WO2019172621A1 - Disease prediction method and disease prediction device using same - Google Patents

Abstract

The present invention provides a disease prediction method and a device using same, the method comprising the steps of: receiving a medical image of a target part; determining a lesion area in the medical image by using a disease prediction model learned to determine, for a plurality of predetermined diseases, lesion areas for a plurality of diseases in medical images; and determining whether the subject is attacked with one disease among the plurality of diseases, on the basis of the determined lesion area.

Description

Disease prediction method and disease prediction device using same

The present invention relates to a disease prediction method and a disease prediction device using the same, and more particularly, to a method and device for predicting a disease based on a diagnostic medical image of a target site.

The medical imaging apparatus is a device for acquiring an internal structure of an object as an image. The medical imaging apparatus is a non-invasive inspection apparatus that is performed without suffering to the human body, and photographs and processes structural details, internal tissues, and fluid flow in the body and shows them to the medical personnel. Medical personnel may diagnose a medical condition and a disease of a patient by using the medical image output from the medical imaging apparatus.

Medical imaging devices include magnetic resonance imaging (MRI) devices, computed tomography (CT) devices, X-ray (X-ray) devices, and ultrasound diagnostic devices to provide magnetic resonance imaging. Etc.

More specifically, the magnetic resonance imaging apparatus is a device for photographing a subject by using a magnetic field, and is widely used for accurate disease diagnosis because bones, disks, joints, nerve ligaments, and the like are shown in three dimensions at a desired angle. The magnetic resonance imaging apparatus acquires a magnetic resonance (MR) signal using a high frequency multi coil, a permanent magnet, a gradient coil, and the like including RF coils. The magnetic resonance signal is sampled to obtain a magnetic resonance image.

Among medical imaging apparatuses, a computed tomography apparatus may provide a cross-sectional image of an object, and have an advantage that internal structures of the object, such as kidneys and lungs, may not be overlapped with each other in comparison with a general x-ray apparatus. . Accordingly, the computed tomography apparatus is widely used for precise diagnosis of a disease. Such a computed tomography apparatus irradiates an X-ray to an object and detects an X-ray passing through the object. Then, an image is acquired using the detected X-rays.

Meanwhile, within the medical image acquired from the medical imaging apparatus, different diseases may appear in a similar form. For this reason, in diagnosis based on medical imaging, medical personnel have difficulty in accurately diagnosing specific diseases. In addition, different treatments may be applied depending on the disease, and thus, diseases that appear in a similar form may have a high risk of misdiagnosis and side effects when the diagnosis is performed based on a medical image.

As the accuracy of diagnosis is required for the prevention of medical accidents and the improvement of medical services, there is a continuous need for the development of new diagnostic methods that can accurately diagnose certain diseases with similar shapes in medical images. It's happening.

The background art of the invention has been created to facilitate understanding of the present invention. It should not be understood that the matters described in the background of the invention exist as prior art.

Meanwhile, in order to solve a problem of the conventional diagnosis system using a medical image, an additional diagnostic test such as a cell test or a histological test may be performed. However, this may not only cause the patient and the medical staff to be troublesome, but may not be able to obtain an accurate diagnosis result by additional examination according to the disease to be diagnosed, and thus may re-perform the medical image diagnosis.

For example, among infectious spondylitis, tuberculosis spondylitis and purulent spondylitis, to which different treatments are applied, are a disease that can be diagnosed by a medical image, but are not easily distinguishable within the medical image. In particular, the diagnosis of spondylitis in the early stages, or diagnosis using a low performance medical imaging apparatus, may not provide a significant difference that can distinguish the two diseases in the medical image, and thus the accuracy of diagnosis may be reduced. Accordingly, the medical staff additionally performs a biopsy. However, the accuracy of histological examination is also very low, so accurate diagnosis of tuberculous spondylitis and purulent spondylitis is very difficult. Eventually, the medical staff re-diagnoses the diagnosis based on the medical image to diagnose spondylitis.

Meanwhile, the inventors of the present invention may use a predictive model learned by medical image data to solve problems caused by a diagnosis using a conventional medical image, in particular, a diagnosis process of a disease having a similar form in the medical image. Could be perceived.

As a result, the inventors of the present invention determine a lesion area present in a medical image for a target site, calculate a probability of a plurality of diseases for the determined area, respectively, and use a predictive model configured to predict the disease. We have come to develop a prediction method.

More specifically, the inventors of the present invention learn disease prediction based on an image in which lesion areas associated with a plurality of predetermined diseases in a medical image are masked, and calculate a probability of the disease for the lesion area and predict the disease. By using the model, it was confirmed that each of the plurality of diseases was classified and predicted with high accuracy.

In particular, the inventors of the present invention, the disease prediction model is configured to calculate the probability of disease occurrence for each of a plurality of medical images for one target site and calculate the final probability score to predict whether the disease onset, within the medical image The accuracy of diagnosis was improved for each of the similarly manifested diseases.

Accordingly, an object of the present invention is to determine a lesion region in a received medical image by using a disease prediction model trained to determine lesion regions for each of a plurality of diseases, and based on this, among the plurality of diseases. It is to provide a disease prediction method and a disease prediction device using the same, which can determine whether one disease occurs.

More specifically, the problem to be solved by the present invention, by using a disease prediction model to determine the lesion area for each of a plurality of medical images for one target site, and to determine the probability of occurrence of each of the plurality of diseases for these areas The present invention provides a disease prediction method and a disease prediction device using the same, by calculating whether or not the disease is caused by one of the plurality of diseases, thereby increasing the accuracy of diagnosis of the disease regardless of the skill of the medical professional.

Another object of the present invention is to provide a disease prediction method and a disease prediction device using the prediction model that can provide a probability of a predetermined disease for a selected region in a medical image.

The objects of the present invention are not limited to the above-mentioned objects, and other objects that are not mentioned will be clearly understood by those skilled in the art from the following description.

In order to solve the above problems, there is provided a disease prediction method according to an embodiment of the present invention. The method may include receiving a medical image of a target site, using a disease prediction model trained to determine lesion areas for a plurality of predetermined diseases in the medical image, and including the lesion region in the medical image. And determining whether to develop one of the plurality of diseases based on the determined lesion area.

According to a feature of the invention, the plurality of diseases are two different diseases, the two diseases being tuberculosis spondylitis and purulent spondylitis, tuberculosis and pneumonia, solid tumors and water bumps, metastatic cancer and abscess, and osteolytic metastatic cancer and degenerative disease. May be a pair of diseases selected from the group consisting of lesions.

