning model using learning data composed of a set of correct answer data that is three-dimensional data indicating a three-dimensional structure and of a plurality of virtual projection images, onto which the three-dimensional structure has been projected by performing pseudo-projection on the three-dimensional structure with the radiation at a plurality of virtual irradiation positions using the three-dimensional data, and which receives the plurality of projection images as an input and outputs an estimated tomographic image group; and
acquiring the estimated tomographic image group output from the tomographic image estimation model,
wherein the correct answer data is a plurality of correct answer tomographic images, each corresponding to a different tomographic plane, and each of the correct answer tomographic images is obtained by reconstructing sinograms derived from radiography performed over a second irradiation angle range that is wider than the first irradiation angle range.
13. A non-transitory computer-readable storage medium storing a learning program that causes a computer to perform a process comprising:
performing pseudo-projection on a three-dimensional structure with radiation at a plurality of virtual irradiation positions, using three-dimensional data indicating the three-dimensional structure, to generate a plurality of virtual projection images onto which the three-dimensional structure has been projected; and
performing machine learning on a machine learning model, using learning data composed of a set of correct answer data which is the three-dimensional data and of the plurality of virtual projection images, to generate a tomographic image estimation model that receives a plurality of projection images as an input and outputs an estimated tomographic image group,
wherein the plurality of projection images is obtained by sequentially irradiating the object with radiation from each of a plurality of irradiation positions being included in a first irradiation angle range and having different irradiation angles, and
wherein the correct answer data is a plurality of correct answer tomographic images, each corresponding to a different tomographic plane, obtained by reconstructing sinograms derived from radiography performed over a second irradiation angle range that is wider than the first irradiation angle range.