KR102044525B1 - Method and apparatus for generating graft rejection detection model - Google Patents

Abstract

According to an embodiment of the present invention, a method for generating a transplant rejection determination model according to an embodiment of the present invention receives a pathological image and selects and outputs a plurality of detection target regions each having a predetermined size or more and having a first size for the pathological image. Training a first neural network; and training a second neural network for determining the PTC + region, the PTC- region, and the non-PTC region included in each region to be detected.

Description

METHOD AND APPARATUS FOR GENERATING GRAFT REJECTION DETECTION MODEL}

The present invention relates to a method and apparatus for generating a transplant rejection discrimination model, and more particularly, to determine whether a rejection reaction occurs in a patient transplanted with an organ by specifying a region used for diagnosing the transplant rejection in a pathological image. The present invention relates to a method and apparatus for generating a transplant rejection determination model for diagnosing types.

As the number of organ donors has recently increased worldwide, the number of patients receiving organ transplants has increased. In this case, if a rejection reaction occurs after transplanting the organ to the patient, the patient needs to undergo a strong immunosuppressive treatment or plasma exchange procedure, and thus, a lot of costs may be incurred, and death may be severe if the rejection reaction is severe. Accordingly, in case of rejection in the transplanted organ, the patient should be treated immediately and appropriately to minimize the graft failure. Therefore, in order to determine the appropriate treatment method as the rejection occurs, it is essential to accurately diagnose the pathology by observing the tissue collected from the transplanted organ of the patient.

Currently, pathological diagnosis of such graft rejection relies on physician visual assessment of pathological images, which are typically very large, at doses of more than 10 gigabytes, and individually for quantitative evaluation of each histological component. Should be done. In addition, only a few pathologists who have specialized training in rejection can diagnose it, even a very long time is needed to diagnose transplant rejection in one patient. As such, the final judgment on the diagnosis of graft rejection depends on the physician's visual assessment and thus tends to be less reproducible and accurate in diagnosis. Therefore, in order to secure the reproducibility and accuracy of diagnosis of graft rejection, a technique for diagnosing graft rejection using objective means is required.

Korean Patent Registration No. 10-1754291 (registered June 29, 2017)

The problem to be solved in the embodiments of the present invention is to provide a technique for diagnosing a transplant rejection by specifying a specific element expressed in the graft using machine learning neural network technology.

However, the technical problem to be achieved by the embodiment of the present invention is not limited to the above-mentioned problem, various technical problems can be derived within the scope apparent to those skilled in the art from the following description.

In the method for generating a transplant rejection determination model according to an embodiment of the present invention, a pathological image is input, and a plurality of detection target regions each having a predetermined size and a first size are selected for the pathological image. Training the first neural network to be output, and for each detection target region, a positive peritubular capillary (PTC +) region, a negative peritubular capillary (PTC-) region, and a non-peritubular capillary (non-PTC) region included in the detection target region. Training the second neural network to determine.

In this case, the pathological image may be an image from which information included in an area other than the tissue area is removed.

The training of the first neural network may include receiving a learning pathology image including a tissue region, selecting a plurality of first samples having the first size within the learning pathology image, and the plurality of first pathologies. Labeling each sample to be selected or not selected as the detection target region, and learning the first neural network based on the label and the plurality of first samples to determine the predetermined ratio. can do.

In addition, the training of the second neural network may include receiving masking information on the plurality of second samples having the first size and the PTC +, PTC-, and non-PTCs included in the plurality of second samples. Specifying PTC +, PTC-, non-PTC in the plurality of second samples based on the information, including the final peripheral region in the specified PTC +, PTC- and labeling them with PTC + region, PTC- region respectively; Labeling a non-PTC region for a specified non-PTC and training the second neural network based on the label and the plurality of second samples.

