KR20220008126A - Method and apparatus for detecting dysphagia automatically based on machine learning - Google Patents

Abstract

The present invention provides a method and an apparatus for automatically detecting dysphagia based on machine learning which can improve determination accuracy and reduce determination time. According to one embodiment of the present invention, the method for automatically detecting dysphagia based on machine learning comprises: a step of receiving image data including a cervical vertebra; a step of inputting the image data into a first machine learning unit to recognize the position of the cervical vertebra, and generating a plurality of reasonable on interest (ROI) images for a plurality of frames in the image data; a step of inputting the plurality of ROI images into a second machine learning unit to identify dysphagia occurrence status in the plurality of ROI images; and a step of determining that a patient corresponding to the image data has dysphagia if the number of frames identified to have dysphagia among prescribed continuous frames included in the plurality of frames is higher than or equal to a first threshold value.

Description

Machine learning-based automatic detection method and device for dysphagia

The present invention relates to a method and apparatus for automatic detection of dysphagia based on machine learning.

Dysphagia is a condition in which food passes from the mouth to the esophagus, causing problems in the passage of food. It is important to detect dysphagia early because it can cause aspiration of food into the airways, which can lead to pneumonia and suffocation.

Meanwhile, in the prior art, in order to determine whether a patient has dysphagia, it is required for a medical practitioner to read image data of a patient swallowing food. In the case of a method in which medical personnel directly view and read image data, there is a disadvantage in that the reading accuracy varies depending on the medical personnel and the reading time is long. Therefore, to solve this problem, research on automatic detection of dysphagia using machine learning is in progress.

Republic of Korea Patent Publication No. 10-2094828 (2020. 4. 27)

The present invention for solving the problems as described above uses the first machine learning unit to recognize the cervical vertebrae, then generates an ROI image including the cervical vertebrae, and uses the second machine learning unit to determine whether swallowing disorder is present in the ROI image. An object of the present invention is to provide a method and apparatus for automatic detection of dysphagia based on machine learning.

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

A method for automatically detecting swallowing disorders based on machine learning according to an embodiment of the present invention for solving the above-described problems includes the steps of receiving image data including cervical vertebrae, and inputting the image data to the first machine learning unit to determine the position of the cervical vertebrae. Recognizing and generating a plurality of ROI (reasonable on interest) images for a plurality of frames in the image data, inputting the plurality of ROI images to a second machine learning unit to determine whether swallowing disorders occur in the plurality of ROI images and determining that the patient corresponding to the image data has dysphagia when the number of frames identified as having dysphagia among predetermined continuous frames included in the plurality of frames is equal to or greater than a first threshold.

In addition, the machine learning-based automatic swallowing disorder detection apparatus according to an embodiment of the present invention for solving the above-described problems is an image data receiving unit for receiving image data including cervical spine, and inputting the image data to the first machine learning unit. A dynamic ROI generator that recognizes the position of the cervical spine and generates a plurality of ROI (reasonable on interest) images for a plurality of frames in the image data and a plurality of ROI images are input to the second machine learning unit, and a plurality of ROI images If the number of frames identified as having dysphagia among predetermined continuous frames included in the plurality of frames is greater than or equal to the first threshold, it is determined that the patient corresponding to the image data has dysphagia. and a swallowing disorder judgment unit.

In addition to this, a program stored in a medium may be further provided to execute the method for implementing the present invention in combination with a computer which is hardware.

In addition, another method for implementing the present invention, another system, and a computer readable recording medium recording a computer program for executing the method may be further provided.

According to the present invention as described above, when the image data is received, the presence or absence of a swallowing disorder can be detected without the intervention of a medical professional, so that the accuracy of judgment can be improved and the reading time can be reduced.

In addition, since the method according to an embodiment of the present invention includes the process of generating an ROI image from image data to detect the presence or absence of a dysphagia, the accuracy can be improved compared to reading the presence or absence of a dysphagia directly from the image data. can have an effect.

