Prognosis diagnosis method and device for predicting prognosis of patient with sudden hearing loss, and learning method and device for training prognosis prediction model

Abstract

Disclosed are a prognosis diagnosis method and device for predicting the prognosis of a patient with sudden hearing loss, and a learning method and device for training a prognosis prediction model. The prognosis diagnosis method for predicting the prognosis of a patient with sudden hearing loss may comprise the steps of: collecting user data and medical data on a patient with sudden hearing loss; detecting a hearing damage frequency region of the patient with sudden hearing loss on the basis of hearing data of the patient with sudden hearing loss included in the medical data; determining a hearing threshold average value including an average value of hearing thresholds of an affected ear and an average value of hearing thresholds of an unaffected ear for the detected hearing damage frequency region, and determining variable values categorizing each of the average value of the hearing thresholds of the affected ears and the average value of the hearing thresholds of the unaffected ears; inputting input data including the hearing threshold average values and the categorized variable values into a prognosis prediction model to obtain prediction information on the prognosis of sudden hearing loss of the patient with sudden hearing loss from each of a plurality of machine learning models included in the prognosis prediction model; and determining prognosis diagnosis data for sudden hearing loss on the basis of the prediction information on the prognosis of sudden hearing loss of the patient with sudden hearing loss obtained from each of the machine learning models included in the prognosis prediction model.

Description

Prognostic diagnostic method and device for predicting the prognosis of patients with sudden hearing loss, and learning method and device for training a prognostic prediction model

[0001] The present invention relates to a prognostic diagnostic method and device for predicting the prognosis of patients with sudden hearing loss, and a learning method and device for learning a prognostic prediction model.

[0002] Sudden hearing loss (SHD) is a sensorineural hearing loss characterized by hearing loss of a specific frequency and range within a three-day period as detected by pure-tone audiometry. Sudden hearing loss may also be accompanied by tinnitus, aural fullness, and vertigo. Diagnosis of sudden hearing loss involves a basic medical history, otoscopic examination, and audiometric testing to determine the severity of hearing loss and to rule out other underlying causes. Treatment includes steroids, other circulatory enhancers, vasodilators, antivirals, and diuretics. The diagnosis of recovery in patients with sudden hearing loss is based on a standard frequency range that is consistent for all patients.

[0003] According to one embodiment, a prognostic diagnosis method for predicting the prognosis of a patient with sudden hearing loss comprises the steps of: collecting user data and medical data of a patient with sudden hearing loss; detecting a hearing impairment frequency range of a patient with sudden hearing loss based on hearing data of the patient with sudden hearing loss included in the medical data; determining a hearing threshold average value including an average value of hearing thresholds of an affected ear and an average value of hearing thresholds of an unaffected ear for the detected hearing impairment frequency range, and a variable value categorized by each of the average value of hearing thresholds of the affected ear and the average value of hearing thresholds of the unaffected ear; obtaining prediction information on the prognosis of the patient with sudden hearing loss from each of a plurality of machine learning models included in the prognosis prediction model by inputting input data including the hearing threshold average value and the categorized variable values ​​into a prognosis prediction model; And it may include a step of determining prognostic diagnostic data for the sudden hearing loss based on the prediction information for the prognosis of the sudden hearing loss patient obtained from each of the machine learning models included in the prognostic prediction model.

[0004] The step of detecting a frequency range of hearing impairment of a patient with sudden hearing loss may include a step of detecting a frequency range of hearing impairment of a patient with sudden hearing loss based on a continuous frequency range in which a hearing threshold is shown to be greater than or equal to a threshold value among hearing data of the patient with sudden hearing loss.

[0005] Predictive information on the prognosis of sudden hearing loss of the patient with sudden hearing loss may include the probability of recovery from sudden hearing loss of the patient with sudden hearing loss.

[0006] The above prognosis prediction model may be characterized as being an ensemble model including the plurality of machine learning models.

[0007] The step of determining the prognostic diagnostic data may include a step of determining the prognostic diagnostic data for sudden hearing loss indicating recovery or non-recovery of sudden hearing loss of the patient with sudden hearing loss based on an average of the recovery probability for sudden hearing loss of the patient with sudden hearing loss included in the prediction information for the prognosis of sudden hearing loss of the patient with sudden hearing loss obtained from each of the machine learning models.

[0008] A learning method for learning a prognostic prediction model according to one embodiment may include the steps of collecting learning data including user data and medical data for each patient with sudden hearing loss; detecting a hearing impairment frequency range of each patient with sudden hearing loss based on hearing data of each patient with sudden hearing loss included in the medical data; generating label data based on the detected hearing impairment frequency range and whether each patient with sudden hearing loss has recovered; and learning a plurality of machine learning models included in the prognostic prediction model based on the learning data and the label data.

