Method of training model for identification of disease, electronic device using method, and non-transitory storage medium
View Patent ↗A method training an AI model in disease identification establishes a disease identifying model, the model includes a convolutional neural network and a pyramid attention network. The pyramid attention network receives output of the convolutional neural network. The method obtains feature map sample set, the sample set being classified into training sets and verification sets. The method inputs each training set into the disease identifying model to train the model and outputs values of degree of confidence in correct identification of diseases. The method further verifies the trained models according to the verification sets. An electronic device and a non-transitory storage medium are also disclosed.
1. A method of training disease identifying model comprising:
establishing a disease identifying model, the disease identifying model comprising a convolutional neural network and a pyramid attention network, the pyramid attention network being connected to an output of the convolutional neural network;
obtaining feature map sample set, the feature map sample set being classified into one or more training sets and one or more verification sets;
inputting each of the training sets into the disease identifying model to train the disease identifying model according to the one or more training sets and output values of degree of confidence in correct identification of diseases; and
verifying the disease identifying models after being trained according to the one or more verification sets and the values of degree of confidence in correct identification of diseases;
wherein the obtaining feature map sample set comprises:
obtaining a preset number of the medical images; and
preprocessing the obtained preset number of the medical images to obtain the feature map sample set;
wherein the preprocessing the obtained preset number of the medical images to obtain the feature map sample set comprises:
resampling the medical images;
converting the medical images after resampling to three-dimensional RGB images;
converting the three-dimensional RGB images into the feature maps of the medical images.
2. The method according to claim 1 , wherein the converting the three-dimensional RGB images into the feature maps of the medical images comprises:
mean normalizing the converted three-dimensional RGB images;
performing an image enhancement on the three-dimensional RGB image after the mean normalization, to convert to the feature maps of the medical images.
3. The method according to claim 1 , wherein:
the establishing the disease identifying model comprises:
selecting one convolutional neural network;
selecting one pyramid attention network;
wherein the disease identifying model comprising the convolutional neural network and the pyramid attention network comprises:
the disease identifying model comprising the convolutional neural network and the pyramid attention network if the pyramid attention network needs to be used in the disease identifying model;
outputting values of degree of confidence in correct identification of diseases comprises:
outputting the values of degree of confidence in correct identification of diseases via the pyramid attention network.
4. The method according to claim 3 , wherein the method further includes:
outputting the values of degree of confidence in correct identification of diseases via a preset function of the convolutional neural network if the pyramid attention network does not need to be used in the disease identifying model.
5. The method according to claim 4 , wherein the method further includes:
determining whether the value of degree of confidence in correct identification of disease is greater than a preset threshold;
outputting the disease corresponding to the value of degree of confidence in correct identification of disease if the value of degree of confidence in correct identification of disease is greater than the prese threshold.
6. An electronic device comprising:
a storage device;
at least one processor; and
the storage device storing one or more programs, which when executed by the at least one processor, cause the at least one processor to:
establish a disease identifying model, the disease identifying model comprising a convolutional neural network and a pyramid attention network, the pyramid attention network being connected to an output of the convolutional neural network;
obtain feature map sample set, the feature map sample set being classified into one or more training sets and one or more verification sets;
input each of the training sets into the disease identifying model to train the disease identifying model according to the one or more training sets and output values of degree of confidence in correct identification of diseases; and
verify the disease identifying models after being trained according to the one or more verification sets and the values of degree of confidence in correct identification of diseases;
obtain a preset number of the medical images; and
preprocess the obtained preset number of the medical images to obtain the feature map sample set;
resample the medical images;
convert the medical images after resampling to three-dimensional RGB images;
convert the three-dimensional RGB images into the feature maps of the medical images.
7. The electronic device according to claim 6 , further causing the at least one processor to:
mean normalize the converted three-dimensional RGB images;
perform an image enhancement on the three-dimensional RGB image after the mean normalization, to convert to the feature maps of the medical images.
8. The electronic device according to claim 6 , further causing the at least one processor to:
select one convolutional neural network;
select one pyramid attention network; where the disease identifying model comprising the convolutional neural network and the pyramid attention network if the pyramid attention network needs to be used in the disease identifying model;
output the values of degree of confidence in correct identification of diseases via the pyramid attention network.
9. The electronic device according to claim 8 , further causing the at least one processor to:
output the values of degree of confidence in correct identification of diseases via a preset function of the convolutional neural network if the pyramid attention network does not need to be used in the disease identifying model.
10. The electronic device according to claim 9 , further causing the at least one processor to:
determine whether the values of degree of confidence in correct identification of disease is greater than a preset threshold;
output the disease corresponding to the values of degree of confidence in correct identification of disease if the values of degree of confidence in correct identification of the disease is greater than the prese threshold.
11. A non-transitory storage medium storing a set of commands, when the commands being executed by at least one processor of an electronic device, causing the at least one processor to:
establish a disease identifying model, the disease identifying model comprising a convolutional neural network and a pyramid attention network, the pyramid attention network being connected to an output of the convolutional neural network;
obtain feature map sample set, the feature map sample set being classified into one or more training sets and one or more verification sets;
input each of the training sets into the disease identifying model to train the disease identifying model according to the one or more training sets and output values of degree of confidence in correct identification of the diseases; and
verify the disease identifying models after being trained according to the one or more verification sets and the values of degree of confidence in correct identification of diseases;
obtain a preset number of the medical images; and
preprocess the obtained preset number of the medical images to obtain the feature map sample set;
resample the medical images;
convert the medical images after resampling to three-dimensional RGB images;
convert the three-dimensional RGB images into the feature maps of the medical images.
12. The non-transitory storage medium according to claim 11 , further causing the at least one processor to:
select one convolutional neural network;
select one pyramid attention network; where the disease identifying model comprising the convolutional neural network and the pyramid attention network if the pyramid attention network needs to be used in the disease identifying model;
output the values of degree of confidence in correct identification of diseases via the pyramid attention network.
13. The non-transitory storage medium according to claim 12 , further causing the at least one processor to:
output the values of degree of confidence in correct identification of disease via a preset function of the convolutional neural network if the pyramid attention network does not need to be used in the disease identifying model.
14. The non-transitory storage medium according to claim 13 , further causing the at least one processor to:
determine whether the value of degree of confidence in correct identification of disease is greater than a preset threshold;
output the disease corresponding to the value of degree of confidence in correct identification of disease if the value of degree of confidence in correct identification of disease is greater than the prese threshold.