KR20220094700A - Masking method and system for building medical image data for artificial intelligence learning - Google Patents

Abstract

The present invention relates to a masking method for building medical image data for artificial intelligence learning. The masking method for building medical image data for artificial intelligence learning, includes the following steps of: acquiring a plurality of medical images; preprocessing the plurality of medical images such that the images can be applied to an artificial intelligence algorithm; selecting a medical image to be masked from among the plurality of preprocessed medical images; recommending a position of a predicted region of interest by applying the selected medical image to the artificial intelligence algorithm; masking the medical image considering the selected medical image and the recommended position information of the region of interest, and generating the completely masked medical image as medical image data for learning; and performing learning and prediction by applying the medical image data for learning to a deep learning model. Therefore, the present invention is capable of improving the accuracy of lesion detection.

Description

MASKING METHOD AND SYSTEM FOR BUILDING MEDICAL IMAGE DATA FOR ARTIFICIAL INTELLIGENCE LEARNING}

The present application relates to a masking method and system for constructing medical image data for artificial intelligence learning.

Cells, the smallest unit constituting the human body, divide and grow, die and disappear due to intracellular regulatory functions when normal, maintaining cell number balance. If a cell is damaged for some reason, it recovers through treatment and functions as a normal cell, but if there is no recovery, it dies on its own. However, for various reasons, abnormal cells whose proliferation and inhibition are not controlled not only proliferate excessively, but also invade surrounding tissues and organs, resulting in mass formation and destruction of normal tissues, which is defined as cancer. Cancer is the proliferation of cells that cannot be suppressed, and it destroys the structure and function of normal cells and organs, so the diagnosis and treatment are very important.

In the case of neoplasms in the existing stomach, the doctor first judged whether it was gastric cancer through a gastroscopy, taking into account the shape and size of the stomach included in the endoscopic image, and confirmed the diagnosis by biopsy. However, in this method, different doctors can make different diagnoses due to different experiences, and there is a problem that an accurate diagnosis cannot be made in an area where there is no doctor.

Recently, as the computer vision recognition problem is applied to various fields using various deep learning models, the importance of constructing an individual neural network (Customizing Model) for each field is increasing. Therefore, it is essential for developers and researchers to perform an annotation process to compose the learning data of the existing machine learning method, and the required labor and time may increase according to the amount of learning data.

The background technology of the present application is disclosed in Korean Patent Application Laid-Open No. 10-2020-0010661.

The present application is intended to solve the problems of the prior art described above, and it is possible to form masking data for RoI (Region of Intrerest) or lesion, which is important for all types of image images that can be imaged in two dimensions. The purpose of the present invention is to provide a masking method and system for constructing medical image data for artificial intelligence learning.

However, the technical problems to be achieved by the embodiments of the present application are not limited to the technical problems described above, and other technical problems may exist.

As a technical means for achieving the above technical task, the masking method for constructing medical image data for artificial intelligence learning according to an embodiment of the present application includes: acquiring a plurality of medical image images; Pre-processing to be applicable to an intelligent algorithm, selecting a medical image to be masked from among a plurality of pre-processed medical image images, applying the selected medical image to the artificial intelligence algorithm to recommend a location of an expected ROI performing the masking in consideration of the selected medical image image and location information of the recommended region of interest, and generating the masked medical image image as medical image data for training, and applying the medical image data for training to a deep learning model It may include the step of performing learning and prediction by applying.

In addition, the masking method for constructing medical image data for AI learning includes the steps of providing a plurality of pre-processed medical image images to a user terminal and user input information on location information of an expected region of interest to be masked on the selected medical image image. The method may further include receiving, wherein the generating of the medical image data for learning includes: when the user input information is received, determining the user input information as location information of the region of interest to generate the medical image data for training. can

Also, in the step of recommending the location of the predicted ROI, the location of the predicted ROI including the predicted lesion region may be recommended using a segmentation algorithm or an object detection algorithm.

