Medical image recognition method, model training method, and computer device

1. A medical image recognition method, applied to a computer device, the method comprising:

obtaining a to-be-recognized original medical image set, the to-be-recognized original medical image set comprising at least one to-be-recognized original medical image;

for each to-be-recognized original medical image of the to-be-recognized original medical image set,

obtaining label information of the respective to-be-recognized original medical image in the to-be-recognized original medical image set, the label information comprising information associated with the respective to-be-recognized original medical image,

determining whether the label information of the respective to-be-recognized original medical image satisfies a sample extraction condition,

matching the respective to-be-recognized original medical image with a target medical image in a case that the label information of the respective to-be-recognized original medical image is determined to satisfy the sample extraction condition, the target medical image being a preset image template, and

determining, in a case that the respective to-be-recognized original medical image is successfully matched with the target medical image, that the to-be-recognized original medical image is one of a to-be-recognized medical image set, the to-be-recognized medical image set comprising at least one to-be-recognized medical image;

extracting, with circuitry of the computer device, a to-be-recognized area corresponding to each to-be-recognized medical image in the to-be-recognized medical image set, the to-be-recognized area being a part of the to-be-recognized medical image; and

determining, with the circuitry of the computer device, a recognition result of each to-be-recognized area through a medical image recognition model, the medical image recognition model being obtained through training according to a medical image sample set, the medical image sample set comprising at least one medical image sample, each medical image sample carrying corresponding annotation information, the annotation information being used for representing a type of the medical image sample, the recognition result being used for representing a type of the to-be-recognized medical image.

2. The method according to claim 1 , wherein the extracting comprises:

performing binarization on each to-be-recognized medical image in the to-be-recognized medical image set according to a preset reflected value, to obtain a binary medical image corresponding to each to-be-recognized medical image;

matching each binary medical image by using a target medical image, to extract a to-be-processed area corresponding to each binary medical image, the target medical image being a preset image template;

performing image smoothing on each to-be-processed area, to generate a to-be-extracted outline corresponding to each to-be-processed area, the image smoothing comprising performing at least one of an opening operation and a closing operation on each to-be-processed area; and

extracting a corresponding to-be-recognized area from each to-be-recognized medical image by using each to-be-extracted outline.

3. The method according to claim 1 , wherein after the extracting, the method further comprises:

scaling down the to-be-recognized area in a case that the to-be-recognized area is larger than or equal to a first preset area; and

scaling up the to-be-recognized area in a case that the to-be-recognized area is smaller than or equal to a second preset area.

4. The method according to claim 1 , wherein after the extracting, the method further comprises:

obtaining a reflected value interval corresponding to each to-be-recognized area, the maximum value of the reflected value interval being a first reflected value, the minimum value of the reflected value interval being a second reflected value; and

performing normalization on each to-be-recognized area according to the reflected value interval, to obtain a normalized area; and

the determining the recognition result comprises:

determining a recognition result corresponding to the normalized area of each to-be-recognized area through the medical image recognition model; and

the performing normalization on each to-be-recognized area according to the reflected value interval, to obtain a normalized area comprises:

obtaining the normalized area as:

x

o

⁢

u

⁢

t

=

x

-

x

min

x

ma

⁢

⁢

x

-

x

min

,

wherein x out represents the normalized area, x represents a pixel value of the to-be-recognized area, x max represents the first reflected value, and x min epresents the second reflected value.

5. A non-transitory computer-readable storage medium, storing computer-readable instructions thereon that, when executed by a processor, cause the processor to perform the method according to claim 1 .

6. A model training method, applied to a computer device, the method comprising:

obtaining a to-be-trained original medical image set, the to-be-trained original medical image set comprising at least one to-be-trained original medical image;

for each to-be-trained original medical image of the to-be-trained original medical image set

obtaining label information of the respective to-be-trained original medical image in the to-be-trained original medical image set, the label information comprising information associated with the respective to-be-trained original medical image,

determining whether the label information of the respective to-be-trained original medical image satisfies a sample extraction condition,

matching the respective to-be-trained original medical image with a target medical image in a case that the label information of the respective to-be-trained original medical image is determined to satisfy the sample extraction condition, the target medical image being a preset image template, and

determining, in a case that the respective to-be-trained original medical image is successfully matched with the target medical image, that the to-be-trained original medical image is one of a to-be-trained medical image set, the to-be-trained medical image sample set comprising at least one to-be-trained medical image, each to-be-trained medical image carrying corresponding annotation information, the annotation information being used for representing a type of the to-be-trained medical image;

extracting, with circuitry of the computer device, a to-be-trained area corresponding to each to-be-trained medical image in the to-be-trained medical image sample set, the to-be-trained area being a part of the to-be-trained medical image; and

obtaining a medical image recognition model through training according to the to-be-trained area corresponding to each to-be-trained medical image sample and the corresponding annotation information carried by each to-be-trained medical image sample.

