KR102062539B1 - Assistance diagnosis method for lumbar disease based on deep learning - Google Patents

Abstract

The present invention relates to an ancillary diagnosis method for a lumbar disease based on deep learning which can accurately and automatically diagnose a lumbar disease. The ancillary diagnosis method comprises: (a) a step of using learning data to perform deep learning on lumbar diseases to generate a diagnosis model; (b) a step of inputting a lumbar disease image to be diagnosed; and (c) a step of diagnosing existence of a lumbar disease of the lumbar disease image to be diagnosed based on the diagnosis model. The step (a) includes: (a1) a step of inputting a plurality of lumbar images and segmentation images obtained by segmenting individual lumbar vertebrae for the lumbar images as first learning data; (a2) a step of using the segmentation images as output data to perform deep learning on the first learning data; (a3) a step of generating a segmentation model by deep learning using the first learning data as an input; (a4) a step of inputting a plurality of normal lumbar patches and a plurality of disease lumbar patches as second learning data; (a5) a step of performing deep learning on the second learning data by a preregistered classification algorithm; and (a6) a step of generating a disease classification model by deep learning using the second learning data as an input. In the step (c), the segmentation model and the disease classification model are applied as the diagnosis model.

Description

Deep learning-based lumbar disease-assisted diagnosis method {ASSISTANCE DIAGNOSIS METHOD FOR LUMBAR DISEASE BASED ON DEEP LEARNING}

The present invention relates to a deep learning-based lumbar disease-assisted diagnosis method, and more particularly, to deep lube-based deep learning that can automatically diagnose lumbar diseases such as compression fracture by learning various lumbar diseases through deep learning technology. It relates to a lumbar disease secondary diagnostic method.

In modern medicine, medical imaging is a very important tool for effective diagnosis and treatment of patients. In addition, advances in imaging technology have enabled the acquisition of more sophisticated medical image data. In exchange for the sophistication, the amount of data is gradually increasing, which makes it difficult to analyze medical image data depending on human vision. Thus, in recent decades, clinical decision support systems and computer-assisted diagnostic systems have played an essential role in automated medical image analysis.

Conventional clinical decision support systems or computer-assisted diagnostic systems detect and display lesion areas or present diagnostic information to medical personnel or healthcare practitioners and the like (hereinafter, users).

For example, in the method and apparatus for calculating medical diagnosis based on medical image disclosed in Korean Patent Laid-Open No. 10-2017-0017614, a region of interest in which an object to be analyzed is photographed is detected, a coefficient of variation is calculated, and a coefficient of variation It includes the step of creating an image and comparing it with a reference sample, and mentions the effect of diagnosing the degree of disease of the patient using medical images acquired through CT, MRI, and ultrasound imaging apparatus.

In particular, artificial intelligence (AI) technology based on machine learning, such as deep learning, has recently been the foundation for bringing dramatic advances in diagnosing patients' diseases using medical imaging. .

Deep learning refers to a machine learning method based on an artificial neural network that simulates a human neuron and allows a machine to learn. Recently, deep learning technology has been rapidly developed in the field of image recognition and widely used in the field of diagnosis using medical images.

In deep learning technology, it is necessary to repeatedly develop the training data to form a diagnostic model for diagnosing a disease. It is important to develop a diagnostic model specialized for each disease because there are various types of diseases used as the training data. This means that even if a diagnostic model for generating a near-perfect diagnosis result is generated in a specific disease, it can be applied to other diseases.

The lumbar spine is the portion of the spinal column between the thoracic and sacrum, which refers to the bone structure that forms the waist of the human body. The human lumbar spine consists of five vertebrae. Compression fractures, which are related to the lumbar spine, are defined as fractures occurring in the forearm of the anterior part of the spine because of the pressing force, that is, the compression force. Other fractures are classified.

Compression fractures of the spine are a disease that frequently occurs in older and postmenopausal women, mainly with osteoporosis. It usually occurs when a patient with severe osteoporosis falls down, but in very severe osteoporosis, it can occur without any extra events such as falling down or sitting down. It can also occur when very strong external force is applied without osteoporosis.

Compression fractures generally cause pain in the back or lower back with fractures, and often have minimal symptoms at the beginning, making it easy to miss treatment. In particular, multiple fractures and severe injuries can lead to kidney loss and scoliosis.