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

1 is a diagram for explaining a method for automatically detecting dysphagia based on machine learning according to an embodiment of the present invention.
2 is a flowchart illustrating a method for automatically detecting dysphagia based on machine learning according to an embodiment of the present invention.
3 is a diagram for explaining pre-processing of image data according to an embodiment of the present invention.
4 is a view for explaining the cervical spine and ROI image recognized by the dynamic ROI generator according to an embodiment of the present invention.
5 is a diagram for explaining a method of generating an ROI image according to an embodiment of the present invention.
6 is a diagram for explaining a method of recognizing a cervical spine by a dynamic ROI generator according to an embodiment of the present invention.
7 to 9 are diagrams for explaining the experimental results of the automatic detection method for dysphagia based on machine learning according to an embodiment of the present invention.
10 is a block diagram for explaining an apparatus for automatic detection of dysphagia based on machine learning according to an embodiment of the present invention.

Advantages and features of the present invention and methods of achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but may be implemented in various different forms, and only the present embodiments allow the disclosure of the present invention to be complete, and those of ordinary skill in the art to which the present invention pertains. It is provided to fully understand the scope of the present invention to those skilled in the art, and the present invention is only defined by the scope of the claims.

The terminology used herein is for the purpose of describing the embodiments and is not intended to limit the present invention. In this specification, the singular also includes the plural unless specifically stated otherwise in the phrase. As used herein, “comprises” and/or “comprising” does not exclude the presence or addition of one or more other components in addition to the stated components. Like reference numerals refer to like elements throughout, and "and/or" includes each and every combination of one or more of the recited elements. Although "first", "second", etc. are used to describe various elements, these elements are not limited by these terms, of course. These terms are only used to distinguish one component from another. Accordingly, it goes without saying that the first component mentioned below may be the second component within the spirit of the present invention.

Unless otherwise defined, all terms (including technical and scientific terms) used herein will have the meaning commonly understood by those of ordinary skill in the art to which this invention belongs. In addition, terms defined in a commonly used dictionary are not to be interpreted ideally or excessively unless specifically defined explicitly.

Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Before the description, the meaning of the terms used in this specification will be briefly described. However, it should be noted that, since the description of the term is for the purpose of helping the understanding of the present specification, it is not used in the meaning of limiting the technical idea of the present invention unless explicitly described as limiting the present invention.

1 is a diagram for explaining a method for automatically detecting dysphagia based on machine learning according to an embodiment of the present invention.

The occurrence of dysphagia is judged according to whether or not the food moves into the airway rather than the esophagus when the patient swallows it. Since it is impossible to predict when dysphagia will occur for each patient, the method according to an embodiment of the present invention first determines whether dysphagia has occurred in each frame from the image data captured by the patient swallowing food, and then based on the result As a result, context information is used to specify the point in time when a swallowing disorder occurs in the corresponding image data.

Referring to FIG. 1 , a method and apparatus for automatically detecting dysphagia based on machine learning according to an embodiment of the present invention may receive image data of a patient swallowing food. 1 and the following drawings, image data is shown as x-ray image data, but the type of image data is not limited thereto.

Thereafter, image preprocessing may be performed to improve the efficiency of machine learning. The pre-processing of the image data may include adjusting pixel brightness and cropping the image.

The pre-processed image data is input to the Dynamic ROI Creator, which recognizes the cervical vertebrae within a plurality of frames of image data and then generates a plurality of ROI images. have. Referring to FIG. 1 , a red region 110 is a region recognized as a cervical spine by the method according to an embodiment of the present invention, and a green region 120 denotes a region to be included in the ROI image.

Thereafter, the plurality of ROI images may be input to the Airway Invasion Detection, and the Airway Invasion Detection may automatically detect the presence or absence of a swallowing disorder.

2 is a flowchart illustrating a method for automatically detecting dysphagia based on machine learning according to an embodiment of the present invention.

The machine learning-based automatic detection method for dysphagia according to an embodiment of the present invention may receive image data including cervical vertebrae in S210.