In addition, the labeling may include generating a first test neural network based on the first labeling information labeled with the first PTC + region and the first PTC- region by including a first peripheral region in the detected PTC + and PTC-, respectively. Based on the second labeling information labeled with the second PTC + region, the second PTC- region, and the second non-PTC region by including a second peripheral region having a different size from the first peripheral region in the detected PTC + and PTC-, respectively. Generating a second test neural network, receiving a test sample having a PTC + region, a PCT- region, and a non-PTC region respectively, and inputting the test samples to the first test neural network and the second test neural network, respectively. Determining a PTC +, PTC-, non-PTC region in the test sample, and determining the PTC + region, PTC- region, non-PTC region in the predetermined test sample and the determined test sample. Comparing the PTC + region, the PTC- region, and the non-PTC region, determining the size of the peripheral region used for the more accurate neural network among the first test neural network and the second test neural network as the final peripheral region. It may include.

In addition, the labeling may include labeling the non-PTC region with a region reduced in a predetermined ratio than the specified non-PTC region.

In this case, the size of the final peripheral area may include an area of 30 pixels or more and 50 pixels or less.

The non-PTC may also generate a test neural network through some of the plurality of second samples and incorrectly identify the PTC + region when entering the test neural network as the rest of the plurality of second samples, tubular and It may comprise at least one of glomerulus.

The method for determining a graft rejection response according to an embodiment of the present invention is performed by the first neural network and the second neural network, and the first neural network receives a target pathological image to detect a plurality of detection target regions having a predetermined size. Outputting, the second neural network receiving the plurality of detection target regions output by the first neural network, and determining PTC + regions, PTC-regions, and non-PTC regions within the plurality of detection target regions; Calculating a C4d score of the target pathological image based on the number of the PTC + region and the PTC- region.

According to an embodiment of the present invention, since the doctor uses a sample divided into a predetermined size for labeling a specific area of the pathological image for learning, the reproducibility and accuracy of the diagnosis may be improved.

In addition, the model is trained not only to identify PTC + regions and PTC- regions but also to distinguish non-PTC regions so that the model can better distinguish PTC + regions and PTC- regions within pathological images. You can.

In addition, the reproducibility and accuracy can be improved by determining the optimal size for labeling each of the PTC + region, PTC- region and non-PTC region included in the pathological image.

Accordingly, the embodiment of the present invention enables a faster and more accurate determination of the diagnosis of the graft rejection reaction, which can greatly reduce the time and labor required for the diagnosis, as well as the convenience of diagnosis to both the doctor and the patient. Can increase safety.

1 is a flowchart illustrating a process of a method for generating a transplant rejection determination model according to an embodiment of the present invention.
FIG. 2 is an exemplary diagram for describing labeling at a specific position in a second sample based on masking information according to an embodiment of the present invention.
FIG. 3 is an exemplary diagram for explaining that the accuracy of a trained model is changed by labeling each of PTC +, PTC-, and non-PTC to include a portion of a peripheral area thereof.
4 is a flowchart illustrating a process of a transplant rejection determination method according to an embodiment of the present invention.

Advantages and features of the present invention, and methods of achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but can be implemented in various forms. The present embodiments merely make the disclosure of the present invention complete, and have the ordinary knowledge in the art to which the present invention pertains. It is provided to fully inform the scope of the invention, and the scope of the invention is defined only by the claims.

In describing the embodiments of the present invention, detailed descriptions of well-known functions or configurations will be omitted unless they are actually necessary in describing the embodiments of the present invention. In addition, terms to be described below are terms defined in consideration of functions in the embodiments of the present invention, which may vary according to intentions or customs of users and operators. Therefore, the definition should be made based on the contents throughout the specification.

The functional blocks shown in the figures and described below are only examples of possible implementations. Other functional blocks may be used in other implementations without departing from the spirit and scope of the detailed description. Also, while one or more functional blocks of the present invention are represented by separate blocks, one or more of the functional blocks of the present invention may be a combination of various hardware and software configurations that perform the same function.

In addition, the expression "comprising" certain components merely refers to the existence of the corresponding components as an open expression, and should not be understood as excluding additional components.

Furthermore, when a component is referred to as being connected or connected to another component, it is to be understood that although the component may be directly connected or connected to the other component, there may be other components in between.

In addition, an expression such as 'first' and 'second' is used only for distinguishing a plurality of components, and does not limit the order or other features between the components.

