Method and program for computing bone age by deep neural network
Provided are a method and program for computing a bone age using a deep neural network. The method for computing a bone age using a deep neural network, including: receiving an analysis target image that is a specific medical image to compute the bone age; and analyzing the analysis target image by at least one computer using the deep neural network to compute the bone age. According to the present disclosure, since the bone age is computed by accumulating medical images of a specific race (particularly, Korean) and analyzing the same, it is possible to compute an accurate bone age that conforms to race.
1. A method for computing a bone age using a deep neural network, comprising:
receiving, by a communication module of at least one computer, an analysis target image that is a specific medical image to compute the bone age; and
analyzing the analysis target image which is received via the communication module, by a processor of the at least one computer, using the deep neural network to compute the bone age, wherein the analysis target image is of bones of entire hand, and
wherein the deep neural network includes a plurality of convolution layers which create a feature map with respect to features in the analysis target image; and a pooling layer in which a sub-sampling is performed between the plurality of convolution layers, whereby different levels of features are extracted.
2. The method of claim 1 , further comprising recognizing, by the processor, the bones of hand of the analysis target image.
3. The method of claim 1 , further comprising generating, by the processor, a plurality of analysis target augmented images by rotating or scaling the analysis target image.
4. The method of claim 1 , further comprising deleting, by the processor, a portion corresponding to noise in the analysis target image.
5. The method of claim 1 , further comprising providing, by the processor, a representative image corresponding to the computed bone age,
wherein in the representative image is an image designated by a medical staff as an image representative of a specific bone age.
6. The method of claim 5 , wherein a representative image of bones of hand is stored by a memory of the computer and the processor of the computer recognizes a shape of bones of hand in the analysis target image to search for a representative image corresponding to the shape of the bones of hand, and provides it.
7. The method of claim 5 , wherein the number of representative images presented to correspond to the analysis target image is adjusted based on the number of persons included in one or more groups obtained by classifying a total number of persons in which bones of hand are imaged, according to a specific classification criterion.
8. The method of claim 5 , further comprising:
a feedback data receiving step of receiving, by the communication module, feedback data from a user based on the representative image or the analysis target image, wherein the feedback data is bone age data determined by the user; and
correcting, by the processor, the bone age of at least one comparative image by reflecting the feedback data received by the communication module.
9. The method of claim 1 , wherein the analyzing the analysis target image further includes:
generating, by the processor, training image data in which comparative images to be used in training by the deep neural network are accumulated, and
wherein the comparative image is a medical image in which a bone age of an imaged bones of hand is determined to be normal by a medical staff.
10. The method of claim 9 , wherein the comparative image further includes a medical image of an existing patient determined to have a specific bone age.
11. The method of claim 9 , wherein the generating training image data includes:
classifying, by the processor, the training image data based on a bone age to generate one or more groups; and
computing, by the processor, a detailed feature through comparison between the generated groups.
12. The method of claim 11 , wherein the generating training image data further includes:
group-changing or excluding, by the processor, a specific comparative image which is outside a range of the detailed feature within each of the groups.
13. The method of claim 11 , further comprising indicating, by the processor, an identification mark in an area of the analysis target area corresponding to the detailed feature.
14. The method of claim 13 , wherein the indicating an identification mark includes searching, by the processor, for the comparative image most similar to each area in which the identification mark is indicated, and providing it.
15. The method of claim 9 , wherein the generating training image data is characterized in that a period range of the training image data to be used in training is limited.
16. The method of claim 9 , wherein the generating training image data includes generating, by the processor, a plurality of comparative augmented images by rotating or scaling the comparative image.
17. A bone age computing program using a deep neural network, wherein the program is stored in a medium and includes instructions implemented to cause a computer, which is hardware, to execute the method of claim 1 .
18. A computer, comprising:
a communication module configured to receive an analysis target image that is a specific medical image to compute the bone age, wherein the analysis target image is of bones of entire hand; and
a processor configured to analyze the analysis target image which is received via the communication module using the deep neural network to compute the bone age, and
wherein the deep neural network includes a plurality of convolution layers which create a feature map with respect to features in the analysis target image; and a pooling layer in which a sub-sampling is performed between the plurality of convolution layers, whereby different levels of features are extracted.