Meanwhile, the method according to an embodiment of the present invention may further include pre-processing the image data after receiving the image data including the cervical spine. Specifically, the pre-processing of the image data may include at least one of uniformizing pixel brightness of the image data and cropping the image data to the same size.

In S220, image data is input to the first machine learning unit to recognize the position of the cervical vertebrae, and a plurality of ROI (reasonable on interest) images may be generated for a plurality of frames in the image data based on the position of the cervical vertebrae. Meanwhile, the ROI image may be dynamically generated for each of a plurality of frames.

In S230, by inputting the plurality of ROI images to the second machine learning unit, it is possible to identify whether swallowing disorders occur in the plurality of ROI images.

When the number of frames identified as having dysphagia among predetermined continuous frames included in the plurality of frames in S240 is equal to or greater than the first threshold, it may be determined that the patient corresponding to the image data has dysphagia.

S240 may include a step of determining whether the swallowing disorder occurs at the number of throat swallowing in the image data. In addition, S240 may include determining how many times the dysphagia has occurred in the image data.

On the other hand, the method according to an embodiment of the present invention further comprises the step of, when the frames above the first threshold value identified as having dysphagia are not continuous, the frame existing between the frames identified as having dysphagia as having occurred is post-selecting further comprising: can do. Here, the step of performing the post-correction may be to use a median filter.

3 is a diagram for explaining pre-processing of image data according to an embodiment of the present invention.

In order to determine whether or not a swallowing disorder is present for each patient, the time point at which the image was taken and the shooting environment may be different. Furthermore, the corresponding image data may have a tendency to be concentrated in a specific area within the dynamic range of the pixel brightness distribution, so it may be difficult to perform machine learning.

Accordingly, the method according to an embodiment of the present invention may perform preprocessing of the received image data so that the convolution filter can be effectively learned during machine learning by matching the conditions of each image data. Specifically, the type of preprocessing may be at least one of uniform pixel brightness and cropping of image size. Here, if all the image data is cropped to the same size, there is an effect of increasing the efficiency of the ROI detection process.

3 is a diagram for explaining a result of preprocessing image data using Contrast Limited Adaptive Histogram Equalization (CLAHE). CLAHE divides the image into small blocks of a certain size and uniformizes the histogram of the pixel brightness for each block, thereby uniformizing the pixel brightness for the entire image.

Fig. 3 (a) is the original image data, and Fig. 3 (b) shows the pre-processing result. Referring to (b) of FIG. 3 , it can be seen that the contrast around the chin and cervical vertebrae is significantly improved through the CLAHE process, so that the structural characteristics of the cervical vertebrae are prominently displayed.

4 is a view for explaining the cervical spine and ROI image recognized by the dynamic ROI generator according to an embodiment of the present invention.

Figure 4 (a) shows the ground truth required for the dynamic ROI generator to recognize the cervical spine included in the image data, the cervical region is shown in red. 4 (b) is an image showing the cervical vertebra recognized by the dynamic ROI generator in blue, and FIG. 4 (c) shows an ROI image generated based on the recognized cervical vertebrae region.

Since dysphagia occurs near the larynx due to the structure of the throat, regions other than around the neck in the image data are unnecessary for diagnosing dysphagia. Rather, the complex texture patterns seen in the skull and cervical vertebrae as shown in (a) and (b) of Figure 4 may cause difficulties in analyzing the movement path of food. Therefore, the method according to an embodiment of the present invention may generate ROI data for the detection of dysphagia.

Referring to FIG. 3 , food swallowed by a patient is generally found in the anterior portion of the cervical spine. Accordingly, the method according to an embodiment of the present invention may select the larynx and its vicinity as an ROI for effective dysphagia detection.

In addition, since the head and neck move whenever the patient swallows food while participating in a videofluoroscopic swallowing study (VFSS), the present invention can generate a dynamic ROI that adaptively moves according to the movement of the patient. To this end, the method according to an embodiment of the present invention may find a cervical vertebra showing structurally prominent features for each image data and use it as a reference point for ROI setting.

