Learning device, learning method, learning program, trained model, radiographic image processing device, radiographic image processing method, and radiographic image processing program

1. A learning device comprising:

at least one processor,

wherein the processor performs machine learning of a learning model by independently inputting each of a plurality of radiographic images that do not include a surgical tool and a plurality of surgical tool images that include only the surgical tool as training data to the learning model, to construct a trained model for detecting a region of the surgical tool from an input radiographic image, and

wherein the learning model comprises a neural network and the processor further performs the machine learning of the neural network, in which when a radiographic image that does not include the surgical tool is input to the neural network as the training data, a probability of being the region of the surgical tool that is output from the neural network becomes 0 in an entire region of the input radiographic image that does not include the surgical tool, and when a surgical tool image is input to the neural network as the training data, the probability of being the region of the surgical tool that is output from the neural network becomes 1 in the region of the surgical tool in the input surgical tool image.

2. The learning device according to claim 1 ,

wherein the surgical tool image is a radiographic image acquired by performing radiography only on the surgical tool.

3. The learning device according to claim 1 ,

wherein the surgical tool image is acquired by a method other than the radiography and has an image quality corresponding to an image acquired by the radiography.

4. The learning device according to claim 3 ,

wherein the processor two-dimensionally projects a three-dimensional model of the surgical tool on the basis of a predetermined parameter to derive the surgical tool image.

5. The learning device according to claim 4 ,

wherein the processor sets the parameter according to at least one of a contrast of the surgical tool in the surgical tool image, a density of the surgical tool in the surgical tool image, or noise included in the surgical tool image.

6. The learning device according to claim 1 ,

wherein the surgical tool includes at least one of gauze, a scalpel, scissors, a drain, a suture needle, a thread, or forceps.

7. The learning device according to claim 6 ,

wherein the surgical tool comprises the gauze and at least a portion of the gauze includes a radiation absorbing thread.

8. A radiographic image processing device comprising:

at least one processor,

wherein the processor acquires a radiographic image and detects a region of a surgical tool from the radiographic image using a trained model constructed by a learning device,

wherein the learning device comprises at least one processor,

wherein the processor of the learning device performs machine learning of a learning model by independently inputting each of a plurality of radiographic images that do not include the surgical tool and a plurality of surgical tool images that include only the surgical tool as training data to the learning model, to construct the trained model for detecting the region of the surgical tool from the radiographic image, and

wherein the learning model comprises a neural network and the processor of the learning device further performs the machine learning of the neural network, in which when a radiographic image that does not include the surgical tool is input to the neural network as the training data, a probability of being the region of the surgical tool that is output from the neural network becomes 0 in an entire region of the input radiographic image that does not include the surgical tool, and when a surgical tool image is input to the neural network as the training data, the probability of being the region of the surgical tool that is output from the neural network becomes 1 in the region of the surgical tool in the input surgical tool image.

9. A learning method comprising:

performing machine learning of a learning model by independently inputting each of a plurality of radiographic images that do not include a surgical tool and a plurality of surgical tool images that include only the surgical tool as training data to the learning model, to construct a trained model for detecting a region of the surgical tool from an input radiographic image, and

wherein the learning model comprises a neural network and the learning method further includes performing the machine learning of the neural network, in which when a radiographic image that does not include the surgical tool is input to the neural network as the training data, a probability of being the region of the surgical tool that is output from the neural network becomes 0 in an entire region of the input radiographic image that does not include the surgical tool, and when a surgical tool image is input to the neural network as the training data, the probability of being the region of the surgical tool that is output from the neural network becomes 1 in the region of the surgical tool in the input surgical tool image.

10. A radiographic image processing method comprising:

acquiring a radiographic image; and

detecting a region of a surgical tool from the radiographic image using a trained model constructed by a learning device,

wherein the learning device comprises at least one processor,

wherein the processor of the learning device performs machine learning of a learning model by independently inputting each of a plurality of radiographic images that do not include the surgical tool and a plurality of surgical tool images that include only the surgical tool as training data to the learning model, to construct the trained model for detecting the region of the surgical tool from the radiographic image, and

wherein the learning model comprises a neural network and the processor of the learning device further performs the machine learning of the neural network, in which when a radiographic image that does not include the surgical tool is input to the neural network as the training data, a probability of being the region of the surgical tool that is output from the neural network becomes 0 in an entire region of the input radiographic image that does not include the surgical tool, and when a surgical tool image is input to the neural network as the training data, the probability of being the region of the surgical tool that is output from the neural network becomes 1 in the region of the surgical tool in the input surgical tool image.

11. A non-transitory computer-readable storage medium that stores a learning program that causes a computer to perform: a procedure of performing machine learning of a learning model by independently inputting each of a plurality of radiographic images that do not include a surgical tool and a plurality of surgical tool images that include only the surgical tool as training data to the learning model, to construct a trained model for detecting a region of the surgical tool from an input radiographic image, and

wherein the learning model comprises a neural network and the learning program further causes the computer to perform the machine learning of the neural network, in which when a radiographic image that does not include the surgical tool is input to the neural network as the training data, a probability of being the region of the surgical tool that is output from the neural network becomes 0 in an entire region of the input radiographic image that does not include the surgical tool, and when a surgical tool image is input to the neural network as the training data, the probability of being the region of the surgical tool that is output from the neural network becomes 1 in the region of the surgical tool in the input surgical tool image.

12. A non-transitory computer-readable storage medium that stores a radiographic image processing program that causes a computer to perform:

a procedure of acquiring a radiographic image; and

a procedure of detecting a region of a surgical tool from the radiographic image using a trained model constructed by a learning device,

wherein the learning device comprises at least one processor,

wherein the processor of the learning device performs machine learning of a learning model by independently inputting each of a plurality of radiographic images that do not include the surgical tool and a plurality of surgical tool images that include only the surgical tool as training data to the learning model, to construct the trained model for detecting the region of the surgical tool from the radiographic image, and

wherein the learning model comprises a neural network and the processor of the learning device further performs the machine learning of the neural network, in which when a radiographic image that does not include the surgical tool is input to the neural network as the training data, a probability of being the region of the surgical tool that is output from the neural network becomes 0 in an entire region of the input radiographic image that does not include the surgical tool, and when a surgical tool image is input to the neural network as the training data, the probability of being the region of the surgical tool that is output from the neural network becomes 1 in the region of the surgical tool in the input surgical tool image.