[0009] The above prognosis prediction model may be characterized as an ensemble model including multiple machine learning models.

[0010] The step of detecting a hearing impairment frequency range of each of the sudden hearing loss patients may include a step of detecting a hearing impairment frequency range of each of the sudden hearing loss patients based on a continuous frequency range in which a hearing threshold is shown to be greater than or equal to a threshold value among the hearing data of each of the sudden hearing loss patients.

[0011] The above-described learning step may include a step of obtaining prediction information on the prognosis of each of the sudden hearing loss patients from each of the plurality of machine learning models included in the prognosis prediction model by inputting the learning data into the prognosis prediction model; and a step of training the plurality of machine learning models included in the prognosis prediction model based on the label data and the prediction information on the prognosis of each of the sudden hearing loss patients.

[0012] According to one embodiment, a prognostic diagnosis device for predicting the prognosis of a patient with sudden hearing loss includes a memory and a processor, wherein the memory stores instructions executable by the processor, and when the instructions are executed by the processor, the processor causes the prognostic diagnosis device to collect user data and medical data for a patient with sudden hearing loss, detect a frequency range of hearing impairment of the patient with sudden hearing loss based on hearing data of the patient with sudden hearing loss included in the medical data, determine an average value of a hearing threshold including an average value of a hearing threshold of an affected ear and an average value of a hearing threshold of an unaffected ear for the detected frequency range of hearing impairment, and a variable value categorized by each of the average value of the hearing threshold of the affected ear and the average value of the hearing threshold of the unaffected ear, and input input data including the average value of the hearing threshold and the categorized variable value into a prognostic prediction model, thereby predicting the prognosis of the patient with sudden hearing loss from each of a plurality of machine learning models included in the prognostic prediction model. The prognosis diagnosis device can be controlled to obtain prediction information on the prognosis and determine prognosis diagnosis data for the sudden hearing loss based on the prediction information on the prognosis of the sudden hearing loss patient obtained from each of the machine learning models included in the prognosis prediction model.

[0013] The processor can control the prognostic diagnosis device to detect a frequency range of hearing impairment of the patient with sudden hearing loss based on a continuous frequency range in which a hearing threshold is shown to be greater than or equal to a threshold value among the hearing data of the patient with sudden hearing loss.

[0014] Predictive information on the prognosis of sudden hearing loss of the patient with sudden hearing loss may include the probability of recovery from sudden hearing loss of the patient with sudden hearing loss.

[0015] The above prognosis prediction model may be characterized as being an ensemble model including the plurality of machine learning models.

[0016] The processor may control the prognostic diagnosis device to determine prognostic diagnosis data for sudden hearing loss indicating recovery or non-recovery of sudden hearing loss of the patient with sudden hearing loss based on an average of the recovery probability for sudden hearing loss of the patient with sudden hearing loss included in the prediction information for the prognosis of sudden hearing loss of the patient with sudden hearing loss obtained from each of the machine learning models.

[0017] A learning device for learning a prognostic model according to one embodiment includes a memory and a processor, wherein the memory stores instructions executable by the processor, and when the instructions are executed by the processor, the processor controls the learning device to collect learning data including user data and medical data for each of the sudden hearing loss patients, detect a hearing loss frequency range of each of the sudden hearing loss patients based on hearing data of each of the sudden hearing loss patients included in the medical data, generate label data based on the detected hearing loss frequency range and whether each of the sudden hearing loss patients has recovered, and learn a plurality of machine learning models included in the prognostic model based on the learning data and the label data.

[0018] The above prognosis prediction model may be characterized as an ensemble model including multiple machine learning models.

[0019] The processor can control the learning device to detect a hearing impairment frequency range of each of the sudden hearing loss patients based on a continuous frequency range in which a hearing threshold is shown to be above a threshold value among the hearing data of each of the sudden hearing loss patients.

[0020] The processor may control the learning device to obtain prediction information on the prognosis of each of the sudden hearing loss patients from each of the plurality of machine learning models included in the prognosis prediction model by inputting the learning data into the prognosis prediction model, and to train the plurality of machine learning models included in the prognosis prediction model based on the label data and the prediction information on the prognosis of each of the sudden hearing loss patients.

[0021] In one embodiment, a method for prognostic diagnosis for hearing impairments that vary from patient to patient may be provided.

[0022] In one embodiment, a personalized recovery prediction for sudden hearing loss can be provided by providing a prognostic diagnostic method for hearing impairment that varies from patient to patient.

[0023] In one embodiment, when a patient first visits the clinic and establishes a treatment plan, the optimal treatment timing can be determined based on the results of prognosis predictions using a machine learning model according to the treatment timing for applying various treatments.

[0024] According to one embodiment, a method for diagnosing hearing loss optimized for each patient can be provided, overcoming the limitations of conventional testing methods that performed hearing loss testing based on the same frequency range for all patients.

