KR20230097692A - Middle ear disease classification method using video pneumatic otoscopy and analysis apparatus - Google Patents

Abstract

A method of evaluating the middle ear condition using an air otoscopy image includes receiving an air otoscopy image of a subject by an analysis device, inputting the image to a deep learning model trained in advance by the analysis device, and analyzing the image. Evaluating, by a device, whether or not the subject has conductive hearing loss based on an output value output from the deep learning model. The image includes a maximum positive pressure image and a maximum negative pressure image, and the analysis device inputs the maximum positive pressure image and the maximum negative pressure image to the deep learning model, respectively.

Description

Middle ear evaluation method and analysis device using air otoscopy images

The technique described below relates to a technique for automatically evaluating the state of the middle ear using an air otoscopy image.

A video pneumatic otoscope is widely used for otolaryngology patients with hearing loss and the like. A conventional examination method was a method in which a medical professional, who is an expert, looked at a patient's image and examined the current condition and symptoms. Meanwhile, in the field of medical imaging, various approaches for analyzing images using artificial intelligence models have recently been attempted.

Korean Patent Publication No. 10-2013-0110993

The technology to be described below is intended to provide a technique for evaluating the state of the middle ear or hearing loss of a subject using a machine learning model for an image produced by a video air otoscope test.

A method of evaluating the middle ear condition using an air otoscopy image includes receiving an air otoscopy image of a subject by an analysis device, inputting the image to a deep learning model trained in advance by the analysis device, and analyzing the image. Evaluating, by a device, whether or not the subject has conductive hearing loss based on an output value output from the deep learning model.

The analysis device that evaluates the middle ear condition using the air otoscopy image learns to calculate the middle ear condition information for the input device by receiving the input device that receives the air otoscope image of the subject, the maximum positive pressure image and the maximum negative pressure image. A storage device for storing the deep learning model, and the image includes a maximum positive pressure image and a maximum negative pressure image, and the subject is based on an output value calculated by inputting the maximum positive pressure image and the maximum negative pressure image to the deep learning model, respectively. It includes an arithmetic device that evaluates whether or not there is conductive hearing loss.

The technology to be described below calculates an accurate evaluation result for the middle ear state of a subject using an image that can be easily obtained with an otoscopy device. The technology described below can be widely used as a diagnostic aid in primary hospitals using easy-to-use testing equipment.

1 is an example of an otoscopy device for determining a middle ear state.
2 is an example of a system for evaluating a middle ear condition using an image obtained from an otoscopy device.
3 is an example of a model for evaluating a middle ear condition using a VPO image.
4 is another example of a model for evaluating a middle ear condition using a VPO image.
5 is another example of a model for evaluating a middle ear condition using a VPO image.
6 is another example of a model for evaluating a middle ear condition using a VPO image.
7 is a performance verification result of models for evaluating middle ear conditions.
8 is an example of an analysis device for evaluating a middle ear condition using a VPO image.

Since the technology to be described below can have various changes and various embodiments, specific embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the technology described below to specific embodiments, and it should be understood to include all modifications, equivalents, or substitutes included in the spirit and scope of the technology described below.

Terms such as first, second, A, B, etc. may be used to describe various elements, but the elements are not limited by the above terms, and are merely used to distinguish one element from another. used only as For example, without departing from the scope of the technology described below, a first element may be referred to as a second element, and similarly, the second element may be referred to as a first element. The terms and/or include any combination of a plurality of related recited items or any of a plurality of related recited items.

In terms used in this specification, singular expressions should be understood to include plural expressions unless clearly interpreted differently in context, and terms such as “comprising” refer to the described features, numbers, steps, operations, and components. , parts or combinations thereof, but it should be understood that it does not exclude the possibility of the presence or addition of one or more other features or numbers, step-action components, parts or combinations thereof.