The machine learning-based automatic swallowing disorder detection apparatus implemented according to an embodiment of the present invention randomly selects 157 image data corresponding to about 50% of the total 322 image data (video files) for learning of the first machine learning unit. The training data set can be configured with a total of 157 images by selecting and selecting an image in which the cervical spine is clearly visible from each image data. On the other hand, the test set can be composed of a total of 165 sheets by selecting one from the remaining video files that are not used for the training data, and for the learning of the first machine learning unit, the cervical spine as shown in the red region of Fig. 4 (a). The pixels of the region may be labeled, but the number of specifically used image data is not limited thereto.

5 is a diagram for explaining a method of generating an ROI image according to an embodiment of the present invention.

The method according to an embodiment of the present invention may generate a rectangular ROI image including the larynx, cervical vertebrae, and adjacent regions based on the position measurement result of the cervical vertebrae. When generating the ROI image, the method according to an embodiment of the present invention may be used as the center coordinates of pixels corresponding to the cervical spine.

The red regions shown in FIGS. 5 (b) and (c) mean pixels classified as cervical vertebrae. When the number of pixels classified as cervical vertebra is N, the coordinates of the reference point of the ROI shown in FIG. 5(c) are the same as in Equation 1.

[Equation 1]

를 기준으로 휴리스틱하게(heuristically) 결정할 수 있다.Here, x _i and y _i mean the coordinates of each pixel when the upper left corner of the image data of FIG. 5 (a) is taken as the origin. On the other hand, the lowermost part of the ROI image is the same as the lowermost part of the image data, and the size of the ROI image is

can be determined heuristically.

The green area shown in FIG. 5C is an ROI image generated by the method according to an embodiment of the present invention, and FIG. 5D shows that the ROI image is resized to a size of 299x299.

The ROI image shown in (d) of FIG. 5 may be input to the dysphagia determination unit. For the determination of dysphagia, the method according to an embodiment of the present invention may utilize a second machine learning unit based on a Convolutional Neural Network (CNN). The dysphagia determination unit may determine whether the food normally passes through the esophagus or whether a dysphagia has occurred based on the position and shape of the food in the ROI image.

6 is a diagram for explaining a method of recognizing a cervical spine by a dynamic ROI generator according to an embodiment of the present invention.

According to an embodiment, U-Net, which is a machine learning model used for segmentation of a biomedical image, may be used to find the position of the cervical spine in the image data. U-Net is a Fully Connected Network (FCN)-based model and has the advantage of being able to learn a network with only a small amount of data.

Meanwhile, in order to effectively learn a network with a small amount of learning data, the method according to an embodiment of the present invention may use a transfer learning technique. 6 shows the structure of a U-Net used to implement a method according to an embodiment of the present invention. In order to implement the method according to an embodiment of the present invention, the encoder consists of 11 convolutional layers, and the decoder uses 4 convolutional layers.

Referring to FIG. 6 , the received image data with a size of 224x224 is passed through a convolutional encoder to generate a feature with a size of 14x14. The feature finds the cervical vertebrae in the image data reduced to 112x112 size while passing through the convolutional encoder. In addition, the U-Net structure shown in FIG. 6 may transmit features extracted from the convolutional encoder to the convolutional decoder.

7 to 9 are diagrams for explaining the experimental results of the automatic detection method for dysphagia based on machine learning according to an embodiment of the present invention.

The method according to an embodiment of the present invention may identify whether or not swallowing disorders have occurred in a plurality of ROI images and output them as binary values. For example, if the ROI image belongs to a positive class in which it is determined that dysphagia has occurred, the method according to an embodiment may output a binary value of 1. In addition, if the ROI image belongs to a negative class in which it is determined that swallowing disorder does not occur, a binary value of 0 may be output.

7 is a view for explaining a class used in the second machine learning unit to read dysphagia. Specifically, as shown in (a) of FIG. 7 , the image determined to have dysphagia is classified as a positive class, and in FIG.