In addition, in the masking method for constructing medical image data for artificial intelligence learning, before performing the learning and prediction, in consideration of the purpose of using the medical image data for learning, any one of a plurality of deep learning models is used with the medical image data for training. It may further include the step of performing data transformation into a form applicable to one.

In addition, the performing of the data conversion includes generating a crop image using the medical image data for training in consideration of the purpose of using the medical image data for training, generating meta data, and a binary map. map), any one of data transformations can be performed.

According to an embodiment of the present application, a masking device for constructing medical image data for artificial intelligence learning includes an acquisition unit for acquiring a plurality of medical image images, and a pre-processing unit for preprocessing the plurality of medical image images to be applicable to an artificial intelligence algorithm. , a recommendation unit that selects a medical image to be masked from among a plurality of preprocessed medical image images, and recommends a location of an expected ROI by applying the selected medical image to an artificial intelligence algorithm, the selected medical image and the recommended medical image. A generator that performs the masking in consideration of location information of the region of interest and generates a masked medical image as medical image data for training, and deep learning that applies the medical image data for training to a deep learning model to perform learning and prediction may include wealth.

In addition, the masking apparatus for constructing medical image data for artificial intelligence learning includes a providing unit that provides a plurality of pre-processed medical image images to a user terminal and a user input for location information of an expected region of interest to be masked on the selected medical image image. The method may further include a receiving unit configured to receive information, wherein, when the user input information is received, the generating unit may determine the user input information as location information of the ROI to generate the training medical image data.

Also, the recommendation unit may recommend the location of the predicted ROI including the predicted lesion region using a segmentation algorithm or an object detection algorithm.

In addition, the masking device for constructing medical image data for artificial intelligence learning performs data conversion into a form applicable to any one of a plurality of deep learning models by converting the medical image data for learning in consideration of the purpose of using the medical image data for learning. It may further include a converter.

In addition, the conversion unit is configured to generate a crop image using the medical image data for training in consideration of the purpose of using the medical image data for training, generate meta data, and generate a binary map. One data transformation can be performed.

According to an embodiment of the present application, a masking system for constructing medical image data for artificial intelligence learning acquires a plurality of medical image images, performs pre-processing to apply the plurality of medical image images to an artificial intelligence algorithm, and pre-processing Selects a medical image to be masked from among a plurality of medical image images, applies the selected medical image to an artificial intelligence algorithm to recommend a location of an expected ROI, and selects a location of the selected medical image and the recommended ROI A masking device that performs the masking in consideration of the information and generates a masked medical image as medical image data for training and a plurality of pre-processed medical image images are provided, and the masking is performed on the selected medical image. A user terminal providing user input information for location information, wherein the masking device may apply the training medical image data to a deep learning model to perform learning and prediction.

The above-described problem solving means are merely exemplary, and should not be construed as limiting the present application. In addition to the exemplary embodiments described above, additional embodiments may exist in the drawings and detailed description.

According to the above-described problem solving means of the present application, it is possible to form masking data for an important RoI (Region of Intrerest) or lesion for all types of image images that can be two-dimensionally imaged.

According to the above-described problem solving means of the present application, by forming masking data for important RoI (Region of Intrerest) or lesion and performing deep learning learning and prediction based on this, time required for lesion detection can be reduced and the accuracy of lesion detection can be improved.

However, the effects obtainable herein are not limited to the above-described effects, and other effects may exist.

1 is a schematic configuration diagram of a masking system for constructing medical image data for AI learning according to an embodiment of the present application.
2 is a schematic block diagram of a masking apparatus for constructing medical image data for artificial intelligence learning according to an embodiment of the present application.
3 is a diagram illustrating an interface provided to a user terminal by a masking apparatus for constructing medical image data for artificial intelligence learning according to an embodiment of the present application.
4 is an operation flowchart of a masking method for constructing medical image data for AI learning according to an embodiment of the present application.