Prior to a detailed description of the drawings, it is to be clarified that the classification of components in the present specification is merely a classification for each main function in charge of each component. That is, two or more components to be described below may be combined into one component, or one component may be divided into two or more for each more subdivided function. In addition, each component to be described below may additionally perform some or all of the functions of other components in addition to its main function, and some of the main functions of each component may be performed by other components. Of course, it may be dedicated and performed by .

In addition, in performing a method or method of operation, each process constituting the method may occur in a different order from the specified order unless a specific order is clearly described in context. That is, each process may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

Hereinafter, it will be described that the otoscopy device or analysis device evaluates the middle ear state using a certain machine learning model. An otoscopy device is basically a device that acquires an image of an eardrum according to a pressure state. Hereinafter, an otoscopy device refers to a device that performs video pneumatic otoscopy (VPO).

The VPO can measure the dynamic movement of the tympanic membrane and ossicles as well as middle ear effusion. VPO is a useful method to assess middle ear status (sound transmission). The VPO image is referred to as an image obtained through an otoscope. The VPO image includes the tympanic region.

The analysis device is a device that evaluates the state of the middle ear by analyzing the VPO image acquired by the otoscopy device. The analysis device may be a separate device that analyzes the image acquired by the otoscopy device. In some cases, the analysis device may be a device mounted inside the otoscope inspection device. Therefore, it is assumed that the analysis device evaluates the middle ear condition by analyzing VPO images according to various pressure states. The analysis device may be implemented with various devices capable of processing data. For example, the analysis device may be implemented as a PC, a server on a network, a smart device, a chipset in which a dedicated program is embedded, and the like.

Machine learning models include decision trees, random forests, K-nearest neighbors (KNNs), Naive Bayes, support vector machines (SVMs), and artificial neural networks (ANNs).

ANN is a statistical learning algorithm that mimics biological neural networks. Various neural network models are being studied. A deep learning network (DNN) can model complex non-linear relationships like a general artificial neural network. Various types of DNN models have been studied. For example, there are a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), a Restricted Boltzmann Machine (RBM), a Deep Belief Network (DBN), a Generative Adversarial Network (GAN), and Relation Networks (RL).

1 is an example of an otoscopy apparatus 100 for determining a middle ear state. The otoscopy apparatus 100 of FIG. 1 analyzes VPO images at the time when different pressures are applied with a deep learning model to generate information for evaluating the state of the middle ear.

The otoscope device 100 includes an air injection unit 110, an image measuring unit 120, a light irradiation unit 130, a pressure adjusting unit 140, a control unit 150, and an information processing unit 160. 1 is an example of a configuration requiring explanation for the otoscopic examination device 100, and other mechanical configurations of the otoscopic examination device 100 may include general configurations.

The air injection unit 110 is connected to an air blocking member inserted into the ear canal to inject air into the ear canal. The air blocking member is formed to correspond to the diameter of the ear canal so that it can be inserted into the ear canal. The air injection unit 110 may include a pump communicating with the air blocking member and an air injection pipe communicating with the pump.

The image measuring unit 120 measures an image of the inside of the ear canal transmitted through the path of the air blocking member. Alternatively, the image measuring unit 120 may be configured to measure an image inside the ear canal separately from the path of the air blocking member. The image measuring unit 120 includes an image sensor (CCD) that captures an image transmitted through a certain path.

The image measurement unit 120 may acquire an image of the eardrum according to the air injection state provided by the air injection unit 110 . For example, the image measuring unit 120 may acquire at least one image of a maximum positive pressure image, a maximum negative pressure image, and a static image in which no pressure is applied to the eardrum.

The pressure adjusting unit 140 is connected to the air injection unit 110 and adjusts to provide a constant pressure.

The light irradiation unit 130 is connected to the air blocking member and illuminates the inside of the ear canal by irradiating a light source into the ear canal so that the image measurement unit 120 can acquire an identifiable image.

