Medical image processing apparatus and medical image processing method

1 . A medical image processing apparatus comprising:

processing circuitry configured to

train an attention model by using a medical image and a mask image that is obtained by performing mask processing on a region other than a region of interest in the medical image,

train an image processing model by using the medical image, a heatmap corresponding to the medical image, and an attention feature that indicates attention of the region of interest in the trained attention model,

combine, as an attention image processing model, an attention extraction module that includes an attention module for outputting the attention feature and the trained image processing model,

perform processing on the medical image by using the attention image processing model,

generate a heatmap for determining a position of a landmark in the medical image,

train a plurality of attention models so as to correspond to a plurality of organs in the medical image,

generate a plurality of attention features corresponding to the plurality of organs,

train the image processing model such that the medical image and the plurality of attention features corresponding to the plurality of organs are adopted as input of the image processing model, the plurality of attention features and a detection feature of the image processing model are concatenated with one another, and heatmaps of landmarks of a plurality of combinations corresponding to the plurality of attention features that correspond to the plurality of organs are adopted as output of the image processing model, and

train the image processing model with respect to the plurality of organs by using mask images corresponding to the plurality of organs.

2 . The medical image processing apparatus according to claim 1 , wherein the processing circuitry is configured to acquire the attention feature, the attention module, and the attention extraction module in the trained attention model.

3 . The medical image processing apparatus according to claim 1 , wherein

the processing circuitry is configured to train the image processing model such that the medical image and the attention feature are adopted as input of the image processing model, the attention feature at corresponding resolution and a detection feature of the image processing model are concatenated with each other, and a heatmap of the landmark in the medical image is adopted as output of the image processing model, and

a value of each of pixels in the heatmap indicates a probability that a located position of the pixel serves as the landmark.

4 . The medical image processing apparatus according to claim 3 , wherein the attention extraction module includes an encoder module and the attention module in the attention model.

5 . The medical image processing apparatus according to claim 4 , wherein the image processing model is one of a landmark position detection model and a region division model for the medical image.

6 . The medical image processing apparatus according to claim 5 , wherein

the medical image is an image in which a pancreas is captured, and

a position of the landmark is a position of the pancreatic tail.

7 . A medical image processing method comprising:

training an attention model by using a medical image and a mask image that is obtained by performing mask processing on a region other than a region of interest in the medical image;

training an image processing model by using the medical image, a heatmap corresponding to the medical image, and an attention feature that indicates attention of the region of interest in the trained attention model;

combining, as an attention image processing model, an attention extraction module that includes an attention module for outputting the attention feature and the trained image processing model;

performing processing on the medical image by using the attention image processing model;

generating a heatmap for determining a position of a landmark in the medical image;

training a plurality of attention models so as to correspond to a plurality of organs in the medical image;

generating a plurality of attention features corresponding to the plurality of organs; and

training the image processing model such that the medical image and the plurality of attention features corresponding to the plurality of organs are adopted as input of the image processing model, the plurality of attention features and a detection feature of the image processing model are concatenated with one another, and heatmaps of landmarks of a plurality of combinations corresponding to the plurality of attention features that correspond to the plurality of organs are adopted as output of the image processing model, wherein the method comprising training the image processing model with respect to the plurality of organs by using mask images corresponding to the plurality of organs.