Embodiments of the present invention generate and use a transplant rejection discrimination model for determining regions used for diagnosis of rejection in pathological images. Here, the pathological image refers to an image including information for determining the cause, occurrence, and progress of the disease. For example, the pathology image may be an image including information about tissues collected from the organs in which the patient is transplanted in order to diagnose whether rejection occurs after transplanting the organs to the patient. However, pathological images are not limited to these examples.

To this end, training data for training a graft rejection determination model to recognize a specific region in a pathological image is required. On the other hand, since the pathological image is very large with a capacity of at least 10 gigabytes or more, it is difficult to label all the specific areas on the pathological image itself. Accordingly, an embodiment of the present invention trains a model by labeling a specific position in a detection target region in which a pathological image is segmented by a size observed by a doctor visually when diagnosing a graft rejection (eg, 1024 x 1024 pixels).

As described above, since the exemplary embodiment of the present invention trains the model based on the detection target region of the size observed when diagnosing the graft rejection response, the pathological image to confirm the graft rejection response also includes the size of the detection target region used in the training. Should be used after dividing to have the same size.

Accordingly, the transplant rejection determination model according to an embodiment of the present invention comprises a first neural network for dividing and outputting a target pathological image to be used for diagnosing the transplant rejection into the same size as the size of the detection target region used for learning; A second neural network is used to determine a specific region used for diagnosing transplant rejection within each detection target region output through the neural network.

On the other hand, the drawings and embodiments of the present invention for explaining the diagnosis of renal transplant rejection using a pathological image of the renal graft for convenience of understanding, but the embodiment of the present invention is not only renal graft but also various pathological images As a technique that can be utilized, it is not limited to the diagnosis of renal transplant rejection. First, the method of learning the first neural network and the second neural network constituting the transplant rejection determination model according to an embodiment of the present invention in conjunction with FIGS. 1 to 3, and then using the trained model with FIG. Describe how to determine transplant rejection.

1 is a flowchart illustrating a process of a method for generating a transplant rejection determination model according to an embodiment of the present invention. Each step of the method for generating a transplant rejection determination model according to FIG. 1 may be performed by one or more processors, which will be described below.

First, a first neural network including a tissue area having a predetermined ratio or more from a pathology image and selecting and outputting a plurality of detection target areas having a first size is trained (S110).

The pathological image to be input to the first neural network may be a pathological image from which information included in a region other than the tissue region is removed through Otsu's thresholding method. Accordingly, the pathological image is divided into a portion corresponding to the tissue of the patient and a non-tissue portion (a portion from which information is removed when using Otsu's thresholding method). In this case, since the non-tissue part does not include an area for determining a transplant rejection response, it does not need to be used as learning data or input data of the second neural network, and inputs an image including the non-tissue area to the second neural network. If so, there is a possibility of slowing down the overall speed of the model. In order to prevent this, a region including a tissue region of the pathological image may be trained to be selected as a detection target region within a first size (eg, 1024 x 1024 pixels) at a predetermined ratio or more.

To this end, the processor may receive a learning pathology image including a tissue region and select a plurality of first samples having a first size in the learning pathology image. For example, a plurality of first samples may be extracted while zigzag scanning the entire learning pathology image through a window of a first size, or a plurality of first samples may be extracted by capturing an arbitrary area of the pathological image. can do.

Then, whether each of the plurality of first samples is selected as a detection target region or not is selected. For example, a doctor may input a label for each of the plurality of first samples to input learning information whether or not to select the first neural network as a detection target region, or use a known image discrimination algorithm. A label representing an output is set to a first sample including a tissue region of a pathology image at a predetermined ratio or more (for example, 50% or more) of the plurality of first samples, and less than a predetermined ratio (for example, 50) within a predetermined size. After setting a label meaning non-output to the first sample including the tissue region of the pathological image (%), the first neural network may be trained based on the plurality of first samples and labels respectively set therein.

Accordingly, the first neural network may be trained to select and output the detection subject region including the tissue region of the pathological image at a predetermined ratio or more within the first size.