The control unit 150 controls the image measuring unit 120, the pressure adjusting unit 140, and the light irradiation unit 130. The control unit 150 may control the operation of the image measuring unit 120 and the timing of image acquisition. The control unit 150 may control the pressure adjusting unit 140 so that the air injection unit 110 provides air at a constant pressure. The control unit 150 may control the operation of the light irradiation unit 130 and the intensity (illuminance) of light. The control unit 150 may control the pressure adjusting unit 140, the light irradiation unit 130, and the image measurement unit 120 to obtain images of the eardrum in each of the maximum positive pressure, maximum negative pressure, and stable state.

The information processing unit 160 may analyze pressure state information transmitted through the control unit 150 and images according to each pressure state. The information processing unit 160 corresponds to an analysis device that analyzes images. For example, the information processing unit 160 may evaluate the middle ear state by analyzing the current pressure state and the movement of the eardrum. The information processing unit 160 may evaluate the state of the middle ear by analyzing VPO images in various pressure states using a deep learning model.

1 shows a form in which the information processing unit 160 is mounted in the otoscopic examination device 100 . That is, it is a case where the information processing unit 160 has a chipset form in which a deep learning model and a program are embedded. In this case, the user (medical staff) may check the evaluation result directly through the otoscope examination device 100 or check the evaluation result through a computer device connected to the otoscope examination device 100 .

2 is an example of a system 200 for evaluating a middle ear condition using an image acquired by an otoscope. In FIG. 2 , the analysis device (corresponding to the information processing unit 160 described above) corresponds to an independent device physically different from the otoscopic examination device 210 . 2 illustrates an example in which the analysis device is a computer terminal 130 and a server 140 .

The otoscopy device 110 performs video air otoscopy (VPO) on a subject. The otoscopy device 110 acquires a VPO image according to the pressure state. The otoscopy apparatus 110 may generate VPO images under maximum positive pressure, maximum negative pressure, and normal pressure conditions. Meanwhile, VPO images for each pressure state may be stored in an Electronic Medical Record (EMR) 120.

In FIG. 2 , the user A may use the computer terminal 130 to evaluate the state of the middle ear by analyzing the VPO image for each pressure state of the subject. The computer terminal 130 may calculate an analysis result by inputting the VPO image for each pressure state of the subject to a pre-learned deep learning model. User A may check the analysis result for the subject.

The server 140 may receive VPO images for each pressure state of the subject from the otoscope 110 or the EMR 120 . The server 140 may calculate an analysis result by inputting the VPO image for each pressure state of the subject to a pre-trained deep learning model. The server 140 may transmit the analysis result to the user A's terminal. User A can check the analysis result.

The computer terminal 130 and/or the server 140 may store the analysis result (whether or not there is a hearing loss) of the target person in the EMR 120 .

The above-described deep learning model evaluates the middle ear state using the image produced by the VPO. The deep learning model may calculate a probability value for a middle ear condition. In this case, the middle ear state is information about sound transmission in the middle ear, and may be information about whether or not there is an air-bone gap. The deep learning model may calculate information about whether the difference between the airway and bone conduction is greater than or equal to a predetermined threshold value. That is, the deep learning model may calculate a probability of whether the hearing loss is conductive.

The following describes the process by which the researcher built the deep learning model. The researcher used a CNN-based deep learning model. The researcher constructed each model using Inception-v3, ResNet-50 and VGG-16. Deep learning models are supervised learning-based models and require training data.

The learning data used by the researcher will be explained.

Conductive hearing loss was defined as an average airway-to-bone difference of 0.5dB or more for 0.5kHz, 1kHz, 2kHz, and 4kHz in pure-tone audiometry.

The researcher used the results of the VPO test and pure tone audiometry conducted at the affiliated hospital from 2007 to 2019 for patients aged 18 years or older.

The learning data includes VPO images of conductive hearing loss patients and normal people.

The patient data includes VPO images of a patient with conductive hearing loss. In patients with conductive hearing loss, as a result of otoscope examination or incision examination, middle ear effusion, ossicular fixation, otosclerosis, middle ear mass, congenital cholesteatoma, and adhesive otitis media were found. A patient diagnosed with a disease such as adhesive otitis media. Patient data were used for patients who complained of hearing loss but had no damage to the eardrum. Therefore, images with eardrum perforation were excluded from the patient data. In addition, among the VPO images in the patient data, images in which the vertebrae were not identified or did not visually indicate otitis media were excluded.