Hereinafter, embodiments of the present application will be described in detail with reference to the accompanying drawings so that those of ordinary skill in the art to which the present application pertains can easily carry out. However, the present application may be implemented in several different forms and is not limited to the embodiments described herein. And in order to clearly explain the present application in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout this specification, when a part is "connected" with another part, it is not only "directly connected" but also "electrically connected" or "indirectly connected" with another element interposed therebetween. "Including cases where

Throughout this specification, when a member is positioned “on”, “on”, “on”, “on”, “under”, “under”, or “under” another member, this means that a member is positioned on the other member. It includes not only the case where they are in contact, but also the case where another member exists between two members.

Throughout this specification, when a part "includes" a certain component, it means that other components may be further included, rather than excluding other components, unless otherwise stated.

1 is a schematic configuration diagram of a masking system for constructing medical image data for AI learning according to an embodiment of the present application. Hereinafter, for convenience of explanation, a masking system for constructing medical image data for AI learning will be referred to as a masking system 1 . In addition, a masking device for constructing medical image data for AI learning will be referred to as a masking device 10 .

Referring to FIG. 1 , a masking system 1 may include a masking device 10 and a user terminal 20 .

According to an embodiment of the present disclosure, the masking apparatus 10 may acquire a plurality of medical image images and perform pre-processing to apply the plurality of medical image images to an artificial intelligence algorithm. Also, the masking apparatus 10 may select a medical image to be masked from among a plurality of pre-processed medical image images. Also, the masking apparatus 10 may apply the selected medical image image to an artificial intelligence algorithm to recommend the location of the expected ROI, and perform masking in consideration of the selected medical image image and location information of the recommended ROI. Also, the masking apparatus 10 may generate the masked medical image image as medical image data for training. Also, the masking apparatus 10 may apply the medical image data for training to the deep learning model to perform learning and prediction.

According to an embodiment of the present disclosure, the user terminal 20 may receive a plurality of pre-processed medical image images from the masking apparatus 10 . The user terminal 20 may output data provided from the masking device 10 through a display. Also, the user terminal 20 may provide the masking apparatus 10 with user input information on location information of an expected ROI to be masked on the selected medical image image.

According to an embodiment of the present disclosure, the masking apparatus 10 may provide a masking creation menu for a Region of Intrerest (RoI) or a lesion to the user terminal 20 . For example, the user terminal 20 downloads and installs an application program provided by the masking device 10, and a masking creation menu for Region of Intrerest (RoI) or Lesion can be provided through the installed application. have.

The masking apparatus 10 may include all kinds of servers, terminals, or devices that transmit and receive data, content, and various communication signals to and from the user terminal 20 through a network, and have functions of storing and processing data.

The user terminal 20 is a device that interworks with the masking apparatus 10 through a network, for example, a smart phone, a smart pad, a tablet PC, a wearable device, etc. PCS (Personal Communication System), GSM (Global System for Mobile communication), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W - It may be all kinds of wireless communication devices such as CDMA (W-Code Division Multiple Access) and Wibro (Wireless Broadband Internet) terminals and fixed terminals such as desktop computers and smart TVs.

An example of a network for sharing information between the masking device 10 and the user terminal 20 is a 3rd Generation Partnership Project (3GPP) network, a Long Term Evolution (LTE) network, a 5G network, and a World Interoperability for Microwave Access (WIMAX) network. , Wired and Wireless Internet, LAN (Local Area Network), Wireless LAN (Wireless Local Area Network), WAN (Wide Area Network), PAN (Personal Area Network), Bluetooth (Bluetooth) network, Wifi network, NFC (Near Field) Communication) network, satellite broadcasting network, analog broadcasting network, DMB (Digital Multimedia Broadcasting) network, etc. may be included, but is not limited thereto.

2 is a schematic block diagram of a masking apparatus for constructing medical image data for artificial intelligence learning according to an embodiment of the present application.