 According to another feature of the present invention, the step of predicting the presence of a disease may be performed by using a disease prediction model, calculating a probability of occurrence of each of a plurality of diseases, for a lesion area determined in a medical image, and for the lesion area. And determining that one of the plurality of diseases is developed based on the probability of developing each of the plurality of diseases.

According to another feature of the present invention, a disease having a probability of occurrence below a threshold among a plurality of newly emerging diseases may further include determining that the predicted reliability of the disease is low.

According to another feature of the invention, the lesion region determined by the ring prediction model is a plurality, the step of predicting the presence of the disease is a plurality of diseases for each of the plurality of lesion areas determined in the medical image using the disease prediction model Calculating a probability of each occurrence, determining a single final onset probability having a maximum value for each of the calculated plurality of diseases, and a final onset probability for each of the plurality of determined diseases In comparison, the method may include determining that one of the plurality of diseases has developed.

According to another feature of the present invention, the medical image includes a plurality of medical images for one target site, and the determining includes determining a lesion area for each of the plurality of medical images using a disease prediction model. The predicting of the disease may include calculating a probability of occurrence of each of the plurality of medical images by using a disease prediction model, and based on a probability of each of the plurality of diseases for the plurality of medical images. And determining whether one of the diseases is developed.

According to another feature of the present invention, the determining of the onset of one of the plurality of diseases may include determining the onset of each of the plurality of diseases, the probability of occurrence of each of the plurality of diseases calculated for the lesion area in the plurality of medical images. Determining a final incidence probability of each one, calculating a sum of the final incidence probability determined in each of the plurality of medical images for each of the plurality of diseases, and calculating a ratio of the sum of the final incidence probability for each of the plurality of diseases. And determining that one of the plurality of diseases has developed based on the ratio of the sum of the calculated final occurrence probability.

According to another feature of the present invention, the determining of the onset of one of the plurality of diseases may include determining the onset of each of the plurality of diseases, the probability of occurrence of each of the plurality of diseases calculated for the lesion area in each of the plurality of medical images. Determining the incidence of one disease as a final incidence probability, replacing the incidence of the other diseases by zero, and calculating a sum of the final incidence probability determined in each of the plurality of medical images for each of the plurality of diseases. Comprising: calculating a ratio of the sum of the final probability of occurrence for each of the plurality of diseases, and based on the calculated ratio of the sum of the final probability of occurrence may include the step of determining that one of the plurality of diseases has developed.

According to another feature of the invention, the step of predicting the presence of a disease in each of the plurality of medical images, calculating the average probability of occurrence for each of the plurality of diseases for the lesion area, a plurality of for each of the plurality of diseases Calculating a sum of the average incidence probability calculated on each of the medical images, calculating a ratio of the sum of the average incidence probability for each of the plurality of diseases, and one of the plurality of diseases based on the calculated ratio of the sum of the average incidence probability The method may further include determining that the disease is caused.

According to another feature of the present invention, the medical image may be a tomography image or a magnetic resonance image, and the predetermined disease may be a disease that can be diagnosed as a tomography image or a magnetic resonance image.

According to another feature of the invention, the predictive model is further configured to calculate the probability of occurrence on a pixel-by-pixel basis for the medical image, receiving a selection for a specific region having a certain pixel of the medical image, using the disease prediction model The method may further include calculating a probability of occurrence of a predetermined disease for the selected specific region, and providing a calculated probability of occurrence of the specific region.

According to another feature of the present invention, when the lesion area is determined by the disease prediction model, the lesion area determined in the medical image is displayed and provided, and when the lesion area is not determined by the disease prediction model, normal diagnostic information is provided. Providing may further provide.

According to another feature of the invention, the disease to determine the lesion area for the predetermined disease in the medical image for the target region based on the learning medical image including the lesion area confirmed as a predetermined disease for the target site Training of the predictive model may be further included.

According to another feature of the invention, the medical image is a T2-weighted image, the training step of cropping (cropping) the training medical image to a predetermined size, normalizing the pixels of the training medical image, Masking the lesion area identified as a predetermined disease in the training medical image, and training the disease prediction model to form a box around the masked lesion area.

In order to solve the above problems, there is provided a disease prediction device according to an embodiment of the present invention. The device includes a receiver configured to receive a medical image of a target site, and a processor operatively connected with the receiver. In this case, the processor determines a lesion area in the medical image by using a predictive model trained to determine lesion areas for a plurality of predetermined diseases in the medical image, and based on the determined lesion area, one of the plurality of diseases. It is configured to determine the onset of the disease.

According to a feature of the present invention, the processor calculates the incidence of each of the plurality of diseases for the lesion area determined in the medical image by using the disease prediction model, and based on the incidence of each of the plurality of diseases for the lesion area. It may be further configured to determine that one of the plurality of diseases is onset.

According to another feature of the invention, the medical image comprises a plurality of medical images for one target site, the processor calculates the probability of the occurrence of each of the plurality of diseases in each of the plurality of medical images using a disease prediction model The onset probability of each of the plurality of diseases of the plurality of medical images may be further determined to determine whether one of the plurality of diseases is developed.

The present invention has an effect of providing accurate diagnostic information on a target site of a subject by providing a disease prediction method and a device using the prediction model configured to predict whether a plurality of predetermined diseases occur.

Specifically, the present invention enables accurate diagnosis of a disease by calculating and providing an incidence of occurrence of each of a plurality of diseases with respect to a lesion region determined by a disease prediction model.

In addition, the present invention has an effect of predicting and providing a probability of occurrence of a disease in a region selected in a medical image.

The present invention is configured to receive a plurality of medical images to calculate the probability of disease occurrence for each of the plurality of diseases and to calculate the final probability score to predict whether or not to develop one disease, the disease appears in a similar form in the medical image There is an effect that can increase the accuracy of the diagnosis for each.

Furthermore, when the lesion region is determined for the target region of the subject by the disease prediction model, the determined lesion region is displayed in the medical image and provided, and when the lesion region is not determined by the disease prediction model, This has the effect of providing diagnostic information. Thus, the present invention has the effect of providing accurate diagnostic results for the target site of the subject regardless of the proficiency of the medical image diagnosis system.

The effects according to the present invention are not limited by the contents exemplified above, and more various effects are included in the present specification.

1 illustrates a configuration of a disease prediction device according to an embodiment of the present invention.

Figure 2a illustrates the procedure of a disease prediction method according to an embodiment of the present invention.

FIG. 2B exemplarily illustrates a procedure for determining a disease occurring in a medical image by a disease prediction method according to an exemplary embodiment of the present invention.

2C exemplarily illustrates various occurrence probability calculation procedures for a lesion area determined by a disease prediction method according to an embodiment of the present invention.

3 exemplarily illustrates a learning procedure for a disease prediction model used in a disease prediction method and a disease prediction device using the same according to an embodiment of the present invention.