Normal person data includes VPO images of a subject with normal hearing and a difference between the airway and bone conduction less than a threshold value.

The researcher used maximum positive pressure images and maximum negative pressure images as input data. The motion of the eardrum can be defined as a point at which the eardrum returns to its original position when positive and negative pressures are given. The maximum positive pressure image is an image of a state in which the eardrum is positioned at the innermost side given the maximum positive pressure, and the maximum negative pressure image is an image in a state in which the eardrum is positioned at the most lateral (outer) side given the maximum negative pressure.

The researchers extracted important frames using the PythonTM image library. The researcher adjusted the image to a size of 200 pixels/inch in the RGB channel and normalized the image to a constant. In addition, in order to increase the amount of learning data, the learning data was augmented 60 times using image rotation.

The researcher built three types of models that receive maximum positive pressure images and maximum negative pressure images as input data and learned the models.

3 is an example of a model for evaluating a middle ear condition using a VPO image. 3 uses a maximum positive pressure image (pos) and a maximum negative pressure image (neg) as input data. The model (Model 1) of FIG. 3 includes a convolution block that extracts features of an input image and a fully connected layer (FC) that receives the extracted features and calculates a final classification result. The model (Model 1) of FIG. 3 includes a first convolution block (conv. 1) and a second convolution block (conv. 2) for the maximum positive pressure image and the maximum negative pressure image, respectively. Each convolution block includes a plurality of convolution layers. The model (Model 1) of FIG. 3 concatenates the feature values calculated in the first convolution block (conv. 1) and the second convolution block (conv. 2), respectively, and inputs them to the heat transfer layer (FC). do. The thermal insulation layer (FC) finally calculates a probability value for whether or not the input image has conductive hearing loss. The analysis device may compare this probability value with a predetermined threshold value to evaluate whether the input image is an image with conductive hearing loss. For example, the analyzer may evaluate that if the output value of model 1 is close to 0, it is normal, and if it is close to 1, it is conductive hearing loss.

4 is another example of a model for evaluating a middle ear condition using a VPO image. 4 uses a maximum positive pressure image (pos) and a maximum negative pressure image (neg) as input data. The model (Model 2) of FIG. 4 includes a convolution block that extracts features of an input image and a full sequence layer that receives the extracted features and calculates a final classification result. The model (Model 2) of FIG. 4 includes a first convolution block (conv. 1) and a second convolution block (conv. 2) for the maximum positive pressure image and the maximum negative pressure image, respectively. Each convolution block includes a plurality of convolution layers. The model (Model 2) of FIG. 4 sums the feature values calculated in the first convolution block (conv. 1) and the second convolution block (conv. 2), or subtracts another value from one value. (subtract) and then input to the electric thermal bonding layer (FC). As will be described later, the model in which the feature values are summed up in Model 2 is displayed as Model 2-sum, and the model in which the feature values are subtracted is displayed as Model 2-susbsract. The thermal insulation layer (FC) finally calculates a probability value for whether or not the input image has conductive hearing loss. The analysis device may compare this probability value with a predetermined threshold value to evaluate whether the input image is an image with conductive hearing loss. For example, the analyzer may evaluate that if the output value of model 1 is close to 0, it is normal, and if it is close to 1, it is conductive hearing loss.

5 is another example of a model for evaluating a middle ear condition using a VPO image. 5 uses a maximum positive pressure image (pos), a maximum negative pressure image (neg), and a residual image (dif) as input data. Here, the residual image is an image representing the difference between the maximum sound pressure image and the maximum sound pressure image.

