Method for predicting disease based on medical image

1. A method for predicting disease based on a medical image, performed by a computing device including one or more processors, the method comprising:

generating a feature vector related to predictive values of brain disease for each of two-dimensional (2D) medical images included in a three-dimensional (3D) medical image, using a pre-trained first model;

estimating importance indicating prediction accuracy for each of the 2D medical images based on the feature vector, using a pre-trained second model; and

selecting at least one model input image suitable for prediction of the brain disease from among the 2D medical images based on the importance.

2. The method of claim 1 , further comprising:

extracting a brain parenchyma region from each of the 2D medical images in which a position of a brain region is aligned with respect to a template, using a pre-trained third model.

3. The method of claim 1 , wherein the first model generates the feature vector based on the 2D medical images, personal information about a subject of the 3D medical image, and number information indicating positions of the 2D medical images in the 3D medical image.

4. The method of claim 3 , wherein the first model includes:

a first neural network extracting features for prediction of the brain disease from the 2D medical images; and

a second neural network that outputs the predictive values of brain disease for each of the 2D medical images, based on the features extracted by the first neural network, the personal information, and the number information.

5. The method of claim 1 , wherein the importance is estimated based on a tree boosting algorithm used in the second model which takes the feature vector as input.

6. The method of claim 1 , wherein the selecting the at least one model input image suitable for prediction of the brain disease includes:

selecting at least one of the 2D medical images as the model input image by comparing the importance and a threshold value; and

obtaining reference number information indicating a position of the selected model input image in the 3D medical image.

7. The method of claim 1 , further comprising:

using the selected at least one model input image as training data for a disease diagnosis model.

8. A method for predicting disease based on a medical image, performed by a computing device including one or more processors, the method comprising:

generating information for parcellating at least one brain region present in each of two-dimensional (2D) medical images included in a three-dimensional (3D) medical image, using a pre-trained segmentation model; and

selecting an image satisfying predetermined condition related to a reference region from among the 2D medical images based on the generated information for parcellating as at least one model input image suitable for prediction of a brain disease,

wherein the reference region is a brain region representing a structure for determining the brain disease.

9. The method of claim 8 , wherein the reference region includes a medial temporal lobe, and

wherein the parcellated at least on brain region includes at least one of a hippocampus, a amygdala, a entorhinal cortex, a parahippocampal cortex, or a ventricle.

10. The method of claim 8 , further comprising:

using the selected at least one model input image as training data for a disease diagnosis model.

11. A method for predicting disease based on a medical image, performed by a computing device including one or more processors, the method comprising:

extracting a brain parenchyma region respectively from two-dimensional (2D) medical images in which a position of a brain region is aligned based on a template, using a pre-trained preprocess model;

selecting at least one image corresponding to reference number information among the 2D medical images from which the brain parenchyma region is extracted as at least one model input image suitable for prediction of a brain disease; and

predicting a presence probability of the brain disease based on the selected model input image, using a pre-trained disease diagnosis model.

12. The method of claim 11 , wherein the selecting the at least one model input image suitable for prediction of the brain disease includes:

wherein the reference number information is predetermined based on importance indicating prediction accuracy of the brain disease for each of the 2D medical images included in a three-dimensional (3D) medical image.

13. The method of claim 11 , wherein the reference number information is predetermined based on number information indicating where the 2D medical image determined as the model input image is located in a three-dimensional (3D) medical image.

14. The method of claim 11 , wherein the disease diagnosis model is trained based on at least one image selected as suitable for prediction of the brain disease among the 2D medical images included in a three-dimensional (3D) medical image.