Method and system for image recognition and computer readable storage medium

1 . A method for image recognition, comprising: executing an image recognition process by a processor, wherein the image recognition process comprises:

obtaining an image to be recognized;

cutting the image to be recognized into a plurality of tiles, wherein a size of each of the tiles conforms to a preset size;

respectively inputting the tiles into a trained artificial intelligence model, respectively obtaining a plurality of specified categories corresponding to the plurality of tiles, and setting the specified categories as a plurality of judgment results of the tiles, wherein each of the specified categories is a no-lesion category or one of a plurality of level categories belonging to a same lesion;

counting a number of the specified categories belonging to the no-lesion category and a number of the specified categories belonging to each of the level categories;

determining a lesion level of the lesion based on the number of the specified categories belonging to the no-lesion category and the number of the specified categories belonging to each of the level categories, wherein in response to all the specified categories are no-lesion categories, the lesion level is set to a lowest, in response to the specified categories include different level categories, the lesion level is set according to a severity of each of the level categories; and

storing the judgment results and the lesion level as a structured report to be output to a digital pathology platform.

2 . The method for image recognition according to claim 1 , wherein the step of obtaining the image to be recognized comprises:

obtaining the image to be recognized from an image server in response to receiving an inference request from the digital pathology platform.

3 . The method for image recognition according to claim 1 , further comprising executing one of the following steps by the processor:

regularly executing the image recognition process based on a preset time; and

executing the image recognition process in response to a trigger event.

4 . The method for image recognition according to claim 1 , wherein the image recognition process further comprises:

after obtaining the image to be recognized and before cutting the image to be recognized, converting a format of the image to be recognized into a specified format.

5 . The method for image recognition according to claim 1 , wherein the step of storing the judgment results as the structured report comprises:

combining the judgment results of the tiles cut from the image to be recognized to obtain an inference result corresponding to the image to be recognized; and

storing the inference result as the structured report.

6 . The method for image recognition according to claim 5 , wherein each of the judgment results further comprises regional coordinate information, and the step of combining the judgment results of the tiles cut from the image to be recognized to obtain the inference result corresponding to the image to be recognized comprises:

obtaining a lesion area corresponding to the image to be recognized based on the plurality of regional coordinate information in the judgment results.

7 . The method for image recognition according to claim 5 , wherein the image recognition process further comprises:

displaying the image to be recognized in the digital pathology platform in a form of a heat map based on the inference result, wherein a plurality of pixels included in a recognized lesion area in the image to be recognized are represented by a first color, and a plurality of pixels not included in the lesion area are represented by a second color.

8 . The method for image recognition according to claim 1 , further comprising: executing a training process on the artificial intelligence model by the processor, wherein the training process comprises:

obtaining a plurality of training images and annotation content corresponding to each of the training images;

cutting each of the training images into a plurality of training tiles, wherein a size of each of the training tiles conforms to the preset size;

classifying the training tiles cut from each of the training images into the specified categories based on the annotation content corresponding to each of the training images; and

inputting the training tiles corresponding to the specified categories into the artificial intelligence model for training.

9 . The method for image recognition according to claim 8 , wherein the training images comprise a plurality of pathological images, and the annotation content of each of the pathological images comprises location information of a pathological area and a pathological label of the pathological area,

the step of classifying the training tiles cut from each of the training images into the specified categories based on the annotation content corresponding to each of the training images comprises:

determining whether each of the training tiles covers at least a part of the pathological area;

classifying the training tiles covering at least a part of the pathological area into a pathology category corresponding to the pathological label, wherein the pathology category is one of the specified categories; and

classifying the training tiles that do not cover at least a part of the pathological area to the no-lesion category.