The model (Model 3-sum) of FIG. 5 includes a convolution block that extracts features of an input image and a full sequence layer that receives the extracted features and calculates a final classification result. The model (Model 3-sum) of FIG. 5 is a first convolution block (conv. 1), a second convolution block (conv. 2), and a third convolution block for the maximum positive pressure image, the maximum negative pressure image, and the residual image, respectively. Contains block (conv. 3). Each convolution block includes a plurality of convolution layers. The model (Model 3 -sum) of FIG. 5 shows feature values calculated in the first convolution block (conv. 1), the second convolution block (conv. 2), and the third convolution block (conv. 3), respectively. After summing, it is input to the electric thermal bonding layer (FC). The thermal insulation layer (FC) finally calculates a probability value for whether or not the input image has conductive hearing loss. The analysis device may compare this probability value with a predetermined threshold value to evaluate whether the input image is an image with conductive hearing loss. For example, the analyzer may evaluate that if the output value of model 1 is close to 0, it is normal, and if it is close to 1, it is conductive hearing loss.

6 is another example of a model for evaluating a middle ear condition using a VPO image. 6 uses a maximum positive pressure image (pos), a maximum negative pressure image (neg), and a residual image (dif) as input data. Here, the residual image is an image representing the difference between the maximum sound pressure image and the maximum sound pressure image.

The model (Model 3-concatenate) of FIG. 6 includes a convolution block that extracts features of an input image and a full sequence layer that receives the extracted features and calculates a final classification result. The model (Model 3-concatenate) of FIG. 6 is a first convolution block (conv. 1), a second convolution block (conv. 2), and a third convolution block for the maximum positive pressure image, the maximum negative pressure image, and the residual image, respectively. Contains block (conv. 3). Each convolution block includes a plurality of convolution layers. The model (Model 3 -sum) of FIG. 5 shows feature values calculated in the first convolution block (conv. 1), the second convolution block (conv. 2), and the third convolution block (conv. 3), respectively. After concatenating, it is input to the electric heating layer (FC). The thermal insulation layer (FC) finally calculates a probability value for whether or not the input image has conductive hearing loss. The analysis device may compare this probability value with a predetermined threshold value to evaluate whether the input image is an image with conductive hearing loss. For example, the analyzer may evaluate that if the output value of model 1 is close to 0, it is normal, and if it is close to 1, it is conductive hearing loss.

Hereinafter, the results of verifying the performance of the above-described model for evaluating the middle ear condition will be described. For performance verification, the AUC of the ROC (Receiver Operating Characteristic) curve for sensitivity and specificity was used as a standard. The researcher compared the evaluation results of each model with the results clinically evaluated by a medical professional with more than 8 years of otolaryngology experience.

The researcher verified the performance of the model with the data of 133 subjects. Among the 144 data, there were 57 conductive hearing loss patients and 76 normal people. Images were obtained while repeatedly applying maximum positive pressure and maximum negative pressure to each subject. That is, there were multiple images of the same subject. The total number of maximum positive pressure images and maximum negative pressure images used was 1,130.

7 is a performance verification result of models for evaluating middle ear conditions. 7 is an evaluation result of a deep learning model implemented with Inception-v3. 7 is a model obtained by subtracting feature values from model 1 (Model 1-concatenate) of FIG. 3, model 2 of FIG. 4 (Model 2-sum), and model 2 of FIG. subtract), the results for Model 3 (Model 3 - sum) in Fig. 5 and Model 3 (Model 3 - concatenate) in Fig. 6 are shown. 7 is a verification result for 10 folds, and looking at the average value, the performance of the model subtracting the feature value was relatively slightly higher than the model using the feature value summed or combined.

Table 1 below is an example of comparison between the above-described middle ear condition evaluation model and an otologost evaluation result. Table 1 shows the evaluation results using the same VPO image. Specialist 1 (otologist 1) had relatively high specificity and PPV (positive predictive value), but overall, the deep learning model showed higher performance than the otologist's evaluation results.

8 is an example of an analysis device for evaluating a middle ear condition using a VPO image. The analyzer 300 corresponds to the aforementioned analyzer (160 in FIG. 1, 230 and 240 in FIG. 2). The analysis device 300 may be physically implemented in various forms. For example, the analysis device 300 may have a form of a computer device such as a PC, a network server, and a chipset dedicated to data processing.