Referring to FIG. 2 , the masking device 10 includes an acquisition unit 11 , a preprocessor 12 , a recommendation unit 13 , a generation unit 14 , a provision unit 15 , a reception unit 16 , and a conversion unit ( 17), a deep learning unit 18 and a database 19 may be included. However, the configuration of the masking device 10 is not limited thereto.

According to an embodiment of the present application, the acquisition unit 11 may acquire a plurality of medical image images. For example, the plurality of medical image image data may include medical ultrasound (Ultrasound), computed tomography (CT), nuclear medicine image, positron emission tomography (PET-CT), X-ray, magnetic resonance imaging (MRI), breast It may include radiology images such as mammography. The format of the image includes jpg, jpeg, jpe, jfif, bmp, gif, heic, dib, tif, tiff, including the medical image data standard, Digital Imaging and Communications in Medicine (DICOM). All computer image compression formats such as , svs, etc. can be input.

Also, the acquirer 11 may upload a plurality of acquired medical image images to the system. The acquisition unit 11 may upload to the system using the HL7 Fast Healthcare Interoperability Resources (FHIR) standard technology, but is not limited thereto.

According to an embodiment of the present application, the preprocessor 12 may preprocess a plurality of medical image images to be applicable to an artificial intelligence algorithm. The preprocessor 12 may convert (preprocess) the plurality of medical image image data provided by the acquirer 11 into an image form finally determined to perform deep learning learning and prediction. In addition, the preprocessor 12 anonymizes and encrypts medical records and medical information included in Digital Imaging and Communications in Medicine (DICOM) to protect the privacy of the user (privacy preserving). have.

Also, the preprocessor 12 may register important labels and additional information of files related to the extracted medical image to the database 19 . For example, the preprocessor 12 may store information of the file, that is, a label to be predicted (missing values if not present), and other necessary information in a json format. Here, the missing value may mean that there is no value due to a missing value, an empty value, or a measurement error. However, the present invention is not limited thereto, and may be stored in a relational database format other than the Json format.

Here, JSON is an open standard format that uses human-readable text to convey data objects consisting of attribute-value pairs or “key-value pairs”. A relational database is a computerized information database with a very simple principle in which simple relationships between keys and values are tabulated.

For example, the preprocessor 12 crops, rotates, flips, rescales, zooms, and zeros a plurality of medical image images to be applicable to an artificial intelligence algorithm. Zero filling, rolling, color jittering, noise addition, contrast adjustment, brightness adjustment (gamma conversion), smoothing (averaging filter), scaling, inversion (left/right/up/down), Shift (horizontal/vertical), partial mask (Cut out and Random Erasing), trim (Random Crop), transform, discolor, background replacement, Gaussian transformation, shear, affine and combinations thereof Pretreatment may be performed by selecting from the group consisting of, but is not limited thereto.

According to an embodiment of the present disclosure, the recommendation unit 13 may select a medical image to be masked from among a plurality of pre-processed medical image images. Also, the recommendation unit 13 may recommend the location of the expected ROI by applying the selected medical image to the artificial intelligence algorithm. In other words, the recommendation unit 13 may recommend the location of the predicted ROI including the predicted lesion region by applying an artificial intelligence algorithm. The predicted ROI may be a two-dimensional definition of a boundary of an object (eg, a lesion) in a rectangular shape.

As an example, the artificial intelligence algorithm may include a segmentation algorithm or an object detection algorithm. However, the present invention is not limited thereto, and the recommender 13 may recommend the location of the predicted ROI by performing various image processing, for example, a Gaussian filter, an affine transformation, and contrast adjustment.