In addition, a first sample comprising two or more specific peritubular capillary (PTC) in a first size of the plurality of first samples, the first sample including an area in which tubular and vessels can be clearly distinguished within a predetermined size A label representing an output is set for at least one of one sample, a first sample including 50% or more of an intact tubular region within a predetermined size, or a blank space within a predetermined size of the plurality of first samples. By setting a label meaning non-output for the first sample including% or more, the characteristic of the detection target region to be output by the first neural network may be determined in various ways.

Next, a second neural network is trained to determine a specific region used to diagnose a transplant rejection in each detection target region output through the first neural network (S120).

For example, if the training data used is a pathological image of a renal graft, the second neural network may be a peritubular capillary (PTC) that the pathological image contains, such as a positive peritubular capillary (PTC +) region and a PTC- ( negative peritubular capillary).

To this end, the processor may perform an operation of receiving a plurality of second samples having a first size (for example, 1024 x 1024 pixels).

In this case, the second sample may be existing data observed when the doctor visually diagnoses the graft rejection reaction, or may be a detection target region outputted by inputting a pathological image to the first neural network.

In addition, the processor may perform an operation of receiving masking information on PTC +, PTC-, and non-PTC included in the plurality of second samples. The masking information is information for labeling PTC +, PTC-, non-PTC positions in the second sample. The processor specifies and labels PTC +, PTC-, and non-PTCs in the plurality of second samples based on this masking information, and learns a second neural network based on the information labeled in the plurality of second samples and the second sample. You can.

FIG. 2 is an exemplary diagram for describing labeling at a specific position in a second sample based on masking information according to an embodiment of the present invention.

Referring to FIG. 2, the masking information has the same size as the second sample and is masked at a position corresponding to a specific region of the second sample. By comparing the masking information with the second sample, the processor may label it as meaning that the inside of the rectangular area including the position corresponding to the masking position in the second sample refers to a specific area. For example, if the masking information is masked at a position corresponding to a tubular region in the second sample, the masking information may be labeled to mean that the inside of the rectangle including the masked position in the second sample is a non-PTC region. PTC + and PTC- can also be labeled in the same manner as in the example above.

At this time, it was confirmed that the trained model did not have high discrimination accuracy between the PTC + region and the PTC- region by setting a rectangular area having a minimum size including only PTC + and PTC- as a label. In the embodiment of the present invention, it is assumed that this is because, when labeling a rectangular area having a minimum size including only PTC + and PTC- as a label, the relationship between the positions where PTC + and PTC- are present and the surrounding area thereof is not considered. As shown in the embodiment of FIG. 3, when the label is set to include the area for discrimination and its surrounding area, it is confirmed that the discrimination accuracy of the model is changed.

FIG. 3 is an exemplary diagram for explaining that the accuracy of the trained model is changed by labeling each of PTC + and PTC- to include a portion of a peripheral area thereof.

Referring to FIG. 3, when the second neural network is trained, labeling information may be generated by varying the size of the rectangle including the masking to include PTC + and PTC- surrounding areas.

For example, the first peripheral region is included in the PTC + and PTC- specified in the second sample based on the masking information, and the first peripheral region is labeled based on the first labeling information labeled with the first PTC + region and the first PTC- region, respectively. A test neural network is generated, and based on the second labeling information labeled with the second PTC + region and the second PTC-region, respectively, by including the second peripheral region having a different size from the first peripheral region in the detected PTC + and PTC- 2 Generate test neural networks.

Subsequently, a test sample having a PTC + region, a PCT- region, and a non-PTC region is input thereto, and the test samples are input to the first test neural network and the second test neural network, respectively, and then PTC +, PTC-, and non- Determine the PTC region and compare the PTC + region, the PTC- region, the non-PTC region in the test sample as a predetermined answer with the PTC + region, the PTC- region, and the non-PTC region in the test sample determined through the neural network, The size of the peripheral region used for the higher accuracy neural network among the first test neural network and the second test neural network may be determined as the final peripheral region to be used for the final learning of the second neural network.