The analysis device 300 may include a storage device 310, a memory 320, an arithmetic device 330, an interface device 340, a communication device 350, and an output device 360.

The storage device 310 may store the VPO image generated by the otoscopic examination device. That is, the storage device 310 may store a maximum positive pressure image and a maximum negative pressure image for a specific subject.

The storage device 310 may store a model for evaluating the state of the middle ear using the VPO image. The storage device 310 may store at least one of the models described in FIGS. 3 to 6 . Accordingly, it is assumed that the analysis device 300 evaluates the input VPO image using any one of the deep learning models having the structure of FIGS. 3 to 6 . Alternatively, in some cases, the analysis device 300 evaluates whether the input VPO image has conductive hearing loss by using a plurality of deep learning models having the structure of FIGS. 3 to 6, respectively, and makes a final decision based on the plurality of evaluation results Evaluation can also be performed (a kind of ensemble model).

The storage device 310 may store the final analysis result (whether or not a specific subject has conductive hearing loss).

The memory 320 may store data and information generated in the course of the analysis device 300 analyzing the VPO image using the deep learning model.

The interface device 340 is a device that receives certain commands and data from the outside. The interface device 340 may receive VPO images (maximum positive pressure image and maximum negative pressure image) of the subject from a physically connected input device, an external storage device, or an otoscope. The interface device 340 may transmit a result of analyzing the input VPO image to an external object.

The communication device 350 refers to a component that receives and transmits certain information through a wired or wireless network. The communication device 350 may receive VPO images (maximum positive pressure image and maximum negative pressure image) of the subject from an external object. Alternatively, the communication device 350 may transmit a result of analyzing the input VPO image to an external object such as a user terminal.

Since the interface device 340 and the communication device 350 are configured to send and receive certain data from a user or other physical object, they can also be collectively referred to as input/output devices. The interface device 340 and the communication device 350 may also be referred to as input devices if the function of receiving the subject's VPO image is limited.

The output device 360 is a device that outputs certain information. The output device 360 may output interfaces and analysis results necessary for data processing.

The computing device 330 may analyze the subject's VPO image using the deep learning model or program code stored in the storage device 310 .

The calculation device 330 inputs the input maximum positive pressure image and maximum negative pressure image to each input layer (convolution block) in the deep learning model, and calculates a probability value (classification result) through the deep learning model.

The calculation device 330 may calculate a probability value by inputting the maximum positive pressure image and the maximum negative pressure image to the deep learning model (FIG. 3 or 4). A value calculated by the deep learning model may be a probability value for conductive hearing loss. A value calculated by the deep learning model may be a probability value of whether the difference between the airway and bone conduction of the subject is equal to or greater than a threshold value.

The arithmetic unit 330 may generate a residual image of the input maximum positive pressure image and maximum negative pressure image.

In addition, the calculation device 330 may calculate a probability value by inputting the maximum positive pressure image, the maximum negative pressure image, and the residual image to the deep learning model (FIG. 5 or 6). A value calculated by the deep learning model may be a probability value for conductive hearing loss. A value calculated by the deep learning model may be a probability value of whether the difference between the airway and bone conduction of the subject is equal to or greater than a threshold value.

The computing device 330 finally generates information on conductive hearing loss based on the probability value output by the deep learning model. The computing device 330 may compare the calculated probability value with a predetermined threshold value to evaluate whether the input image is an image with conductive hearing loss. For example, the calculation device 330 may evaluate that the output value of the model is normal when the output value is close to 0, and conductive hearing loss when the output value is close to 1.

The arithmetic device 330 may be a device such as a processor, an AP, or a chip in which a program is embedded that processes data and performs certain arithmetic operations.

In addition, the deep learning model construction method for evaluating the middle ear condition and the method for evaluating the middle ear condition using VPO images as described above may be implemented as a program (or application) including an executable algorithm that can be executed on a computer. The program may be stored and provided in a temporary or non-transitory computer readable medium.