For example, referring to area A of FIG. 3 , the recommendation unit 13 may select a first image (CT image) as a medical image image to be masked. The recommendation unit 13 may recommend the location of the predicted ROI by applying the selected first image (CT image) to the artificial intelligence algorithm. Here, the predicted ROI may include a predicted lesion region. That is, the recommendation unit 13 may recommend the location of the lesion by displaying the expected ROI in a rectangular shape at a location predicted to be a lesion in the selected medical image. In FIG. 3 , only one location of the predicted ROI is displayed (recommended), but the present invention is not limited thereto, and a plurality of locations may be displayed (recommended). That is, when there are a plurality of predicted lesions, the recommender 13 may recommend a plurality of predicted ROIs corresponding thereto.

In addition, the recommender 13 may provide the location of the predicted ROI through the providing unit 15 , that is, location information on which annotation is performed, to the user terminal 20 . The user may input correction and additional selection information related to the location of the predicted ROI through the user terminal 20 . The user terminal 20 may provide the user's input information through the receiving unit 16 , and the receiving unit 16 may provide the user's input information to the recommending unit 13 when the user's input information is received. The recommender 13 may change the location of the predicted ROI in consideration of the user's input information. For example, the recommender 13 may recommend the first location in the selected medical image as the location of the expected ROI. On the other hand, when the second location in the medical image selected by the user terminal 20 is provided as location information of the predicted ROI, the recommender 13 may change the second location information to the location of the predicted ROI. The recommender 13 may provide a more accurate ROI 131 to the generator 14 by correcting the location of the predicted ROI by reflecting the user's input information.

According to an embodiment of the present disclosure, the generator 14 may perform masking in consideration of the selected medical image image and the recommended location information of the ROI 131 . In other words, the generator 14 may perform masking on the selected medical image based on the selected medical image and location information of the ROI recommended by the recommender 13 . In this case, the recommended location information of the ROI may be location information of the ROI recommended by the recommender 13 . On the other hand, the location information of the recommended region of interest may be location information of the region of interest changed in consideration of user input information.

In addition, the generator 14 may generate medical image data for learning from the masked medical image image. For example, the medical image data for training may include information related to a selected medical image image, location information of a recommended ROI (eg, x and y coordinate locations, horizontal length, vertical length), and the like. The generator 14 may store the medical image data for training in the database 19 . As an example, individual images may be automatically registered in the database 19 when annotation is performed. series) can be saved.

According to an embodiment of the present application, the providing unit 15 may provide a plurality of pre-processed image images to the user terminal 20 . The providing unit 15 may provide a plurality of pre-processed image images including a user interface (UI). The user may select a medical image to be masked from among a plurality of preprocessed image images displayed on the user terminal 20 .

According to an embodiment of the present application, the receiver 16 may receive medical semantic selection information to be masked from the user terminal 20 . Also, the receiver 16 may receive user input information on location information of an expected ROI to be masked on the selected medical image. In this case, when the user input information is received from the receiver 16 , the generator 14 may determine the user input information as location information of the ROI to generate image data for learning. In other words, the generator 14 may generate medical image data for training by determining the predicted ROI recommended by the recommender 13 as location information of the ROI to be masked. On the other hand, the generator 14 may generate the medical image data for learning by determining the user input information as location information of the region of interest to be masked.

According to an embodiment of the present application, the transforming unit 17 may perform data transformation on the medical image data for training into a form applicable to any one of a plurality of deep learning models in consideration of the purpose of using the medical image data for training. In addition, in consideration of the purpose of use of the medical image data for training, the converter 17 generates a crop image using the medical image data for learning, generates meta data, or generates a binary map. One data transformation can be performed.

For example, the transform unit 17 may perform image cropping if it is a form to be applied to a deep learning learning model through specific region information. In addition, if it is for object detection, the transformation unit 17 may perform data transformation for generating metadata for storing image coordinates in XML or Json format. Also, if the purpose of use is to generate a segmentation map, the conversion unit 17 may perform data conversion to generate a binary map corresponding thereto.