10 . A system for image recognition, comprising:

a storage device storing a trained artificial intelligence model; and

a processor coupled to the storage device and configured to execute an image recognition process, wherein the image recognition process comprises:

obtaining an image to be recognized;

cutting the image to be recognized into a plurality of tiles, wherein a size of each of the tiles conforms to a preset size;

respectively inputting the tiles into the trained artificial intelligence model, respectively obtaining a plurality of specified categories corresponding to the plurality of tiles, and setting the specified categories as a plurality of judgment results of the tiles, wherein each of the specified categories is a no-lesion category or one of a plurality of level categories belonging to a same lesion;

counting a number of the specified categories belonging to the no-lesion category and a number of the specified categories belonging to each of the level categories;

determining a lesion level of the lesion based on the number of the specified categories belonging to the no-lesion category and the number of the specified categories belonging to each of the level categories, wherein in response to all the specified categories are no-lesion categories, the lesion level is set to a lowest, in response to the specified categories include different level categories, the lesion level is set according to a severity of each of the level categories; and

storing the judgment results and the lesion level as a structured report to be output to a digital pathology platform.

11 . The system for image recognition according to claim 10 , wherein the processor is configured to:

obtain the image to be recognized from an image server in response to receiving an inference request from the digital pathology platform.

12 . The system for image recognition according to claim 10 , wherein the processor is configured to execute one of the following steps:

regularly executing the image recognition process based on a preset time; and

executing the image recognition process in response to a trigger event.

13 . The system for image recognition according to claim 10 , wherein the processor is configured to:

in the image recognition process, after obtaining the image to be recognized and before cutting the image to be recognized, convert a format of the image to be recognized into a specified format.

14 . The system for image recognition according to claim 10 , wherein the processor is configured to:

in the image recognition process, combine the judgment results of the tiles cut from the image to be recognized to obtain an inference result corresponding to the image to be recognized; and

store the inference result as the structured report.

15 . The system for image recognition according to claim 14 , wherein each of the judgment results further comprises regional coordinate information, and the processor is configured to:

in the image recognition process, obtain a lesion area corresponding to the image to be recognized based on the plurality of regional coordinate information in the judgment results.

16 . The system for image recognition according to claim 14 , wherein the processor is configured to:

in the image recognition process, display the image to be recognized in the digital pathology platform in a form of a heat map based on the inference result, wherein a plurality of pixels included in a recognized lesion area in the image to be recognized are represented by a first color, and a plurality of pixels not included in the lesion area are represented by a second color.

17 . The system for image recognition according to claim 10 , wherein the processor is configured to execute a training process on the artificial intelligence model, and the training process comprises:

obtaining a plurality of training images and annotation content corresponding to each of the training images;

cutting each of the training images into a plurality of training tiles, wherein a size of each of the training tiles conforms to the preset size;

classifying the training tiles cut from each of the training images into the specified categories based on the annotation content corresponding to each of the training images; and

inputting the training tiles corresponding to the specified categories into the artificial intelligence model for training.

18 . The system for image recognition according to claim 17 , wherein the training images comprise a plurality of pathological images, and the annotation content of each of the pathological images comprises location information of a pathological area and a pathological label of the pathological area,

the processor is configured to:

in the training process, determine whether each of the training tiles covers at least a part of the pathological area;

classify the training tiles covering at least a part of the pathological area into a pathology category corresponding to the pathological label, wherein the pathology category is one of the specified categories; and

classify the training tiles that do not cover at least a part of the pathological area to a no-lesion category.

19 . A non-transitory computer readable medium, storing a plurality of program instructions, wherein the program instructions are loaded through an electronic device to execute following steps comprising:

obtaining an image to be recognized;

cutting the image to be recognized into a plurality of tiles, wherein a size of each of the tiles conforms to a preset size;

respectively inputting the tiles into a trained artificial intelligence model, respectively obtaining a plurality of specified categories corresponding to the plurality of tiles, and setting the specified categories as a plurality of judgment results of the tiles, wherein each of the specified categories is a no-lesion category or one of a plurality of level categories belonging to a same lesion;

counting a number of the specified categories belonging to the no-lesion category and a number of the specified categories belonging to each of the level categories;

determining a lesion level of the lesion based on the number of the specified categories belonging to the no-lesion category and the number of the specified categories belonging to each of the level categories, wherein in response to all the specified categories are no-lesion categories, the lesion level is set to a lowest, in response to the specified categories include different level categories, the lesion level is set according to a severity of each of the level categories; and

storing the judgment results and the lesion level as a structured report to be output to a digital pathology platform.