In the case of FIG. 3, when the rectangle is the smallest size containing PTC + and PTC- and does not include the peripheral area, the peripheral area is added to the pixel of the minimum size including PTC + and PTC- in addition to 10 pixel, 20 pixel, and 30 pixel, respectively. For the case of including, 40 pixels and 50 pixels, six models were generated and the test samples for knowing the correct answer for the PTC region were inputted to compare the accuracy. In this case, the accuracy of determining PTC + was highest when the label included 40 pixels in the surrounding area, and the accuracy of determining PTC- was highest when the label included 30 pixels in the surrounding area, and the PTC +, PTC-, The accuracy of discriminating non-PTC as a whole was highest when the label included 50 pixels of the surrounding area.

Accordingly, in order to improve the specific accuracy of PTC +, PTC- and non-PTC, when the label including the peripheral area of 30 pixels or more and 50 pixels or less, it can be confirmed that the accuracy of the trained model is the highest. In this case, in the C4d scoring method for diagnosing transplant rejection of renal grafts, it is important to accurately determine the number of PTC +. Therefore, it may be confirmed that the label is set to include 40 pixels of the surrounding area. Therefore, according to the purpose of the specific accuracy of PTC +, the specific accuracy of PTC-, or the overall accuracy from the pathological image, the size of the final peripheral region may be set to include an area of 30 pixels or more and 50 pixels or less.

Meanwhile, the embodiment of the present invention not only learns to discriminate PTC + and PTC- included in the second sample input to the second neural network, but also additionally includes PTC + / PTC- as an area for improving the accuracy of the discrimination. The second neural network is trained to identify and output three classes, including non-peritubular capillary (PTC) classes.

In this case, the non-PTC may be defined as a non-PTC tubular and a glomerulus that does not include PTC. In addition, non-PTC trains a test neural network with a part of a second sample used for learning a second neural network, and specifies a non-PTC region that is incorrectly determined as a PTC + area when the other part of the second sample is input to the test neural network. Accuracy can be improved by eliminating ambiguity.

In the case of non-PTC, as shown in FIG. 2, since the masked area may be relatively large compared to PTC + and PTC-, PTC + or PTC- is included in the masking area for specifying non-PTC. It could be wrong. Therefore, in the embodiment of the present invention, when labeling a non-PTC, the size of the masking area is labeled after the non-PTC area is labeled as a non-PTC area, which is reduced by a predetermined ratio or a predetermined pixel, compared to the area masked as the non-PTC. The accuracy of the model can be improved by using reduced and labeled non-PTC regions for training. For example, in FIG. 2, the area of the rectangle reduced in the ratio of 70% to 80% of the size of the rectangle including the masking may be labeled to specify that the area is a non-PTC area. The range of a predetermined ratio for labeling by reducing the size is not limited to this example.

Meanwhile, the neural network used in the present specification means an algorithm implemented by simulating the manner in which the human brain recognizes a pattern. Accordingly, while the first neural network and the second neural network include a plurality of layers, one layer may consist of one or more nodes. At this time, each node of the neural network may perform classification and clustering by performing an operation on input data based on weights and bias values learned through a learning algorithm. However, the configuration of the neural network can be implemented in a variety of ways, and a variety of known algorithms for implementing this, so a detailed description thereof will be omitted. Accordingly, the learning algorithms of the first neural network and the second neural network may use various algorithms such as Faster R-CNN, SDD, and Yolo, and are not limited to any one, and the layers constituting the first and second neural networks are also various. Can be implemented.

The transplant rejection determination model including the first neural network and the second neural network, which has been learned according to the above-described embodiment, may be used to diagnose the actual transplant rejection as shown in FIG. 4 below.

4 is a flowchart illustrating a process of a transplant rejection determination method according to an embodiment of the present invention. The transplant rejection determination method shown in FIG. 4 may be performed by a first neural network, a second neural network, and a processor generated according to the above-described embodiment. In this case, the first neural network and the second neural network may be operated by a computing device including a memory and a microprocessor storing the same. Referring to each step of the transplant rejection determination method shown in Figure 4 as follows.

First, the first neural network learned according to an embodiment of the present invention receives a target pathological image and outputs a plurality of detection target regions having a predetermined size (S410).