The conversion unit 17 may convert the converted data into a form that can be used as an input of a deep learning model (deep learning framework) and store it. The conversion unit 17 may generate information that can be stored in different folders according to the label or accessed through a file name in the form of metadata in a data frame in the form of a file name. For example, deep learning frameworks include tensorflow, pytorch, keras, caffe, Sonnet, MXNet, ONN, Gluon, DL4J, Chainer, tensorflow.js, etc. may be included.

According to an embodiment of the present application, the deep learning unit 18 may apply the medical image data for training to the deep learning model to perform learning and prediction. Here, learning and prediction may mean lesion determination based on a medical image,

According to another embodiment of the present application, the deep learning unit 18 may diagnose whether the thyroid cancer has metastasized to the lymph nodes by applying the medical image data for training to the deep learning model.

First, in order to diagnose whether the thyroid cancer has metastasized to the lymph node, the converter 17 may generate a crop image to diagnose whether the thyroid cancer has metastasized to the lymph node. The conversion unit 17 may augment the cropped image. Since the training data has a higher number of benign lymph nodes than the metastatic lymph node image, the data can be augmented so that the ratio of the two labels is similar, preventing overfitting of the model and improving the model performance. That is, by augmenting the cropped image, the amount of training images is effectively increased based on a limited training data set.

The transform unit 17 crops, rotates, flips, rescales, zooms, zero filling, rolling, etc. to enhance the image. Color jittering, Noise Addition, Contrast Adjustment, Brightness Adjustment (Gamma Conversion), Smoothing (Averaging Filter), Scaling, Invert (Left/Left/Up/Down), Shift (Horizontal/Vertical), Partial Mask (Cut) out and Random Erasing), trimming (Random Crop), transformation, discoloration, background replacement, Gaussian transformation, shear, affine, and combinations thereof However, the present invention is not limited thereto.

For example, the transform unit 17 randomly sets the image angle within about ±20º, randomly changes the size of the region around the lymph node by about 20%, and randomly converts the image horizontally and vertically while including the lymph node itself. Inverting may be performed, but is not limited thereto.

In addition, the deep learning unit 18 may generate learning information by learning the augmented image using a convolutional neural network (CNN). For example, the convolutional neural network may be a model selected from the group consisting of VGG16, VGG19, Xception, InceptionV3, InceptionResNetV2, DenseNet121, DenseNet169, ResNet, EfficientNet, MobileNet, and combinations thereof, but is not limited thereto. .

Also, the deep learning unit 18 may diagnose whether the thyroid cancer has metastasized to the lymph nodes based on the learning information. Also, the deep learning unit 18 may generate a heat map. The heatmap can show where the CNN model is focused to differentiate between benign and metastatic lymph nodes. This suggests that the abnormal characteristics of malignant tumors were learned by CNN and used as the basis for the diagnosis of lymph node metastasis of thyroid cancer. The generated heat map is superimposed on the original CT image image to compare the actual lymph node location and the area used by the CNN model for metastasis diagnosis. And there is an advantage of further increasing the sensitivity.

In the case of conventional CT images, lymph node metastasis and remaining tissue regions are generally affected by noise artifacts and overlap with other regions. Although there was a limitation in that it was difficult to directly use the deep learning algorithm, the masking method for constructing medical image data for artificial intelligence learning according to the present application extracts the lesion area of the cervical lymph node from the CT image and learns the crop image for the extracted lesion area. It is possible to provide a method for diagnosing lymph node metastasis of thyroid cancer based on deep learning with high sensitivity, specificity and accuracy.

In the case of the conventional diagnosis method using deep learning, it was difficult to explain the internal relationship between the input value and the result predicted by deep learning, so it was difficult to evaluate the diagnostic accuracy of deep learning. of thyroid cancer lymph node metastasis diagnosis method generates and provides a heat map to infer which part of the CT image the deep learning model is focusing on. method can be provided.

3 is a diagram illustrating an interface provided to a user terminal by a masking apparatus for constructing medical image data for artificial intelligence learning according to an embodiment of the present application.