Therefore, the compression fracture should be able to be diagnosed quickly from the medical image. However, due to various forms of the spine, it is often difficult to diagnose compression fractures of the spine by simple imaging such as X-rays, which requires additional diagnosis such as CT or MRI.

Therefore, it would be desirable to develop a deep learning technology that can provide the function of assisting the medical decision making by X-ray image only, so that it is easier to identify not only the initial diagnosis but also additional imaging such as CT or MRI. . In addition, the development of a deep learning model that can be applied to lumbar diseases other than compression bone is urgently needed.

Accordingly, the present invention has been made in order to solve the above problems, deep learning-based lumbar disease aid that can automatically diagnose more lumbar diseases such as compression fracture by learning various lumbar diseases through deep learning technology The purpose is to provide a diagnostic method.

According to the present invention, in the deep learning-based lumbar disease-assisted diagnosis method, (a) deep learning lumbar disease using learning data to generate a diagnostic model, and (b) the lumbar image to be diagnosed Inputting and (c) diagnosing whether or not a lumbar disease of the lumbar image to be diagnosed is diagnosed based on the diagnostic model; The step (a) comprises (a1) inputting a plurality of lumbar vertebral images, segmentation images obtained by segmenting individual lumbar vertebrae for each of the lumbar vertebral images as first training data, and (a2) using the segmentation image as output data. Deep learning of the first training data; (a3) generating a segmentation model through deep learning using the first training data; and (a4) a plurality of normal lumbar patches and a plurality of diseases. Inputting the lumbar patch as second learning data, (a5) deep learning the second learning data through a pre-registered classification algorithm, and (a6) deep learning using the second learning data as input. Through which a disease classification model is generated; In the step (c), the segmentation model and the disease classification model are achieved by a deep learning-based lumbar disease assisted diagnosis method, characterized in that applied to the diagnostic model.

Here, the step (c) is (c1) generating a segmentation image corresponding to an individual lumbar spine from the diagnosis lumbar image through the segmentation model, and (c2) using the segmentation image, the diagnosis lumbar image Extracting the center point of the individual lumbar spine, (c3) extracting a rotation angle of the individual lumbar spine based on the center point of the individual lumbar spine, and (c4) the lumbar spine to be diagnosed based on the segmentation image and the center point. Extracting an individual lumbar patch corresponding to each individual lumbar from an image, (c5) rotating each individual lumbar patch based on the rotation angle, and (c6) the individual lumbar patch is classified into the disease. The method may include inputting a model to diagnose whether each individual lumbar patch has a disease.

In addition, step (c) is performed between step (c1) and step (c2), generating a confidence map using the segmentation image, and removing segments other than the lumbar spine based on the confidence map. It may further include.

In the step (c2), a convergence map algorithm may be applied to extract the center point.

In the step (c3), a connection line connecting the center points of adjacent individual lumbar vertebrae may be generated, and an angle formed by a virtual line orthogonal to the connection line and a horizontal direction may be extracted as the rotation angle.

According to the configuration as described above, according to the present invention, by using the deep learning technology in two of the diagnostic process, that is, the segmentation process and classification process, by ensuring the accuracy of segmentation and disease classification accuracy, Can increase the accuracy.

In addition, by extracting the center point of the individual lumbar spine, and using it to determine the rotation angle, learning with the learning data of the same rotation state, and through the rotation of the extracted individual lumbar spine to increase the accuracy of the diagnosis results by diagnosis under the same conditions It becomes possible.

1 is a view for explaining a deep learning-based lumbar disease secondary diagnostic method according to the present invention,
2 and 3 is a diagram for explaining a method for generating a diagnostic model in a deep learning-based lumbar disease secondary diagnostic method according to the present invention,
4 and 5 are views showing examples of a normal lumbar patch and a disease lumbar patch used as training data for deep learning in a deep learning based lumbar disease assisted diagnostic method,
6 to 9 are diagrams for explaining a diagnostic process of the deep learning-based lumbar disease auxiliary diagnostic method according to the present invention.

As the inventive concept allows for various changes and numerous embodiments, particular embodiments will be illustrated in the drawings and described in detail in the written description.

However, this is not intended to limit the present invention to specific embodiments, it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. Singular expressions include plural expressions unless the context clearly indicates otherwise. In this application, the terms "comprise" or "have" are intended to indicate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, and one or more other features. It is to be understood that the present invention does not exclude the possibility of the presence or the addition of numbers, steps, operations, components, components, or a combination thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art. Terms such as those defined in the commonly used dictionaries should be construed as having meanings consistent with the meanings in the context of the related art and shall not be construed in ideal or excessively formal meanings unless expressly defined in this application. Do not.