A non-transitory readable medium is not a medium that stores data for a short moment, such as a register, cache, or memory, but a medium that stores data semi-permanently and can be read by a device. Specifically, the various applications or programs described above are CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM (read-only memory), PROM (programmable read only memory), EPROM (Erasable PROM, EPROM) Alternatively, it may be stored and provided in a non-transitory readable medium such as EEPROM (Electrically EPROM) or flash memory.

Temporary readable media include static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), and enhanced SDRAM (Enhanced SDRAM). SDRAM, ESDRAM), Synchronous DRAM (Synclink DRAM, SLDRAM) and Direct Rambus RAM (DRRAM).

This embodiment and the drawings accompanying this specification clearly represent only a part of the technical idea included in the foregoing technology, and those skilled in the art can easily understand it within the scope of the technical idea included in the specification and drawings of the above technology. It will be obvious that all variations and specific examples that can be inferred are included in the scope of the above-described technology.

Claims (10)

receiving an air otoscopy image of a subject by an analysis device;
Step of the analysis device inputting the image to a pre-learned deep learning model; and
Evaluating, by the analysis device, whether or not the subject has conductive hearing loss based on an output value output from the deep learning model,
The image includes a maximum positive pressure image and a maximum negative pressure image, and the analysis device evaluates the middle ear state using an image of an air otoscopy inputting the maximum positive pressure image and the maximum negative pressure image to the deep learning model, respectively. . According to claim 1,
The deep learning model is an air otoscope that calculates the output value by inputting a value obtained by concatenating, summing, or subtracting characteristic values of each of the maximum positive pressure image and the maximum negative pressure image to a full connection layer. A method for evaluating the condition of the middle ear using images of. According to claim 1,
The analysis device further comprises generating a residual image representing a difference between the maximum positive pressure image and the maximum negative pressure image,
The method of evaluating the state of the middle ear using an air otoscopy image in which the analysis device further inputs the residual image to the deep learning model to calculate the output value. According to claim 3,
The deep learning model is an air otoscopy image that calculates the output value by inputting a value obtained by concatenating or summing features of each of the maximum positive pressure image, the maximum negative pressure image, and the residual image to a full connection layer. How to evaluate the middle ear condition using. According to claim 1,
The deep learning model is a method for evaluating the middle ear condition using an image of an air otoscope that calculates a probability value for whether the value of the air-bone gap of the subject is greater than or equal to a threshold value. an input device for receiving an air otoscopy image of a subject;
a storage device for receiving a maximum positive pressure image and a maximum negative pressure image and storing a deep learning model trained to calculate middle ear state information for the input image; and
The image includes a maximum positive pressure image and a maximum negative pressure image, and an arithmetic device for evaluating whether the subject has conductive hearing loss based on an output value calculated by inputting the maximum positive pressure image and the maximum negative pressure image to the deep learning model, respectively An analysis device for evaluating the state of the middle ear using an image of an air otoscopy comprising: According to claim 6,
The deep learning model is an air otoscope that calculates the output value by inputting a value obtained by concatenating, summing, or subtracting characteristic values of each of the maximum positive pressure image and the maximum negative pressure image to a full connection layer. An analysis device that evaluates the condition of the middle ear using images of According to claim 6,
The operation device generates a residual image representing a difference between the maximum positive pressure image and the maximum negative pressure image, and further inputs the residual image to the deep learning model to calculate the output value by using an air otoscope image to determine the state of the middle ear. An analysis device that evaluates. According to claim 8,
The deep learning model is an air otoscopy image that calculates the output value by inputting a value obtained by concatenating or summing features of each of the maximum positive pressure image, the maximum negative pressure image, and the residual image to a full connection layer. An analysis device that evaluates the middle ear condition by using. According to claim 6,
The deep learning model is an analysis device for evaluating the middle ear state using an image of an air otoscope that calculates a probability value for whether the value of the air-bone gap of the subject is greater than or equal to a threshold value.

Images