In this case, the first neural network may output a plurality of detection target regions including a tissue region of the target pathology image having a predetermined size (for example, 1024 x 1024 pixels) and having a predetermined ratio or more (for example, 50% or more) of the entire region. have.

Next, the second neural network learned according to an embodiment of the present invention receives a plurality of detection target regions and receives a PTC + region, a PTC- region, and a non-peritubular capillary (PTC) region included in each detection target region. Determine (S420).

Thereafter, the processor calculates a C4d score of the target pathological image based on the number of PTC + regions and PTC- regions determined by the second neural network (S430). For example, the C4d score can be calculated from the number of PTC + regions for the sum of the number of PTC + regions and the number of PTC- regions. At this time, the algorithm for calculating the C4d score may be calculated based on the Banff criteria scoring algorithm. Therefore, through the present embodiment, the user can check the C4d score directly from the pathological image, thereby quickly and easily diagnosing the transplant rejection.

According to the embodiment described above, since the doctor uses a sample divided into a predetermined size that can display a label on a specific area of the pathological image for learning, it is possible to improve the reproducibility and accuracy of the diagnosis.

In addition, the model is trained not only to identify PTC + regions and PTC- regions but also to distinguish non-PTC regions so that the model can better distinguish PTC + regions and PTC- regions within pathological images. You can.

In addition, the reproducibility and accuracy can be improved by determining the optimal size for labeling each of the PTC + region, PTC- region and non-PTC region included in the pathological image.

Accordingly, the embodiment of the present invention enables a faster and more accurate determination of the diagnosis of the transplant rejection, which can greatly reduce the time and labor required for the diagnosis, as well as the convenience of diagnosis to both the doctor and the patient. Can increase safety.

Embodiments of the present invention described above may be implemented through various means. For example, embodiments of the present invention may be implemented by hardware, firmware, software, or a combination thereof.

In the case of a hardware implementation, the method according to embodiments of the present invention may include one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing elements (DSPs). Digital Signal Processing Devices (DSPs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, and the like.

In the case of an implementation by firmware or software, the method according to the embodiments of the present invention may be implemented in the form of a module, procedure, or function that performs the functions or operations described above. The computer program in which the software code or the like is recorded may be stored in a computer readable recording medium or a memory unit and driven by a processor. The memory unit may be located inside or outside the processor, and may exchange data with the processor by various known means.

In addition, the combination of each block of the block diagram and each step of the flowchart attached to the present invention may be performed by computer program instructions. These computer program instructions may be embedded in an encoding processor of a general purpose computer, special purpose computer, or other programmable data processing equipment such that the instructions executed by the encoding processor of the computer or other programmable data processing equipment are not included in each block or block diagram. It will create means for performing the functions described in each step of the flowchart. These computer program instructions may be stored in a computer usable or computer readable memory that can be directed to a computer or other programmable data processing equipment to implement functionality in a particular manner, and thus the computer usable or computer readable memory. It is also possible for the instructions stored in to produce an article of manufacture containing instruction means for performing the functions described in each block or flowchart of each step of the block diagram. Computer program instructions may also be mounted on a computer or other programmable data processing equipment, such that a series of operating steps may be performed on the computer or other programmable data processing equipment to create a computer-implemented process to create a computer or other programmable data. Instructions that perform processing equipment may also provide steps for performing the functions described in each block of the block diagram and in each step of the flowchart.

In addition, each block or step may represent a portion of a module, segment or code that includes one or more executable instructions for executing a specified logical function. It should also be noted that in some alternative embodiments the functions noted in the blocks or steps may occur out of order. For example, the two blocks or steps shown in succession may in fact be executed substantially concurrently or the blocks or steps may sometimes be performed in the reverse order, depending on the functionality involved.

As such, those skilled in the art will appreciate that the present invention can be implemented in other specific forms without changing the technical spirit or essential features thereof. Therefore, the above-described embodiments are to be understood as illustrative in all respects and not as restrictive. The scope of the present invention is shown by the following claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. .

