Method and apparatus for bone age assessment

1. A method for bone age assessment using a neural network performed by a computing device, the method comprising:

receiving an analysis image which is a target of bone age assessment; and

assessing bone age of the target by inputting the analysis image into a bone age analysis model comprising one or more neural networks,

wherein the bone age analysis model, which is trained by supervised learning based on an attention guide label, comprises at least one attention module for analyzing a main region of the analysis image,

wherein the attention guide label comprises importance of each pixel obtained based on a distance between the each pixel included in a training image for training the bone age analysis model and the main region,

wherein the supervised learning is performed based on a comparison result of a spatial attention map generated regarding the training image using the bone age analysis model and the attention guide label corresponding to the training image,

wherein the supervised learning is performed based on:

calculating a loss function by using a label of each pixel included in the attention guide label as a ground truth value and using predicted importance of each pixel included in the spatial attention map as a prediction value, and

updating weights of the attention module so the loss function is minimized.

2. The method of claim 1 , wherein the attention guide label is generated based on a detection result obtained as a result of inputting the training image to a main region detection model comprising at least one neural network, and

wherein the detection result comprises location information on at least one main region comprised in the training image.

3. The method of claim 2 , wherein when the main region is in a form of a bounding box, the detection result comprises coordinates of a center point of at least one main region comprised in the training image, a width of the at least one main region comprised in the training image, and a height of the at least one main region comprised in the training image.

4. The method of claim 1 , wherein the attention guide label comprises an importance of at least one pixel comprised in the training image, and

wherein the pixel is obtained as a result of substituting a distance between coordinates of the pixel and coordinates of a center point of one or more main regions comprised in the training image to an equation based on a gaussian distribution.

5. The method of claim 1 , wherein the bone age analysis model is trained by supervised learning based on at least one training image and the attention guide label which corresponds to each training image and comprises at least one main region.

6. The method of claim 5 , wherein the supervised learning is performed based on a result calculated by substituting the spatial attention map and the attention guide label into a binary cross-entropy loss function.

7. The method of claim 1 , wherein when the bone age analysis model comprises at least two attention modules, the supervised learning on the bone age analysis model is performed based on a result of multiplying a result of a loss function calculated in each attention module by the weight for a loss value of an attention module according to a predetermined attention module and then summing the multiplication results.

8. The method of claim 1 , wherein the attention module comprises a channel attention neural network model for generating a channel attention map regarding a feature map input to the attention module, and a spatial attention neural network model for generating a spatial attention map regarding a modified feature map, and

wherein the modified feature map is a feature map generated by multiplying the feature map input to the attention module by the channel attention map for each element.

9. A method for bone age assessment using a neural network performed by a computing device, the method comprising:

receiving an analysis image which is a target of bone age assessment;

assessing bone age of the target by inputting the analysis image into a bone age analysis model comprising at least one neural network; and

providing a user interface screen comprising a heat map generated based on a spatial attention map for the analysis image,

wherein the bone age analysis model, which is trained by supervised learning based on an attention guide label, comprises at least one attention module for analyzing a main region of the analysis image,

wherein attention guide label comprises importance of each pixel obtained based on a distance between the each pixel included in a training image for training the bone age analysis model and the main region,

wherein the supervised learning is performed based on a comparison result of a spatial attention map generated regarding the training image using the bone age analysis model and the attention guide label corresponding to the training image,

wherein the supervised learning is performed based on:

calculating a loss function by using a label of each pixel included in the attention guide label as a ground truth value and using predicted importance of each pixel included in the spatial attention map as a prediction value, and

updating weights of the attention module so the loss function is minimized.

10. A computing device for bone age assessment, comprising:

one or more processors;

a memory for storing a bone age analysis model comprising one or more neural networks; and

a network unit for receiving an analysis image which is a target of bone age assessment,

and wherein the one or more processors is configured to:

assess bone age of the target by inputting the analysis image into a bone age analysis model comprising one or more neural networks,

wherein the bone age analysis model, which is trained by supervised learning based on an attention guide label, comprises at least one attention module for analyzing a main region of the analysis image,

wherein attention guide label comprises importance of each pixel obtained based on a distance between the each pixel included in a training image for training the bone age analysis model and the main region,

wherein the supervised learning is performed based on a comparison result of a spatial attention map generated regarding the training image using the bone age analysis model and the attention guide label corresponding to the training image,

wherein the supervised learning is performed based on:

calculating a loss function by using a label of each pixel included in the attention guide label as a ground truth value and using predicted importance of each pixel included in the spatial attention map as a prediction value, and

updating weights of the attention module so the loss function is minimized.