FIG. 4A illustrates a lesion area of tuberculous spondylitis and purulent spondylitis determined by a disease prediction model used in a disease prediction method and a disease prediction device using the same according to an embodiment of the present invention.

4B illustrates the results of a disease prediction method and a disease prediction model used in a disease prediction device using the same, and a predictive evaluation of tuberculous spondylitis and purulent spondylitis by three radiologists according to an embodiment of the present invention.

Figure 4c illustrates the specificity and accuracy of the disease prediction method and the disease prediction model used in the disease prediction device using the same, and the diagnosis of tuberculous spondylitis and purulent spondylitis of three radiologists according to an embodiment of the present invention .

5A illustrates a plurality of medical images predicted as tuberculosis by a disease prediction model used in a disease prediction method and a disease prediction device using the same according to an embodiment of the present invention.

5B illustrates a plurality of medical images predicted as pneumonia by a disease prediction model used in a disease prediction method and a disease prediction device using the same according to an embodiment of the present invention.

Advantages of the 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 will be implemented in various forms, and only the present embodiments are intended to complete the disclosure of the present invention, and the general knowledge in the art to which the present invention pertains. It is provided to fully convey the scope of the invention to those skilled in the art, and the present invention is defined only by the scope of the claims.

Shapes, sizes, ratios, angles, numbers, and the like disclosed in the drawings for describing the embodiments of the present invention are exemplary, and the present invention is not limited to the illustrated items. In addition, in describing the present invention, if it is determined that the detailed description of the related known technology may unnecessarily obscure the subject matter of the present invention, the detailed description thereof will be omitted. When 'comprises', 'haves', 'consists of' and the like mentioned in the present specification are used, other parts may be added unless 'only' is used. In case of singular reference, the plural number includes the plural unless specifically stated otherwise.

In interpreting a component, it is interpreted to include an error range even if there is no separate description.

Each of the features of the various embodiments of the present invention may be combined or combined with each other in part or in whole, various technically interlocking and driving as can be understood by those skilled in the art, each of the embodiments may be implemented independently of each other It may be possible to carry out together in an association.

For clarity of interpretation of the present specification, the terms used herein will be defined below.

As used herein, the term "medical image" may refer to all images taken for diagnosis of a target site. For example, the medical image may be a magnetic resonance image, a computed tomography image, an x-ray image, or an ultrasound image. Preferably, within the present specification, the medical image may be, but is not limited to, a magnetic resonance image, a tomography image, or a T2 weighted image. The medical image of the target region may mean a 2D image, a 3D image, a still image of one cut, a moving image image including a plurality of cuts, and the like including the target portion. For example, when the medical image is a moving image including a plurality of cuts, the detection of the lesion area for each of the plurality of medical images and the prediction of the onset of the disease according to the disease prediction method according to an embodiment of the present invention are performed. It may be possible.

On the other hand, disease prediction based on the medical image of the target site may be particularly important in the early diagnosis of the disease. In this case, the term "purpose site" as used herein may be a specific body part of a subject to predict a condition such as the presence or absence of a disease. For example, the target site may be the chest, spine, upper abdomen, lower abdomen, lungs, brain, liver, varicose veins, uterus, prostate, testes, musculoskeletal system, thyroid or breast. However, the target site is not limited thereto and may be various sites as long as the image is acquired by the imaging apparatus.

As used herein, the term “predetermined disease” may mean any disease or lesion that can be diagnosed based on a medical image obtained from an imaging device. In this case, the predetermined disease may be a plurality. More specifically, the predetermined diseases are two different diseases from each other, and may be diseases that appear in a similar form in a medical image. For example, two predetermined diseases may be tuberculosis spondylitis and purulent spondylitis, or tuberculosis and pneumonia, or solid tumors and water bumps, or metastatic cancer and abscess, or at least one pair of osteolytic metastatic cancer and degenerative lesions. However, it is not limited thereto.

As used herein, the term "lesion area" may refer to an area having a lesion for a particular disease that is different from normal tissue in a medical image for a target site. For example, the lesion area may have different pixel values, textures for different areas. In this case, the lesion area may include an area for cyst, inflammation, or mass tissue appearing at the target site. Meanwhile, a plurality of lesion areas may be independently distributed in a medical image of one target site.

As noted above, tuberculosis spondylitis and purulent spondylitis, or tuberculosis and pneumonia, or solid tumors and water bumps, or metastatic cancers and abscesses, or osteolytic metastases and degenerative lesions, respectively, can all appear similarly within a medical image, Depending on the skill, the accuracy of diagnosis of the disease may vary.

In order to solve this problem, a predictive model trained to predict a lesion area in a medical image and predict the onset of a disease may be used.

As used herein, the term “disease prediction model” may be a predictive model trained to predict a lesion area with respect to a target site of a subject and determine whether to develop one of a plurality of diseases. For example, the disease prediction model may be a model trained using a data set of each of the medical images including a lesion region confirmed as two diseases predetermined by the medical image for the target region. In this case, the disease prediction model may be trained to determine the lesion area for each of the plurality of medical images of the target site and to calculate the probability of occurrence of the plurality of predetermined diseases for the lesion area. As a result, the disease prediction model may determine the onset of one of the plurality of diseases in the target region by calculating a final incidence probability based on the disease incidence probability calculated for each of the lesion areas in the plurality of medical images.

For example, the disease prediction model may include a probability of developing tuberculous spondylitis and purulent spondylitis, or a probability of developing tuberculosis and pneumonia, respectively, or a probability of each of solid tumors and water bumps, for lesion areas determined in a plurality of input medical images. Or the probability of developing metastatic cancer and abscess respectively, or the probability of developing osteolytic metastatic cancer and degenerative lesion, respectively. Finally, the disease prediction model includes one of tuberculosis spondylitis and purulent spondylitis, or one of tuberculosis and pneumonia, or one of solid tumors and water bumps, or one of metastatic cancer and abscess, or osteolytic metastasis. The onset of one of the cancerous and degenerative lesions can be determined.

 As used herein, the term "probability of occurrence" may mean the probability that the lesion area determined by the disease prediction model is a lesion site for a predetermined disease. As described above, the disease prediction model may calculate an incidence of a plurality of diseases for the lesion area, respectively, and finally determine whether to develop one disease.

Finally, the medical practitioner may acquire a medical image including the probability of occurrence of the predicted lesion area through the disease prediction model, and further, determine whether the disease is finally predicted through the disease prediction model. Based on this, accurate diagnosis may be possible.

Hereinafter, a disease prediction method and a device using the same will be described with reference to FIGS. 1 to 2C.