Claims (12)

In the method for generating a transplant rejection determination model performed by one or more processors,
Training a first neural network that receives a pathological image and selects and outputs a plurality of detection target regions each having a first size and having a tissue area more than a predetermined ratio with respect to the pathological image; And
Training a second neural network for determining a positive peritubular capillary (PTC +) region, a negative peritubular capillary (PTC-) region, and a non-peritubular capillary (non-PTC) region included in each detection target region; Containing
How to Create a Graft Rejection Identification Model.
The method of claim 1,
The pathological image,
An image from which information included in an area other than the tissue area is removed.
How to Create a Graft Rejection Identification Model.
The method of claim 1,
Learning the first neural network,
Receiving a learning pathology image including a tissue region;
Selecting a plurality of first samples having the first size in the learning pathology image;
Labeling each of the plurality of first samples to set whether to be selected as the detection subject region or not to select a label; And
Learning the first neural network based on the label and the plurality of first samples to determine the predetermined ratio.
How to Create a Graft Rejection Identification Model.
The method of claim 1,
Learning the second neural network,
Receiving masking information on the plurality of second samples having the first size and the PTC +, PTC-, and non-PTCs included in the plurality of second samples;
Specifying PTC +, PTC-, non-PTC in the plurality of second samples based on the masking information;
Labeling the specified PTC + and PTC- areas with a final peripheral region and labeling the PTC + region and the PTC- region, respectively, and labeling the specified non-PTC with a non-PTC region; And
Training the second neural network based on the label and the plurality of second samples.
How to Create a Graft Rejection Identification Model.
The method of claim 4, wherein
Setting the label,
Including the first peripheral region in the specified PTC +, PTC- to generate a first test neural network based on the first labeling information labeled with the first PTC + region and the first PTC- region, respectively, and the specified PTC +, PTC -Including a second peripheral region of a different size from the first peripheral region to generate a second test neural network based on second labeling information labeled with the second PTC + region, the second PTC- region, respectively;
Receiving a test sample in which a PTC + region, a PCT- region, and a non-PTC region are respectively defined;
Inputting the test sample into the first test neural network and the second test neural network, respectively, to determine PTC +, PTC-, and non-PTC regions within the test sample; And
The comparison between the first test neural network and the PTC + region, the PTC- region, and the non-PTC region in the determined test sample and the PTC + region, the PTC- region, and the non-PTC region in the determined test sample is performed. Determining the size of the peripheral region used for the higher accuracy neural network among the second test neural networks as the final peripheral region.
How to create a model for determining transplant rejection.
The method of claim 4, wherein
Setting the label,
Labeling the non-PTC region by narrowing the region reduced by a predetermined ratio than the specified non-PTC region to set a label.
How to Create a Graft Rejection Identification Model.
The method of claim 4, wherein
The size of the final peripheral area,
30 pixels or more and 50 pixels or less
How to Create a Graft Rejection Identification Model.
The method of claim 4, wherein
The non-PTC,
At least one of a location, tubular and glomerulus that was incorrectly identified as a PTC + region when a test neural network was generated through some of the plurality of second samples and entered into the test neural network as the remaining portion of the plurality of second samples Containing one
How to Create a Graft Rejection Identification Model.
An apparatus comprising a processor for performing the method of any one of claims 1 to 8.
A computer-readable recording medium having recorded thereon a computer program comprising instructions for causing a processor to perform the method of any one of claims 1 to 8.
A computer program stored in a computer readable recording medium for causing a processor to perform the method of any one of claims 1 to 8.
In the method for determining a transplant rejection response performed by the first neural network and the second neural network learned according to any one of claims 1 to 8,
Receiving, by the first neural network, a target pathological image and outputting a plurality of detection target regions having a predetermined size;
The second neural network receives the plurality of detection target regions output by the first neural network, and includes a PTC + (positive peritubular capillary) region, a PTC- (negative peritubular capillary) region, and a non-PTC (non-PTC) region within the plurality of detection target regions. determining a peritubular capillary) area; And
Calculating a C4d score of the target pathological image based on the determined PTC + region and the number of PTC- regions.
How to determine transplant rejection.

Images