On the other hand, the method according to an embodiment of the present invention can be implemented using a CNN using the Xception architecture. In order to test the performance of the method according to an embodiment of the present invention, images corresponding to 70% of each of 16,062 positive class images and 269,579 negative class images included in 322 video files were used as training data. was used as In addition, 10% of the images were used for verification of the learned network, and the remaining 20% of images were used as test data for evaluating the performance of the method according to an embodiment of the present invention. Table 1 shows the total number of images used in the experiment and the number of image data for each class used in Training, Validation, and Test.

[Table 1]

Since dysphagia occurs for a short period of time in the entire image data, the number of negative samples in Table 1 is much higher than that of the positive samples. In order to alleviate the problem of imbalance in the amount of data between these classes, a focal loss function may be used.

In the case of focal loss, loss update is almost impossible by giving a small loss to a class that can be classified relatively frequently due to its large number, and learning for a class that is difficult to classify by increasing the update of the loss for a class that is difficult to classify allow it to be focused.

In addition, both the process of recognizing the cervical spine by the first machine learning unit and the process of detecting the dysphagia by the second machine learning unit may be regarded as a relatively difficult task to find a positive sample compared to finding a negative sample. Accordingly, the method and apparatus according to an embodiment may learn a network using a focus loss instead of a cross entropy loss.

8 shows the classification results of CNN for each image data of patients participating in the experiment. The horizontal axis of each graph of FIG. 8 lists the image data in the order of shooting time, and the vertical axis is the class classification result for each image data. When it is determined that the corresponding image data corresponds to the positive class, 1 is mapped, and when it is determined that the corresponding image data corresponds to the negative class, 0 is mapped.

Fig. 8 (a) shows the experimental results for a normal person without dysphagia, and Fig. 8 (b) shows the experimental results for a patient with dysphagia. 8 (c) and (d) show the results of post-correction of the experimental results of the normal person and the patient through the median filter, respectively.

Referring to (a) and (b) of Figure 8, it can be determined that even a normal person has instantaneous dysphagia, and even in a patient with dysphagia, it can be determined that there is no instantaneous dysphagia in any frame. However, since swallowing food is a continuous motion, the images at the point of occurrence of dysphagia should be continuously classified as positive samples on the time axis. In order to reflect the temporal context information of the image data in the final determination of whether or not swallowing disorders occur in the image data, the method according to an embodiment of the present invention may perform post-correction using a median filter. .

For example, if the image data is image data shot at 24 fps, the positive sample 810 in the impulse form shown in FIG. it may be reasonable Accordingly, in the method according to an embodiment of the present invention, a positive sample may be post-corrected to a negative sample as shown in FIG. 8C using a median filter.

In addition, the method according to an embodiment of the present invention may determine that dysphagia has occurred in the corresponding section only when at least two or more of the seven consecutive images are classified as positive samples. To this end, a convolution operation can be performed on a vector containing the classification result of CNN for image data and a filter of length 7 in which all components have a value of 1. Fig. 8(d) shows the result of performing such a post-correction.

Meanwhile, Table 2 shows the cervical spine detection results using U-Net.

[Table 2]

The performance was evaluated for 39 images not used for U-Net training, and the figures in Table 2 are the results measured at the pixel level. For example, recall refers to the proportion of pixels in the ground truth of the cervical spine (pixels labeled as cervical by humans) predicted by the model to be cervical. Referring to Table 2, it can be seen that U-Net exhibits a recall rate of 90.4%, an accuracy of 99.6%, and a precision of 85.6% for the cervical spine.

Table 3 and FIG. 9 show the image data classification results and ROC curves based on the Xception architecture, respectively.

[Table 3]

The recall rate in Table 3 means the ratio of the image data detected that dysphagia also occurred in the method according to an embodiment of the present invention among the image data read that dysphagia occurred by two medical personnel. Referring to Table 3, it can be seen that a recall rate of 90.4%, an accuracy of 99.6%, and a precision of 85.6% are shown.