First, a disease prediction device according to an embodiment of the present invention will be described in detail with reference to FIG. 1. 1 illustrates a configuration of a disease prediction device according to an embodiment of the present invention.

Referring to FIG. 1, the disease prediction device 100 includes a receiver 110, an inputter 120, an outputter 130, a storage 140, and a processor 150.

In detail, the receiver 110 may receive a medical image of a target part of a subject that can be obtained from an imaging apparatus. For example, the receiver 110 may receive a medical image of the spine, chest, upper abdomen, lower abdomen, lung, brain, liver, varicose vein, uterus, prostate, testes, musculoskeletal system, thyroid, or breast of the subject. On the other hand, the medical image of the target site of the subject received through the receiver 110 may be a plurality, it may include a region for the lesion tissue for a particular disease.

The input unit 120 is not limited to a keyboard, a mouse, a touch screen panel, and the like. The input unit 120 may set the disease prediction device 100 and instruct an operation of the disease prediction device 100. For example, the medical person may directly determine the lesion area within the medical image received by the receiver 110 through the input unit 120.

The output unit 130 may visually display the medical image received by the receiver 110. In addition, the output unit 130 may be configured to display, by the processor 150, the lesion area determined in the medical image, further, the probability of the disease predicted for the lesion area, and the diagnostic information of the finally determined disease. In addition, the output unit 130 may be configured to display normal diagnostic information when the lesion area is not determined in the medical image of the target site by the processor 150.

The storage unit 140 may be configured to store a medical image of the target part of the subject received through the receiver 110 and store an indication of the disease prediction device 100 set through the input unit 120. In addition, the storage 140 may store the lesion area determined by the processor 150 to be described later, and store the probability of occurrence of a plurality of diseases calculated for the determined lesion area and diagnostic information about one disease finally determined. Is configured to. However, without being limited to the above, the storage 140 may store various pieces of information determined by the processor 150.

The processor 150 may be a component for providing accurate prediction results for the disease prediction device 110. In this case, for disease prediction, the processor 150 may be configured to determine a lesion area in a medical image of a target region of a subject and use a disease prediction model configured to predict a disease occurrence probability for the lesion area. For example, the processor 150 may input a medical image obtained through the receiver 110 using a disease prediction model configured to predict a lesion area with respect to a predetermined disease with respect to a target part of a subject, within the medical image. It can be configured to determine the lesion area. In addition, the processor 150 may calculate the probability of occurrence of the plurality of predetermined diseases for the determined lesion area using the disease prediction model configured to calculate the probability of occurrence of the plurality of predetermined diseases. The processor 150 may be configured to determine that one of the normal or a plurality of diseases is developed based on the probability of occurrence of the disease finally calculated by the disease prediction model. In this case, the disease prediction model configured to determine the lesion area in the medical image and predict the onset of Xalhwan may be based on various learning models learned based on the image. For example, a prediction model used in various embodiments of the present invention may be a deep neural network (DNN), a convolutional neural network (CNN), a deep convolution neural network (DNN), a current recurrent neural network (RNN), or a restricted boltzmann machine (RBM). ), But may be a prediction model based on a deep belief network (DBN), a single shot detector (SSD) model, or a U-net, but is not limited thereto.

Meanwhile, the processor 150 determines a lesion area in the medical image on a pixel-by-pixel basis using a disease prediction model, calculates a probability of occurrence of the disease in the lesion area having a plurality of pixel units, and then selects one of the normal or multiple diseases. It can be further configured to predict the disease of.

Hereinafter, a disease prediction method according to an embodiment of the present invention will be described in detail with reference to FIGS. 2A to 2C. In this case, it is explained by using an example of determining whether the tuberculous spondylitis (TB) or pyogenic spondylitis (Pyo) is developed on the basis of MRI images of the TL spine (Thoracic-Lumbar spine) of the subject. However, the present invention is not limited thereto, and the disease prediction method according to an exemplary embodiment of the present invention may be used to predict whether various diseases occur in a medical image of a more various target site.

Figure 2a illustrates the procedure of a disease prediction method according to an embodiment of the present invention. FIG. 2B exemplarily illustrates a procedure for determining a disease occurring in a medical image by a disease prediction method according to an exemplary embodiment of the present invention. 2C exemplarily illustrates various occurrence probability calculation procedures for a lesion area determined by a disease prediction method according to an embodiment of the present invention.

Referring to Figure 2a, the disease prediction procedure according to an embodiment of the present invention is as follows. First, a medical image of a target part of a subject is received (S210). Next, the lesion area is determined in the medical image by using a disease prediction model configured to predict lesion areas for a plurality of predetermined diseases in the medical image (S220). Finally, the disease prediction model is used to determine whether the disease region determined in the medical image is normal or one of the plurality of diseases is developed (S230).

For example, referring to FIG. 2B, in operation S210 of receiving a medical image, a plurality of 15 medical images 212 for a thoracic-lumbar region set as a target site may be obtained. In this case, the obtained plurality of medical images 212 may be a T2-weighted axial plane image that is easy to analyze the lesion area.

In various embodiments of the present disclosure, in the receiving of the medical image (S210), the medical image, which has been pre-processed, may be further received to enable fast analysis of the plurality of medical images 212. Otherwise, the plurality of medical images may be adjusted to have a predetermined pixel unit or to adjust contrast, resolution, contrast, or symmetry of the plurality of medical images 212 received after receiving the medical image (S210). A preprocessing step for the medical image 212 may be further performed. As a result of the preprocessing step, the plurality of medical images 212 may have a resolution or size required by the disease prediction model 222 to be described later, and may have a smaller resolution or size than a plurality of original medical images, thereby predicting a disease. The processing speed in model 222 can be improved.

Next, referring to FIG. 2B, in the determining of the lesion area (S220), the plurality of medical images 212 are included in the thoracic-lumbar region obtained in the step S210 of receiving the medical image in the disease prediction model 222. Is entered. In this case, the disease prediction model 222 may determine the lesion area for the disease based on at least one of pixel values, textures, and pixel differences with respect to the surrounding areas of the plurality of areas existing in the plurality of medical images 212. You can decide. For example, the disease prediction model 222 can determine the lesion area for tuberculous spondylitis or purulent spondylitis in each of the plurality of medical images 212.

According to another embodiment of the present invention, a box may be formed around the lesion area determined by the disease prediction model 222 in determining the lesion area (S220). For example, referring to the plurality of medical images 224 in which the lesion area of FIG. 2B is determined, lesions for tuberculous spondylitis or purulent spondylitis determined for each of the plurality of medical images 212 by the disease prediction model 222. A box may be formed around the area. Thus, the medical person can easily recognize the lesion area determined for the disease.

