Learning device, method, and program, medical image processing apparatus, method, and program, and discriminator
An information acquisition unit acquires a learning image including a disease region and a first teacher label that specifies the disease region included in the learning image. A teacher label generation unit generates at least one second teacher label of which a criterion for specifying the disease region is different from the first teacher label. A learning unit trains a discriminator that detects a disease region included in a target image on the basis of the learning image, the first teacher label, and the at least one second teacher label.
1. A learning device comprising at least one processor, wherein the processor is configured to:
acquire a learning image including a disease region and a first teacher label that specifies the disease region included in the learning image;
generate at least one second teacher label of which a criterion for specifying the disease region is different from the first teacher label on the basis of a standard deviation of a signal value of a region in the first teacher label; and
train a discriminator that detects a disease region included in a target image on the basis of the learning image, the first teacher label, and the at least one second teacher label, wherein the first teacher label and the second teacher label are respectively inputted to the discriminator for training.
2. The learning device according to claim 1 ,
wherein the generation of the at least one second teacher label is further on the basis of a centroid position of the first teacher label in the learning image.
3. The learning device according to claim 2 ,
wherein the processor is configured to derive a representative value of the signal values in the region of the first teacher label in the learning image, and generates a region corresponding to the first teacher label and a region in which signal values of a region adjacent to the region in the first teacher label in the learning image are within a predetermined range with respect to the representative value, as the second teacher label.
4. The learning device according to claim 1 ,
wherein the processor is configured to train the discriminator by inputting the learning image to the discriminator to detect a learning disease region, deriving a first loss between the learning disease region and the first teacher label and a second loss between the learning disease region and the second teacher label, deriving a total loss from the first loss and the second loss, and using the total loss in training the discriminator.
5. The learning device according to claim 1 ,
wherein the processor is configured to acquire an image of a brain, and the disease region is a region of a brain disease.
6. A medical image processing apparatus comprising at least one processor to which the discriminator trained by the learning device according to claim 1 is applied,
wherein the processor is configured to detect a disease region included in a target medical image in a case where the target medical image is input.
7. The medical image processing apparatus according to claim 6 , wherein the processor is further configured to:
perform labeling of the disease region detected from the target medical image; and
cause a display to display the labeled target medical image.
8. A discriminator which is trained by the learning device according to claim 1 , and which detects a disease region included in a target medical image in a case where the target medical image is input.
9. A learning method comprising:
acquiring a learning image including a disease region and a first teacher label that specifies the disease region included in the learning image;
generating at least one second teacher label of which a criterion for specifying the disease region is different from the first teacher label on the basis of a standard deviation of a signal value of a region in the first teacher label; and
training a discriminator that detects a disease region included in a target image on the basis of the learning image, the first teacher label, and the at least one second teacher label, wherein the first teacher label and the second teacher label are respectively inputted to the discriminator for training.
10. A medical image processing method of detecting a disease region included in a target medical image in a case where the target medical image is input, using the discriminator trained by the learning method according to claim 9 .
11. A non-transitory computer-readable storage medium that stores a learning program causing a computer to execute:
a procedure of acquiring a learning image including a disease region and a first teacher label that specifies the disease region included in the learning image;
a procedure of generating at least one second teacher label of which a criterion for specifying the disease region is different from the first teacher label on the basis of a standard deviation of a signal value of a region in the first teacher label; and
a procedure of training a discriminator that detects a disease region included in a target image on the basis of the learning image, the first teacher label, and the at least one second teacher label, wherein the first teacher label and the second teacher label are respectively inputted to the discriminator for training.
12. A non-transitory computer-readable medium, storing a medical image processing program causing a computer to execute a procedure of detecting a disease region included in a target medical image in a case where the target medical image is input, using the discriminator trained by the learning method according to claim 9 .