In addition, referring to FIG. 9 , as a result of performing the method according to an embodiment of the present invention, a false positive rate (FPR) value indicating a ratio for a case in which a negative sample is incorrectly identified as a positive sample is close to 0, It can be seen that the TPR (True Positive Rate) value, which is the rate at which a positive sample is properly identified as a positive sample, is close to 1.

Table 4 shows the final detection result of dysphagia and the ROC curve after the results of identifying whether or not dysphagia has occurred by using a convolution filter to reflect context information.

[Table 4]

Referring to Table 4, it can be seen that the recall rate is 83.3%, the accuracy is 93.9%, and the precision is 98.0%.

The machine learning-based automatic swallowing disorder detection method according to the embodiment of the present invention described above differs from the machine learning-based automatic swallowing disorder detection method described with reference to FIGS. Examples are omitted.

10 is a block diagram illustrating an apparatus for automatically detecting a dysphagia based on machine learning according to an embodiment of the present invention.

The machine learning-based automatic swallowing disorder detection apparatus 1000 according to an embodiment of the present invention may include an image data receiver 1010 , a dynamic ROI generator 1020 , and a swallowing disorder determiner 1030 .

The image data receiver 1010 may receive image data including cervical vertebrae.

Meanwhile, the image data receiver 1010 may include an image preprocessor that pre-processes the image data. In this case, the image preprocessor may perform at least one of uniform pixel brightness of image data and cropping image data to the same size.

The dynamic ROI generator 1020 inputs the image data received by the image data receiver 1010 to the first machine learning unit to recognize the position of the cervical vertebrae, and a plurality of frames in the image data based on the position of the cervical vertebrae. It is possible to generate a plurality of ROI (reasonable on interest) images with respect to the .

The dysphagia determination unit 1030 inputs the plurality of ROI images to the second machine learning unit, identifies whether or not swallowing disorders occur in the plurality of ROI images, and determines whether dysphagia occurs among predetermined continuous frames included in the plurality of frames. When the number of frames identified as occurring is equal to or greater than the first threshold, it may be determined that the patient corresponding to the image data has dysphagia.

On the other hand, the dysphagia determination unit 1030 may postulate that if the frames above the first threshold value identified as having the dysphagia are not continuous, frames existing between the frames identified as having the dysphagia have occurred. In this case, the dysphagia determination unit 1030 may use a median filter for the post-correction.

In addition, the swallowing disorder determination unit 1030 may determine at which number of throat swallows in the image data the swallowing disorder occurs.

Also, the swallowing disorder determination unit 1030 may determine how many times the swallowing disorder occurs in the image data.

The method according to an embodiment of the present invention described above may be implemented as a program (or application) to be executed in combination with a server, which is hardware, and stored in a medium.

The above-described program is C, C++, JAVA, machine language, etc. that a processor (CPU) of the computer can read through a device interface of the computer in order for the computer to read the program and execute the methods implemented as a program It may include code (Code) coded in the computer language of Such code may include functional code related to a function defining functions necessary for executing the methods, etc., and includes an execution procedure related control code necessary for the processor of the computer to execute the functions according to a predetermined procedure. can do. In addition, the code may further include additional information necessary for the processor of the computer to execute the functions or code related to memory reference for which location (address address) in the internal or external memory of the computer to be referenced. have. In addition, when the processor of the computer needs to communicate with any other computer or server located remotely in order to execute the above functions, the code uses the communication module of the computer to determine how to communicate with any other computer or server remotely. It may further include a communication-related code for whether to communicate and what information or media to transmit and receive during communication.

The storage medium is not a medium that stores data for a short moment, such as a register, a cache, a memory, etc., but a medium that stores data semi-permanently and can be read by a device. Specifically, examples of the storage medium include, but are not limited to, ROM, RAM, CD-ROM, magnetic tape, floppy disk, and optical data storage device. That is, the program may be stored in various recording media on various servers that the computer can access or in various recording media on the user's computer. In addition, the medium may be distributed in a computer system connected to a network, and a computer-readable code may be stored in a distributed manner.