Furthermore, according to another embodiment, when the lesion region is not determined by the disease prediction model 222 in determining the lesion region (S220), the disease prediction model 222 determines that there is no disease at the target site. Normal diagnostic information can be provided to a healthcare practitioner.

Next, referring to FIG. 2B, in step S230 of determining whether the disease occurs, in each of the plurality of medical images 224 in which the lesion area is determined by the disease prediction model 222, the onset of tuberculous spondylitis in the determined lesion area Probability (TB score) and Pyo score of purulent spondylitis are calculated. Finally, in the step of determining whether the disease occurs (S230), the incidence of TB spondylitis (TB score) and the incidence of purulent spondylitis (Pyo) calculated for each of the plurality of medical images 212 by the disease prediction model 222. based on the score), a tuberculous spondylitis having a high final probability score may be determined as the diseased disease, and a medical image 232 in which the diseased disease is predicted may be provided.

According to another exemplary embodiment of the present disclosure, in the determining of whether the disease occurs (S230), in each of the plurality of medical images 212, one of the probabilities of occurrence of each of the plurality of diseases calculated for the lesion area may be determined. The final onset probability is determined separately, the sum of the final incidence probability determined in each of the plurality of medical images for each of the plurality of diseases is calculated, the ratio of the sum of the final incidence probability for each of the plurality of diseases is calculated, and the calculated final onset And may determine that one of the plurality of diseases has developed on the basis of the ratio of the sum of the probabilities.

For example, referring to (a) of FIG. 2C, a slice of four medical images (Slice # 1, Slice # 2, Slice # 3, and Slice # 4) of the thoracic and lumbar spine by the disease prediction model 222 may be used. In case of # 1, 0.9 and Lesion # 3 purulent spondylitis of TB, which is the incidence of TB spondylitis of Lesion # 2 having the maximum of a plurality of lesion areas (Lesion # 1, Lesion # 2 and Lesion # 3) A Pyo score of 0.72 can be determined as the final probability of onset. In the same way, the disease prediction model 222 can determine TB incidence (TB score) and Pyo score of pyogenic spondylitis (Pyo score) in each of the remaining three medical images (Slice # 2, Slice # 3 and Slice # 4). Final probability of onset can be determined. The disease prediction model 222 then calculates the sum of tuberculous spondylitis final incidence probability (3.53) and the final incidence probability of purulent spondylitis (3.05) for the four medical images, followed by Calculate the ratio (3.53 / (3.53 + 3.05)) of the incidence of spondylitis. As a result, the final probability score is calculated to be 0.54, and as the probability of developing tuberculous spondylitis is relatively high, the disease prediction model 222 may determine that tuberculous spondylitis is developed at the target site.

According to another embodiment of the present invention, in the step of determining whether the disease (S230), in each of the plurality of medical images 212, one of the probability of the occurrence of each of the plurality of diseases calculated for the lesion area, the one having the maximum value Determine the incidence of onset of the disease as the final incidence, replace the incidence of the remaining disease with zero, calculate the sum of the final incidence determined in each of the plurality of medical images for each of the plurality of diseases, And calculating a ratio of the sum of the final probability of occurrence for each disease, and determining that one of the plurality of diseases is on the basis of the calculated percentage of the sum of the final probability of occurrence.

For example, referring to (b) of FIG. 2C, a slice of four medical images (Slice # 1, Slice # 2, Slice # 3, and Slice # 4) of the thoracic and lumbar spine by the disease prediction model 222 may be used. In the case of # 1, 0.9, which is the probability of occurrence of spondylitis (TB score) having the maximum value among the plurality of lesion areas (Lesion # 1, Lesion # 2, and Lesion # 3), may be determined as the final occurrence probability. At this time, the probability of occurrence of purulent spondylitis (Pyo score) can be replaced with zero. In the same way, the disease prediction model 222 can determine the final incidence probability for one spondylitis having a maximum for total spondylitis in each of the remaining three medical images (Slice # 2, Slice # 3, and Slice # 4). . The disease prediction model 222 then calculates the sum of tuberculous spondylitis final incidence probability (2.74) and the final incidence probability of purulent spondylitis (0.88) for four medical images, and then the tuberculosis for overall spondylitis incidence probability. Calculate the ratio (2.74 / (2.74 + 0.88)) of the probability of developing spondylitis. As a result, the final probability score is calculated to be 0.76, which indicates that the incidence of tuberculous spondylitis is relatively high, so that the disease prediction model 222 may determine that tuberculous spondylitis is developed at the target site.

According to another exemplary embodiment of the present disclosure, in the determining of whether the disease occurs (S230), in each of the plurality of medical images 212, an average occurrence probability of each of the plurality of diseases calculated for the lesion area is calculated, and the plurality of Calculate the sum of the average incidence probability calculated from each of the plurality of medical images for each of the disease, calculate the ratio of the sum of the average incidence probability for each of the plurality of diseases, and determine the plurality of It can be configured to determine that one of the diseases of the dog is onset.

For example, referring to (c) of FIG. 2C, a slice of four medical images (Slice # 1, Slice # 2, Slice # 3, and Slice # 4) of the thoracic-lumbar spine by the disease prediction model 222 may be used. In case of # 1, the average incidence of 0.82 and TB was 0.82 and ascites for TB incidence of TB spondylitis (TB score) calculated in each of the multiple lesion areas (Lesion # 1, Lesion # 2 and Lesion # 3), respectively. The mean onset probability 0.67 for the Pyo scores of 0.6, 0.7, and 0.72, which is calculated in each of the lesion areas (Lesion # 1, Lesion # 2, and Lesion # 3), is calculated. In the same way, the disease prediction model 222 can determine TB incidence (TB score) and Pyo score of pyogenic spondylitis (Pyo score) in each of the remaining three medical images (Slice # 2, Slice # 3 and Slice # 4). The average probability of onset can be calculated. The disease prediction model 222 then calculates the sum of tuberculosis spondylitis onset probability (3.40) and the sum of the mean onset probability of purulent spondylitis (2.94) on four medical images, and then the tuberculosis of overall spondylitis probability Calculate the ratio (3.40 / (3.40 + 2.94)) of the incidence of spondylitis. As a result, the final probability score is calculated to be 0.54, and as the probability of developing tuberculous spondylitis is relatively high, the disease prediction model 222 may determine that tuberculous spondylitis is developed at the target site.

According to another embodiment of the present invention, among the plurality of emerging diseases, a disease having a probability of occurrence below the threshold may be determined to have low predicted reliability of the disease. For example, referring to (a), (b) and (c) of FIG. 2C, the TB probability and the probability of developing purulent spondylitis (TB score) calculated for each lesion of a plurality of medical images ( Pyo score) is greater than 0.6. If a value having a probability of 0.6 or less among the calculated TB score or Pyo score of purulent spondylitis is determined to be non-discriminatory, it is determined that the final probability score for determining the onset is determined. May be excluded.