7. The method according to claim 6 , wherein the obtaining a medical image recognition model comprises:

training the to-be-trained area corresponding to each to-be-trained medical image and the corresponding annotation information carried by each to-be-trained medical image through a residual network resnet-18 structure by using a stochastic gradient descent (SGD) algorithm, to obtain a training result;

obtaining a plurality of authentication set loss values according to the training result; and

determining the medical image recognition model according to the plurality of authentication set loss values.

8. The method according to claim 7 , wherein the determining the medical image recognition model comprises:

determining a target authentication set loss value from the plurality of authentication set loss values, the target authentication set loss value being a minimum value in the plurality of authentication set loss values; and

determining a training result corresponding to the target authentication set loss value as the medical image recognition model.

9. The method according to claim 6 , wherein after the obtaining the medical image recognition model, the method further comprises:

obtaining a to-be-recognized medical image set, the to-be-recognized medical image set comprising at least one to-be-recognized medical image;

extracting a to-be-recognized area corresponding to each to-be-recognized medical image in the to-be-recognized medical image set, the to-be-recognized area being a part of the to-be-recognized medical image; and

determining a recognition result of each to-be-recognized area through the medical image recognition model, the recognition result being used for representing a type of the to-be-recognized medical image.

10. A non-transitory computer-readable storage medium, storing computer-readable instructions thereon that, when executed by a processor, cause the processor to perform the method according to claim 6 .

11. A computer device, comprising:

a memory and a processor configured to,

obtain a to-be-recognized original medical image set, the to-be-recognized original medical image set comprising at least one to-be-recognized original medical image;

for each to-be-recognized original medical image of the to-be-recognized original medical image set,

obtain label information of the respective to-be-recognized original medical image in the to-be-recognized original medical image set, the label information comprising information associated with the respective to-be-recognized original medical image,

determine whether the label information of the respective to-be-recognized original medical image satisfies a sample extraction condition,

match the respective to-be-recognized original medical image with a target medical image in a case that the label information of the respective to-be-recognized original medical image is determined to satisfy the sample extraction condition, the target medical image being a preset image template, and

determine, in a case that the respective to-be-recognized original medical image is successfully matched with the target medical image, that the to-be-recognized original medical image is one of a to-be-recognized medical image set, o the to-be-recognized medical image set comprising at least one to-be-recognized medical image;

extract a to-be-recognized area corresponding to each to-be-recognized medical image in the to-be-recognized medical image set, the to-be-recognized area being a part of the to-be-recognized medical image; and

determine a recognition result of each to-be-recognized area through a medical image recognition model, the medical image recognition model being obtained through training according to a medical image sample set, the medical image sample set comprising at least one medical image sample, each medical image sample carrying corresponding annotation information, the annotation information being used for representing a type of the medical image sample, the recognition result being used for representing a type of the to-be-recognized medical image.

12. The computer device according to claim 11 , wherein the processor is configured to:

perform binarization on each to-be-recognized medical image in the to-be-recognized medical image set according to a preset reflected value, to obtain a binary medical image corresponding to each to-be-recognized medical image;

match each binary medical image by using a target medical image, to extract a to-be-processed area corresponding to each binary medical image, the target medical image being a preset image template;

perform image smoothing on each to-be-processed area, to generate a to-be-extracted outline corresponding to each to-be-processed area, the image smoothing comprising performing at least one of an opening operation and a closing operation on each to-be-processed area; and

extract the corresponding to-be-recognized area from each to-be-recognized medical image by using each to-be-extracted outline.

13. The computer device according to claim 11 , wherein the processor is further configured to:

scale down the to-be-recognized area in a case that the to-be-recognized area is larger than or equal to a first preset area; and

scale up the to-be-recognized area in a case that the to-be-recognized area is smaller than or equal to a second preset area.

14. The computer device according to claim 11 , wherein the processor is further configured to:

obtain a reflected value interval corresponding to each to-be-recognized area, the maximum value of the reflected value interval being a first reflected value, the minimum value of the reflected value interval being a second reflected value.

15. The computer device according to claim 14 , wherein the processor is further configured to:

performing normalization on each to-be-recognized area according to the reflected value interval, to obtain a normalized area.

16. The computer device according to claim 15 , wherein the processor is further configured to:

determine a recognition result corresponding to the normalized area of each to-be-recognized area through the medical image recognition model; and

the processor is configured to obtain the normalized area as:

x

o

⁢

u

⁢

t

=

x

-

x

min

x

ma

⁢

⁢

x

-

x

min

,

wherein x out represents the normalized area, x represents a pixel value of the to-be-recognized area, x max represents the first reflected value, and x min r epresents the second reflected value.