The steps of a method or algorithm described in connection with an embodiment of the present invention may be implemented directly in hardware, as a software module executed by hardware, or by a combination thereof. A software module may include random access memory (RAM), read only memory (ROM), erasable programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, hard disk, removable disk, CD-ROM, or It may reside in any type of computer-readable recording medium well known in the art to which the present invention pertains.

As mentioned above, although embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art to which the present invention pertains know that the present invention may be embodied in other specific forms without changing the technical spirit or essential features thereof. you will be able to understand Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive.

1000: machine learning-based automatic detection device for dysphagia
1010: image data receiving unit
1020: dynamic ROI generator
1030: swallowing disorder judgment unit

Claims (13)

Receiving image data including cervical spine;
inputting the image data to a first machine learning unit to recognize a position of the cervical vertebrae, and generating a plurality of ROI (reasonable on interest) images for a plurality of frames in the image data based on the position of the cervical vertebra;
inputting the plurality of ROI images to a second machine learning unit to identify whether swallowing disorders occur in the plurality of ROI images; and
determining that the patient corresponding to the image data has dysphagia if the number of frames identified as having dysphagia is greater than or equal to a first threshold among predetermined continuous frames included in the plurality of frames;
Including, machine learning-based automatic detection method of dysphagia. According to claim 1,
If the frames above the first threshold value identified as having dysphagia are not continuous,
preliminarily setting a frame existing between the frames identified as having dysphagia as having dysphagia;
Further comprising, a machine learning-based automatic detection method of dysphagia. 3. The method of claim 2,
The step of pre-correction is to use a median filter, a machine learning-based automatic detection method for dysphagia. According to claim 1,
The steps to determine that you have dysphagia are:
determining at which number of swallows in the image data a swallowing disorder occurred;
Including, machine learning-based automatic detection method of dysphagia. According to claim 1,
The steps to determine that you have dysphagia are:
determining how many times the dysphagia has occurred in the image data;
Including, machine learning-based automatic detection method of dysphagia. According to claim 1,
After receiving the image data including the cervical vertebrae, comprising the step of pre-processing the image data,
The pre-processing of the image data comprises at least one of equalizing the pixel brightness of the image data and cropping the image data to the same size, machine learning-based automatic detection method for dysphagia . An image data receiving unit for receiving image data including cervical vertebrae;
Dynamic, which inputs the image data to a first machine learning unit to recognize the position of the cervical vertebrae, and generates a plurality of reasonable on interest (ROI) images for a plurality of frames in the image data based on the position of the cervical vertebrae ROI generator; and
By inputting the plurality of ROI images to a second machine learning unit, it is identified whether or not swallowing disorders have occurred in the plurality of ROI images, and frames identified as having dysphagia among predetermined continuous frames included in the plurality of frames are identified. a dysphagia determination unit that determines that a patient corresponding to the image data has a dysphagia when the number of is greater than or equal to a first threshold;
Including, machine learning-based automatic detection of dysphagia. 8. The method of claim 7,
The swallowing disorder determination unit
When the frames above the first threshold value identified as having dysphagia are not continuous, the frame existing between the frames identified as having dysphagia is pre-selected as having dysphagia, a machine learning-based automatic detection device for dysphagia. 9. The method of claim 8,
A machine learning-based automatic detection device for dysphagia that uses a median filter for the post-correction. 8. The method of claim 7,
The swallowing disorder determination unit
A machine learning-based automatic swallowing disorder detection device that determines whether a swallowing disorder occurs in the number of throat swallows in the image data. 8. The method of claim 7,
The swallowing disorder determination unit
To determine how many times the dysphagia has occurred in the image data, a machine learning-based automatic detection of dysphagia. 8. The method of claim 7,
The image data receiving unit includes an image pre-processing unit for pre-processing the image data,
The image preprocessor, machine learning-based automatic detection of dysphagia, which performs at least one of equalizing the pixel brightness of the image data and cropping the image data to the same size. A program that is combined with a computer which is hardware and is stored in a medium for executing the method of any one of claims 1 to 6.

Images