As a result of various embodiments, diagnostic information for one disease, determined based on the probability of occurrence of each of the plurality of diseases calculated for each of the plurality of medical images, is provided so that the medical practitioner can obtain the target with higher accuracy regardless of skill level. Determination of the onset of the site can be made.

Hereinafter, a learning method of a disease prediction model will be described with reference to FIG. 3. In this case, the medical image used for learning was used for the patient who was finally diagnosed with tuberculous spondylitis or purulent spondylitis for the thoracic-lumbar region, but the medical image includes various target areas depending on the disease to be predicted. This can be used for learning.

3 exemplarily illustrates a learning procedure for a disease prediction model used in a disease prediction method and a disease prediction device using the same according to an embodiment of the present invention.

Referring to FIG. 3, a disease prediction model used in various embodiments of the present invention is trained with a learning set of a plurality of medical images 312 for the thoracic-lumbar region. More specifically, in order to train the disease prediction model, the plurality of medical images 312 may be acquired as the T2 weighted image 313 in which the lesion area is displayed with high luminance. The T2-weighted medical image 313 can then be normalized to have a pixel between 0 and 255 and cropped to a size of about 10 × 10 cm. The lesion area for tuberculous spondylitis or purulent spondylitis may then be masked within the cropping and standardized medical image 314. At this time, the disease prediction model may be trained to form a lesion area box 318 around the lesion area with respect to the medical image 315 where the lesion area is masked, including the masked lesion area 316. On the other hand, as lesion areas less than 1.5 cm ² may be false positive, the disease prediction model of the present invention may be trained not to form lesion area box 318 for lesion areas less than 1.5 cm ^2. Can be. The predictive model trained by the above method may form a lesion area box 318 for each of the plurality of medical images for the input target region, and as a result, the medical image in which the lesion area box is formed for each of the plurality of diseases. 317 may be provided.

Meanwhile, the disease prediction model of the present invention may be a model configured to determine a lesion area on the basis of an SSD performing recognition in a unified framework by recognizing a medical image of a target region as multiple boxes. However, the present invention is not limited thereto, and the disease prediction model used in various embodiments of the present disclosure may be based on various algorithms based on images.

Example 1 Evaluation of Disease Prediction Models Tuberculosis Spondylitis or Purulent Spondylitis

In Example 1 below, an evaluation method and results of a prediction model used in various embodiments of the present invention will be described with reference to FIGS. 4A to 4C.

In this evaluation, the predictive rate for tuberculous spondylitis or purulent spondylitis determined by each of the disease prediction model of the present invention and each of three radiology specialists is compared and explained.

FIG. 4A illustrates a lesion area of tuberculous spondylitis and purulent spondylitis determined by a disease prediction model used in a disease prediction method and a disease prediction device using the same according to an embodiment of the present invention. 4B illustrates a disease prediction method and a disease prediction model used in a disease prediction device using the same, and a prediction evaluation of tuberculous spondylitis and purulent spondylitis by three radiologists according to one embodiment of the present invention. . Figure 4c illustrates the specificity and accuracy of the disease prediction method and the disease prediction model used in the disease prediction device using the same, and the diagnosis of tuberculous spondylitis and purulent spondylitis of three radiologists according to an embodiment of the present invention .

Referring to FIG. 4A (a), a thoracic-lumbar spine image finally determined as tuberculous spondylitis by a disease prediction model used in various embodiments of the present invention is shown. More specifically, the disease prediction model determined lesion areas with respect to the input thoracic-lumbar spine images to generate cyan boxes around the lesion areas, and calculated the probability of developing tuberculous spondylitis for the lesion areas within the boxes. As a result, the incidence of tuberculous spondylitis for each of the lesion areas was 1.00 and 0.91, and the disease prediction model could finally determine that tuberculous spondylitis was developed for the target site.

Referring to FIG. 4A (b), a thoracic-lumbar spine image finally determined as purulent spondylitis by a disease prediction model used in various embodiments of the present invention is shown. More specifically, the disease prediction model determined lesion areas for tuberculous spondylitis and purulent spondylitis on input thoracic-lumbar spine images, and generated cyan boxes and red boxes around the lesion areas, respectively. The disease prediction model then calculated the probability of developing tuberculous spondylitis and the probability of purulent spondylitis for the lesion area present in the box. As a result, the incidence of tuberculous spondylitis and the probability of purulent spondylitis were calculated to be 0.75 and 0.88 for each of the two lesion regions, and the disease prediction model finally showed the development of purulent spondylitis with a relatively high probability for the target site. Can be determined.

Referring to (c) of FIG. 4, a thoracic-lumbar spine image finally determined to be normal by a disease prediction model used in various embodiments of the present invention is shown. More specifically, when the lesion region is not determined by the disease prediction model or the lesion area box is not formed, the disease prediction model may finally be determined to be normal to the target site.

Referring to (a) of FIG. 4B, in the fourth evaluation of the disease prediction model, an area under the curve (AUC) value, which means a result hit ratio, associated with excellent diagnostic ability, was respectively measured. At this time, the disease prediction model was finally determined as tuberculous spondylitis or purulent spondylitis based on the final probability scores of tuberculous spondylitis and purulent spondylitis calculated on thoracolumbar lumbar spine images. Then, the AUC value in each evaluation was calculated, and the sensitivity (Sen, sensitivity) specificity (Spe, specificity) was calculated by comparing the predetermined correct disease for the image with the disease determined by the disease prediction model in various ways. . As a result, it was found to have high AUC values of 0.723, 0.856, 0.853, and 0.761 in the 1st, 2nd, 3rd, and 4th phases respectively, and the average AUC value for the total 4th evaluation was 0.802. Furthermore, when a straight line with a slope of 1 at each point of the ROC curve was drawn, the sensitivity and specificity of the (Youden index) were 85.0 and 67.9, respectively. The sensitivity and specificity by the optimal threshold according to the equation for the figure were 73.8 and 75.3, respectively.

On the other hand, referring to Figure 4b (b), the AUC value measured according to the prediction results of tuberculous spondylitis and purulent spondylitis by three radiologists (Reader 1, Reader 2 and Reader 3) in the same manner as described above Appears. As a result, the average AUC value according to the prediction result of three radiologists has a 0.729 lower than the 0.80 AUC value of the disease prediction model of the present invention. These results may mean that the disease prediction model of the present invention has a diagnosis ability superior to that of a radiologist in the diagnosis of tuberculous spondylitis or purulent spondylitis.