Hereinafter, with reference to the accompanying drawings will be described embodiments of the present invention;

1 is a view for explaining a deep learning-based lumbar disease secondary diagnostic method according to the present invention. Referring to Figure 1 in the lumbar disease auxiliary diagnostic method according to the present invention, first, when the training data is input (S10), the lumbar disease disease is deep learning using the training data (S20). The diagnostic model is generated as a result of the deep learning (S30).

Referring to FIGS. 2 and 3, a deep learning process in a method of assisting lumbar spine disease according to the present invention will be described. In the method of assisting lumbar disease in accordance with the present invention, a segmentation model 110 and a disease classification model ( 120 is generated through deep learning.

First, the segmentation model 110 will be described. As shown in FIG. 2, a plurality of lumbar lumbar images and a segmentation image obtained by segmenting individual lumbar segments for each lumbar lumbar image are input as first training data. The upper left image of FIG. 2 is a lumbar spine image, and the lower left is an intensitymentment image for the lumbar spine image.

When a plurality of lumbar images and segmentation images for each lumbar image are input, the segmentation learning unit 10 proceeds with deep learning using the lumbar image and the segmentation image as learning data using the segmentation image as output data. When the diagnosis target lumbar image is input in the diagnosis process to be described later through the deep learning process, the segmentation model 110 to perform the segmentation is generated, which will be described in detail later.

Meanwhile, the disease classification model 120 is generated through deep learning using the plurality of normal lumbar patches and the plurality of disease lumbar patches as second learning data. FIG. 4 is a diagram showing examples of diseased lumbar patches. An example of an image of an individual lumbar spine having a compression fracture is shown. FIG. 5 is a diagram illustrating examples of a normal lumbar patch.

As such, when a plurality of normal lumbar patches and a plurality of disease lumbar patches are input as the second learning data, the classification learning unit 20 performs deep learning through a pre-registered classification algorithm. A disease classification model 120 is generated using the patch as an input.

As described above, in the present invention, the segmentation model 110 to be performed in the segmentation process of the diagnosis process and the lumbar patch are classified as inputs, that is, the disease classification model 120 to be diagnosed through deep learning is generated. The more accurate diagnosis is possible in the course of diagnosis, the description thereof will be described later.

Referring back to FIG. 1, when the diagnostic model is generated through deep learning as described above, an actual diagnosis process is performed. When the diagnosis target lumbar image is input (S40), whether the lumbar disease of the diagnosis target lumbar image is diagnosed based on the segmentation model 110 and the disease classification model 120 (S50).

Specifically, referring to FIG. 6, the segmentation process is performed (S52). As shown in (a) of FIG. 7 prior to the segmentation process, if a diagnosis target lumbar image having a different size from the lumbar image used as the training data is input, a resizing process is performed to correspond to the lumbar image used as the training data. (S51), as shown in Fig. 7B, a resized image is generated.

When the resizing of the lumbar image to be diagnosed is completed, the segmentation process is performed as described above. Here, the segmentation process of the diagnosis target lumbar image is a segmentation corresponding to an individual lumbar spine, as shown in (c) of FIG. 7, from the diagnosis target lumbar image through a segmentation model 110 generated through deep learning. Create an image.

When the segmentation image is generated as described above, a noise removing process for an area except the lumbar region may be performed. FIG. 7D is a diagram illustrating an example of segmentation of the lumbar spine in regions other than the lumbar region, and as shown in FIG. 7E, a confidence map is generated (S54) and illustrated in FIG. 7F. As shown, the actual lumbar region is analyzed by removing segments other than the lumbar spine (S55).

Then, the center point of the individual lumbar spine is extracted from the lumbar spine image analyzed using the segmentation image (S56). In the present invention, an example in which a confidence map algorithm is applied to extracting a center point of an individual lumbar spine is illustrated. FIG. 8A illustrates an example of a center point of an individual lumbar spine extracted through a confidence map algorithm.

When the center point of the individual lumbar spine is extracted, the rotation angle of the individual lumbar spine is extracted using the center point of the individual lumbar spine (S57). Extraction of the rotation angle using the center point generates a connection line connecting the center points of the adjacent individual lumbar spine, and extracts the angle formed by the crosswise orthogonal connection line and the horizontal direction as the rotation angle.