Referring to FIG. 3 , the interface provided to the user terminal by the masking device for constructing medical image data for artificial intelligence learning, that is, the masking interface, includes an image and a masking canvas area (A), a help area (B), and calling It may include a wrong area (C), a history area (D), a current image masking information confirmation area (E), and a list area (F). However, the present invention is not limited thereto, and according to the exemplary embodiment of the present application, only some of the above-described plurality of items may be displayed or other types of additional items may be displayed.

For example, the image and masking canvas area A may be an area provided so that a user can draw a desired rectangle masking on the image through left-click of the mouse. The rectangular masking may be the region of interest 141 . Two or more rectangles may be formed in the image and masking canvas area A. The masking apparatus 10 may delete the most recently drawn rectangle maskin when mouse right-click information is received as user input information.

In addition, information related to this program, that is, a method that a user should perform in order to construct medical image data for AI learning, may be displayed in the help area B. Also, the call area C may include a call window for calling data desired by the user. In other words, the masking device 10 calls data for constructing medical image data for artificial intelligence learning from the image stored in the database 19 based on user input information input to the call window included in the call area C can do. Also, in the history area D, information related to the name of the recently modified data and the name of the data being modified may be displayed.

In addition, the current image masking information confirmation area E may display information on all rectangle masking generated by left-clicking on the masking canvas. The masking device 10 may Multiple rectangle masking data information can be stored through the x, y coordinate position, horizontal length, and vertical length of the upper left In addition, the list area (F) is used to output list information that calls the recently modified file. It can be an area.

By providing a masking web tool for constructing medical image data for artificial intelligence learning using a user interface in the masking device 10, the user can more conveniently generate masking data for an important region of interest and lesion.

Through this process, the masking apparatus 10 may form masking data for RoI or Lesion, which is important for all types of image images (Ultrasound, CT, MRI, X-ray) that can be imaged in 2D.

Hereinafter, an operation flow of the present application will be briefly reviewed based on the details described above.

4 is an operation flowchart of a masking method for constructing medical image data for AI learning according to an embodiment of the present application.

The masking method for constructing medical image data for artificial intelligence learning shown in FIG. 4 may be performed by the masking apparatus 10 described above. Therefore, even if omitted below, the description of the masking apparatus 10 may be equally applied to the description of the masking method for constructing medical image data for AI learning.

In operation S110 , the masking apparatus 10 may acquire a plurality of medical image images.

In operation S120, the masking apparatus 10 may pre-process the plurality of medical image images to be applicable to the artificial intelligence algorithm.

In operation S130 , the masking apparatus 10 may select a medical image image to be masked from among a plurality of preprocessed medical image images.

In operation S140 , the masking apparatus 10 may recommend the location of the predicted ROI by applying the selected medical image to an artificial intelligence algorithm.

In operation S150 , the masking apparatus 10 may perform masking in consideration of the selected medical image image and location information of the recommended ROI, and generate the masked medical image as medical image data for training.

In step S160, the masking apparatus 10 may apply the medical image data for training to the deep learning model to perform learning and prediction.

In the above description, steps S110 to S160 may be further divided into additional steps or combined into fewer steps, according to an embodiment of the present application. In addition, some steps may be omitted if necessary, and the order between steps may be changed.

The masking method for constructing medical image data for AI learning according to an embodiment of the present application may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

In addition, the above-described masking method for constructing medical image data for artificial intelligence learning may be implemented in the form of a computer program or application executed by a computer stored in a recording medium.

The foregoing description of the present application is for illustration, and those of ordinary skill in the art to which the present application pertains will understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present application. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may be implemented in a combined form.

The scope of the present application is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present application.