More specifically, referring to (a) of FIG. 4C, the specificity of diagnosis by the disease prediction model of the present invention and the diagnostic radiology specialists (Reader 1, Reader 2 and Reader 3) of the present invention appears to be similar. Furthermore, referring to (b) of FIG. 4C, in the disease prediction model, when the threshold of the final probability score for determining the onset of the disease is set to 0.5, the accuracy of diagnosis for the patient is increased even though the epoch increases. It appears to remain at 0.7.

As a result of Example 1, it was confirmed that the disease prediction model has a high discrimination ability against two diseases that appear similarly in a medical image, for example, tuberculous spondylitis and purulent spondylitis. Thus, the disease prediction device of the present invention using the disease prediction model may determine whether the disease occurs with high accuracy with respect to the lesion area determined in the medical image of the target site.

Example 2 Evaluation of Disease Prediction Models Pneumonia or Tuberculosis

In Example 2 below, an evaluation method and results of a prediction model used in various embodiments of the present invention will be described with reference to FIGS. 5A and 5B.

In the evaluation of this disease prediction model, evaluation of the prediction of pneumonia or tuberculosis was performed based on ten computed tomography images (Image 1 to Image 10) of the chest lung. However, the present invention is not limited thereto, and the disease prediction model of the present invention may predict each disease with high accuracy for two diseases having similar shapes in the medical image.

5A illustrates a plurality of medical images predicted as tuberculosis by a disease prediction model used in a disease prediction method and a disease prediction device using the same according to an embodiment of the present invention. 5B illustrates a plurality of medical images predicted as pneumonia by a disease prediction model used in a disease prediction method and a disease prediction device using the same, according to an embodiment of the present invention.

Referring to FIG. 5A, a red box and a cyan box and tuberculosis formed around the lesion area of each of tuberculosis (Tb, tuberculosis) and pneumonia (Pn, pneumonia) determined in a medical image of a chest lung by a disease prediction model And 10 images showing the probability of developing each of pneumonia.

More specifically, referring to Table 1 below, tuberculosis (Tb), which is calculated for lesion areas determined in each of 10 images (Images 1 to Image 10) acquired for the lung of the target region, was obtained. Onset probability and onset of pneumonia (Pn, pneumonia).

The disease prediction model yields the ratio of the incidence of tuberculosis (5.84 / (1.56 + 5.84)) to the probability of total disease (tuberculosis and pneumonia) at 1.56 and 5.84, the sum of both the probability of developing pneumonia and tuberculosis. do. The final probability score was calculated to be about 0.789 by the disease prediction model, indicating that the incidence of tuberculosis is relatively high, and thus, the target site having the computed tomography image of the chest lung shown in FIG. 5A. Can be determined to have tuberculosis.

	pneumonia	tb
image 1	0	0
image 2	0.95	0
image 3	0	0.9
image 4	0	0.99
image 5	0.61	0.98
image 6	0	One
image 7	0	0.99
image 8	0	0.98
image 9	0	0
image 10	0	0
sum	1.56	5.84
final score (favoring tb)		0.78918919

Referring to FIG. 5B, a red box and a cyan box and tuberculosis formed around the lesion area of each of tuberculosis (Tb, tuberculosis) and pneumonia (Pn, pneumonia) determined in a medical image of a chest lung by a disease prediction model And 10 images showing the probability of developing each of pneumonia.

More specifically, referring to [Table 2] below, tuberculosis (Tb, tuberculosis) calculated for the lesion area determined in each of 10 images (Image 1 to Image 10) acquired for the lung of the target region. Probability of onset and probability of developing pneumonia (Pn, pneumonia).

	pneumonia	tb
image 1	0.82	0
image 2	0.81	0
image 3	0.98	0
image 4	0.99	0
image 5	0.8	0.71
image 6	0	0.67
image 7	0	0
image 8	0	0
image 9	0	0
image 10	0.9	0
sum	5.3	1.38
final score (favoring tb)		0.206586826

The disease prediction model is the ratio of the incidence of tuberculosis to the probability of total disease (tuberculosis and pneumonia) at 5.3 and 1.38, which is the sum of the incidence of each pneumonia and the probability of tuberculosis (1.38 / (5.3 + 1.38)) To calculate. The final probability score was calculated to be about 0.206 by the disease prediction model, indicating that the incidence of tuberculosis was relatively low, so that the target site having the computed tomography image of the chest lung shown in FIG. 5B. Can be determined to have developed pneumonia.

As a result of Example 2, the disease prediction method and the device using the same according to an embodiment of the present invention based on a disease prediction model to determine the lesion area in the medical image of the target site of the subject with high accuracy, There is an effect that can predict the onset.

In particular, the disease prediction method and the device using the same according to an embodiment of the present invention, by determining the disease with a high accuracy for a plurality of similar diseases having a similar shape in the medical image, thereby obtaining an accurate diagnosis result regardless of the skill of the medical practitioner There is an effect that can be provided.

On the other hand, the disease prediction method according to an embodiment of the present invention and the use range and effects of the device using the same are not limited. For example, a disease prediction method and a device using the same according to an embodiment of the present invention may provide a probability of occurrence of a predetermined disease with respect to a specific area in a medical image selected by a medical person, and develop the disease based on the occurrence probability. Alternatively, the healthcare provider may be provided with information that is normally predicted.

 Furthermore, the disease prediction method and the device using the same according to an embodiment of the present invention, as well as the thoracic-lumbar region and lungs, the upper abdomen, lower abdomen, liver, brain, varicose veins, uterus, prostate, testicles, musculoskeletal system, or breast For various sites where an ultrasound scan can be performed, lesion areas can be predicted.

Although the embodiments of the present invention have been described in more detail with reference to the accompanying drawings, the present invention is not necessarily limited to these embodiments, and various modifications can be made without departing from the spirit of the present invention. Therefore, the embodiments disclosed in the present invention are not intended to limit the technical idea of the present invention but to describe the present invention, and the scope of the technical idea of the present invention is not limited by these embodiments. Therefore, it should be understood that the embodiments described above are exemplary in all respects and not restrictive. The protection scope of the present invention should be interpreted by the following claims, and all technical ideas within the equivalent scope should be interpreted as being included in the scope of the present invention.