Then, based on the segmentation imager and the center point, as shown in (d) of FIG. 8, individual lumbar patches corresponding to each individual lumbar spine are extracted from the lumbar image to be diagnosed (S58). 8E is an enlarged view of the individual lumbar spine, and individual lumbar patches are extracted as shown in FIG. 8F.

At this time, as described above, each individual lumbar patch may be inclined in the horizontal direction when it is extracted, and rotates the individual lumbar patch based on the rotation angle extracted for each individual lumbar spine described above. The individual lumbar patch is input to the disease classification model 120 and classified according to the disease (S59).

Then, as shown in Figure 1, whether the disease of each individual lumbar spine determined according to the classification of the disease classification model 120, that is, the diagnosis result is output (S60). FIG. 9 is a diagram illustrating an example of output of a diagnosis result. FIG. 9A is a diagram showing numbers of individual lumbar vertebrae based on segmentation images and center points, and FIG. 9B is a diagnosis result of disease. It is a diagram showing an example in which a red box is displayed on the lumbar spine, and a green box is displayed on the normal lumbar spine. 9 (c) is a diagram illustrating an example of displaying a result in a lumbar spine image together with a description of a diagnosis result.

As described above, by using deep learning techniques in two processes of the diagnosis process, that is, the segmentation process and the classification process, the accuracy of the segmentation and the disease classification can be secured, thereby increasing the accuracy of the overall diagnosis.

In addition, by extracting the center point of the individual lumbar spine, and using it to determine the rotation angle, learning with the learning data of the same rotation state, and through the rotation of the extracted individual lumbar spine to increase the accuracy of the diagnosis results by diagnosis under the same conditions It becomes possible.

Although some embodiments of the invention have been shown and described, it will be apparent to those skilled in the art that the embodiments may be modified without departing from the spirit or spirit of the invention. . It is intended that the scope of the invention be defined by the claims appended hereto and their equivalents.

10: segmentation learning unit 20: classification learning unit
110: segmentation model 120: disease classification model

Claims (5)

In the deep learning based diagnostic method for lumbar spine disease,
(a) deep learning-based information processing apparatus generating a diagnostic model by deep learning lumbar disease using learning data;
(b) inputting a diagnosis lumbar image to the diagnostic model;
(c) diagnosing the lumbar disease of the lumbar image to be diagnosed by the diagnostic model;
Step (a) performed by the information processing apparatus
(a1) inputting a plurality of lumbar vertebrae images and segmentation images segmenting individual lumbar vertebrae for each of the lumbar vertebral images as first training data;
(a2) deep learning the first training data using the segmentation image as output data;
(a3) generating a segmentation model through deep learning using the first training data as an input;
(a4) inputting a plurality of normal lumbar patches and a plurality of disease lumbar patches as second learning data;
(a5) deep learning the second learning data through a pre-registered classification algorithm;
(a6) generating a disease classification model through deep learning using the second learning data as an input;
In the step (c), the segmentation model and the disease classification model are applied to the diagnostic model;
Step (c) performed by the diagnostic model
(c1) generating a segmentation image corresponding to an individual lumbar spine from the diagnosis lumbar image through the segmentation model;
(c2) extracting a center point of each lumbar spine from the lumbar image to be diagnosed using the segmentation image;
(c3) extracting the rotation angle of the individual lumbar vertebra based on the center point of the individual lumbar vertebra,
(c4) extracting an individual lumbar patch corresponding to each individual lumbar from the diagnosis lumbar image based on the segmentation image and the center point;
(c5) each said lumbar patch is rotated based on said angle of rotation,
(c6) deep learning-based lumbar disease assisted diagnosis method comprising the step of inputting the individual lumbar patch into the disease classification model to diagnose whether each individual lumbar patch.
delete The method of claim 1,
Step (c) is
It is performed between the step (c1) and the step (c2), and further comprising the step of generating a confidence map using the segmentation image, and removing segments other than the lumbar spine based on the confidence map. Deep learning based diagnostic method for lumbar spine disease.
The method of claim 1,
In the step (c2), deep learning-based lumbar disease assisted diagnostic method, characterized in that the center point is extracted by applying a confidence map algorithm.
The method of claim 1,
In the step (c3), the deep learning-based lumbar disease, characterized in that for generating a connection line connecting the center point of the adjacent individual lumbar spine, the angle formed by the imaginary line orthogonal to the connecting line and the horizontal direction as the rotation angle Assisted diagnostic method.

Images