1: Masking system
10: masking device
11: Acquisition unit 12: Pre-processing unit
13: recommendation section 14: generator section
15: providing unit 16: receiving unit
17: transformation unit 18: deep learning unit
19: Database
20: user terminal

Claims (12)

acquiring a plurality of medical imaging images;
pre-processing the plurality of medical image images to be applicable to an artificial intelligence algorithm;
selecting a medical image to be masked from among a plurality of preprocessed medical image images;
applying the selected medical image to the artificial intelligence algorithm to recommend a location of an expected ROI;
performing the masking in consideration of the selected medical image image and location information of the recommended region of interest, and generating the masked medical image data as medical image data for training; and
performing learning and prediction by applying the medical image data for training to a deep learning model;
Including, a masking method.
According to claim 1,
providing a plurality of preprocessed medical image images to a user terminal; and
receiving user input information on location information of an expected region of interest to be masked on the selected medical image;
further comprising,
The step of generating the medical image data for learning includes:
When the user input information is received, the masking method of generating the medical image data for learning by determining the user input information as location information of the ROI.
According to claim 1,
The step of recommending the location of the predicted ROI includes:
The masking method of claim 1, wherein the location of the predicted ROI including the predicted lesion region is recommended using a segmentation algorithm or an object detection algorithm.
According to claim 1,
Before performing the learning and prediction,
In consideration of the purpose of using the medical image data for training, the method further comprising the step of performing data transformation of the medical image data for training into a form applicable to any one of a plurality of deep learning models, the masking method.
5. The method of claim 4,
The data conversion step includes:
In consideration of the purpose of using the medical image data for training, any one of data conversion among generating a crop image, generating meta data, and generating a binary map is performed using the medical image data for training What to do, the masking method.
an acquisition unit configured to acquire a plurality of medical image images;
a pre-processing unit for pre-processing the plurality of medical image images to be applicable to an artificial intelligence algorithm;
a recommendation unit that selects a medical image to be masked from among a plurality of preprocessed medical image images, and recommends a location of an expected ROI by applying the selected medical image to an artificial intelligence algorithm;
a generator for performing the masking in consideration of the selected medical image image and location information of the recommended region of interest, and generating the masked medical image as medical image data for training; and
A deep learning unit that performs learning and prediction by applying the medical image data for learning to a deep learning model;
Including, a masking device.
7. The method of claim 6,
a provider providing a plurality of pre-processed medical image images to a user terminal; and
a receiver for receiving user input information on location information of an expected region of interest to be masked on the selected medical image;
further comprising,
The generating unit,
When the user input information is received, determining the user input information as location information of the region of interest to generate the training medical image data.
7. The method of claim 6,
The recommendation is
The masking apparatus of claim 1, wherein the location of the predicted ROI including the predicted lesion region is recommended using a segmentation algorithm or an object detection algorithm.
7. The method of claim 6,
In consideration of the purpose of using the medical image data for training, the masking device further comprising: a converter configured to convert the medical image data for training into a form applicable to any one of a plurality of deep learning models.
10. The method of claim 9,
The conversion unit,
In consideration of the purpose of using the medical image data for training, any one of data conversion among generating a crop image, generating meta data, and generating a binary map is performed using the medical image data for training What to do is a masking device.
Obtaining a plurality of medical image images, performing pre-processing to apply the plurality of medical image images to an artificial intelligence algorithm, selecting a medical image to be masked from among the plurality of pre-processed medical image images, and selecting the selected medical image The image is applied to an artificial intelligence algorithm to recommend the location of the expected ROI, and the masking is performed in consideration of the selected medical image image and location information of the recommended ROI, and the masked medical image is used as medical image data for learning. generating masking devices; and
a user terminal that receives a plurality of pre-processed medical image images and provides user input information on location information of an expected region of interest to be masked on the selected medical image image;
including,
The masking device,
A masking system that performs learning and prediction by applying the medical image data for training to a deep learning model.
A computer-readable recording medium in which a program for executing the method of any one of claims 1 to 5 on a computer is recorded.

Images