* Description of the sign

100: disease prediction device

110: receiver

120: input unit

130: output unit

140: storage unit

212, 312: multiple medical images

222: disease prediction model

224: A plurality of medical images of the lesion area and the probability of onset

232: Medical image predicted disease

313: T2 highlighted medical imaging

314: Cropped and Standardized Medical Imaging

315: Medical image masking lesion area

316: masking lesion area

317: Medical image with lesion area box formed

318: lesion area box

* National R & D project supporting this invention

[Operation Unique Number] 1711051031, [Department Name] Ministry of Science and ICT, [Research Management Specialized Institution] Korea Research Foundation, [Research Project Name] Personal Basic Research (Future Part), [Research Title] Study of Automated Fracture Fractures in Women, [Contribution Rate] 1/1

Claims (18)

1. Receiving a medical image of a target site;

   Determining a lesion region in the medical image using a disease prediction model trained to determine lesion regions for a plurality of predetermined diseases in the medical image, and

   And determining whether the disease occurs for one of the plurality of diseases based on the determined lesion area.
2. The method of claim 1,

   The plurality of diseases are two different diseases from each other,

   Wherein said two diseases are a pair of diseases selected from the group consisting of tuberculous spondylitis and purulent spondylitis, tuberculosis and pneumonia, solid tumors and water bumps, metastatic cancer and abscess, and osteolytic metastatic cancer and degenerative lesions.
3. The method of claim 1,

   Predicting whether the disease is,

   Calculating a probability of occurrence of each of the plurality of diseases with respect to the lesion area determined in the medical image by using the disease prediction model, and

   And determining that one of the plurality of diseases is developed based on a probability of occurrence of each of the plurality of diseases for the lesion area.
4. The method of claim 3,

   The disease prediction method of the said several diseases which have a probability of onset below a threshold further including determining that the predicted reliability of a disease is low.
5. The method of claim 3,

   There are a plurality of lesion regions determined by the disease prediction model,

   Predicting whether the disease is,

   Calculating a probability of occurrence of each of the plurality of diseases for each of the plurality of lesion areas determined in the medical image by using the disease prediction model;

   Determining, in all of the plurality of lesion areas, one final onset probability having a maximum value for each of the calculated plurality of diseases, and

   Comparing the final probability of occurrence for each of the determined plurality of diseases, and determining that one of the plurality of diseases has developed.
6. The method of claim 3,

   The medical image includes a plurality of medical images of one target region,

   The determining step,

   Determining a lesion area for each of the plurality of medical images by using the disease prediction model;

   Predicting whether the disease is,

   Calculating a probability of occurrence of each of the plurality of diseases in each of the plurality of medical images by using the disease prediction model, and

   And determining whether one of the plurality of diseases occurs on the basis of the occurrence probability of each of the plurality of diseases with respect to the plurality of medical images.
7. The method of claim 6,

   Determining whether one of the plurality of diseases is onset,

   Determining, in each of the plurality of medical images, one final onset probability having a maximum value among the incidences of occurrence of each of the plurality of diseases calculated for the lesion area;

   Calculating, for each of the plurality of diseases, a sum of the final probability of occurrence determined in each of the plurality of medical images;

   Calculating a ratio of the sum of the final incidence probability for each of the plurality of diseases, and

   And determining that one of the plurality of diseases is on the basis of the calculated percentage of the sum of the final probability of occurrence.
8. The method of claim 6,

   Determining whether one of the plurality of diseases is onset,

   In each of the plurality of medical images, the probability of occurrence of one disease having a maximum value among the probability of occurrence of each of the plurality of diseases calculated for the lesion area is determined as a final occurrence probability, and the probability of occurrence of the remaining diseases. Replacing with 0;

   Calculating, for each of the plurality of diseases, a sum of the final probability of occurrence determined in each of the plurality of medical images;

   Calculating a ratio of the sum of the final incidence probability for each of the plurality of diseases, and

   And determining that one of the plurality of diseases is on the basis of the calculated percentage of the sum of the final probability of occurrence.
9. The method of claim 6,

   Predicting whether the disease is,

   Calculating an average probability of occurrence of each of the plurality of diseases in the lesion area in each of the plurality of medical images;

   Calculating, for each of the plurality of diseases, a sum of average occurrence probabilities calculated from each of the plurality of medical images;

   Calculating a ratio of the sum of the average probability of occurrence for each of the plurality of diseases, and

   And determining that one of the plurality of diseases is on the basis of the calculated ratio of the sum of the average occurrence probability.
10. The method of claim 1,

    The medical image is a tomography image or a magnetic resonance image,

    The predetermined disease is a disease prediction method, which can be diagnosed by the tomography image or magnetic resonance image.
11. The method of claim 1,

    The prediction model is further configured to calculate a probability of occurrence on a pixel-by-pixel basis with respect to the medical image.

    Receiving a selection of a specific area having a predetermined pixel of the medical image;

    Calculating a probability of occurrence for the predetermined disease for the selected specific region using the disease prediction model, and

    And providing the probability of occurrence calculated for the particular region.
12. The method of claim 1,

    When the lesion area is determined by the disease prediction model,

    Displaying and providing the lesion area determined in the medical image,

    When the lesion area is not determined by the disease prediction model,

    Providing normal diagnostic information.
13. The method of claim 1,

    Training the disease prediction model to determine a lesion area for the predetermined disease in the medical image for the target area based on a training medical image including the lesion area confirmed as the predetermined disease for the target area. Further comprising the step of, disease prediction method.
14. The method of claim 13,

    The medical image is a T2-weighted image,

    The training step,

    Cropping the training medical image to a predetermined size;

    Normalizing pixels of the training medical image;

    Masking a lesion area confirmed as the predetermined disease in the learning medical image, and

    Training the disease prediction model to form a box around the lesion area masked.
15. A receiver configured to receive a medical image of a target site, and

    A processor operatively connected with the receiver,

    The processor,

    Determine a lesion area for the plurality of diseases in the medical image by using a predictive model trained to determine the lesion area for the plurality of predetermined diseases in the medical image,

    And determining whether to develop onset of one of a plurality of diseases based on the determined lesion area.
16. The method of claim 15,

    The processor,

    Calculating a probability of occurrence of each of the plurality of diseases with respect to the lesion area determined in the medical image by using the disease prediction model,

    And determine, based on the probability of developing each of the plurality of diseases for the lesion area, that one of the plurality of diseases is onset.
17. The method of claim 15,

    The medical image includes a plurality of medical images of one target region,

    The processor,

    The onset probability of each of the plurality of diseases in each of the plurality of medical images is calculated using the disease prediction model, and based on the incidence probability of each of the plurality of diseases in the plurality of medical images, The disease prediction device further configured to determine whether the disease of one of the diseases is onset.
18. The method of claim 15,

    The plurality of diseases are two different diseases from each other,

    Wherein said two diseases are a pair of diseases selected from the group consisting of tuberculous spondylitis and purulent spondylitis, tuberculosis and pneumonia, solid tumors and water bumps, metastatic cancer and abscess, and osteolytic metastatic cancer and degenerative lesions.

Images