[0025] In one embodiment, the predictive model can provide a prognostic diagnostic method that overcomes the limitations of conventional testing methods that rely on the results of hearing tests.

[0026] FIG. 1 is a schematic diagram illustrating an overview of a prognostic diagnostic system for predicting the prognosis of a patient with sudden hearing loss according to one embodiment.

[0027] FIG. 2 is a flowchart illustrating a prognostic diagnostic method for predicting the prognosis of a patient with sudden hearing loss according to one embodiment.

[0028] Figure 3 is a flowchart illustrating a learning method for learning a prognostic prediction model according to one embodiment.

[0029] FIG. 4 is a diagram for explaining a prognosis prediction model according to another embodiment.

[0030] Figures 5 to 7 are drawings for explaining a process of generating label data according to one embodiment.

[0031] FIG. 8 is a diagram illustrating a configuration of a prognostic diagnostic device according to one embodiment.

[0032] Fig. 9 is a diagram illustrating the configuration of a learning device according to one embodiment.

[0033] Specific structural or functional descriptions of the embodiments are disclosed for illustrative purposes only and may be modified and implemented in various forms. Therefore, the actual implementation is not limited to the specific embodiments disclosed, and the scope of this specification includes modifications, equivalents, or alternatives within the technical concepts described in the embodiments.

[0034] Although terms such as "first" or "second" may be used to describe various components, these terms should be interpreted solely to distinguish one component from another. For example, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component.

[0035] When it is said that a component is "connected" to another component, it should be understood that it may be directly connected or connected to that other component, but there may also be other components in between.

[0036] Singular expressions include plural expressions unless the context clearly dictates otherwise. In this document, phrases such as "A or B", "at least one of A and B", "at least one of A or B", "A, B, or C", "at least one of A, B, and C", and "at least one of A, B, or C" can each include any one of the items listed together in that phrase, or all possible combinations thereof. In this specification, it should be understood that the terms "comprises" or "has" and the like are intended to specify the presence of a described feature, number, step, operation, component, part, or combination thereof, but do not exclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0037] Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the art. Terms defined in commonly used dictionaries should be interpreted to have a meaning consistent with their meaning in the context of the relevant technology, and will not be interpreted in an idealized or overly formal sense unless explicitly defined herein.

[0038] The term "module" as used herein may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit. A module may be an integral component, or a minimum unit or portion of such a component that performs one or more functions. For example, according to one embodiment, a module may be implemented in the form of an application-specific integrated circuit (ASIC).

[0039] The term "~part" as used in this document refers to a software or hardware component such as an FPGA or ASIC, and the "~part" performs certain roles. However, the "~part" is not limited to software or hardware. The "~part" may be configured to reside on an addressable storage medium and may be configured to execute one or more processors. For example, the "~part" may include components such as software components, object-oriented software components, class components, and task components, as well as processes, functions, attributes, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided within the components and "~parts" may be combined into a smaller number of components and "~parts" or further separated into additional components and "~parts." Furthermore, the components and "~parts" may be implemented to execute one or more CPUs within a device or a secure multimedia card. Additionally, '~bu' may include one or more processors.

[0040] Hereinafter, embodiments will be described in detail with reference to the attached drawings. In the description with reference to the attached drawings, identical components are assigned the same reference numerals regardless of the drawing numbers, and redundant descriptions thereof will be omitted.

[0041] FIG. 1 is a schematic diagram illustrating an overview of a prognostic diagnostic system for predicting the prognosis of a patient with sudden hearing loss according to one embodiment.

[0042] The prognostic diagnosis system described herein can provide a machine learning-based, patient-specific prognostic diagnosis method for sudden hearing loss, based on the patient's individual hearing impairment frequency range, which serves as the basis for the diagnosis of sudden hearing loss. The prognostic diagnosis system can use a prognostic prediction model, which is an ensemble model comprising multiple machine learning models, to determine prognostic data indicating whether or not the patient will recover from sudden hearing loss from user data and medical data.

[0043] Referring to FIG. 1, a prognostic diagnosis system may include a prognostic diagnosis device (120). The prognostic diagnosis device (120) may collect user data and medical data (110) for a patient with sudden hearing loss. The prognostic diagnosis device (120) may detect a frequency range of hearing impairment of a patient with sudden hearing loss based on hearing data of the patient with sudden hearing loss included in the medical data. The prognostic diagnosis device (120) may determine an average value of hearing thresholds including an average value of hearing thresholds of the affected ear and an average value of hearing thresholds of the non-affected ear for the detected hearing impairment frequency range, and variable values ​​categorized by each of the average value of hearing thresholds of the affected ear and the average value of hearing thresholds of the non-affected ear. The prognostic diagnosis device (120) can obtain prediction information on the prognosis of sudden hearing loss of a patient with sudden hearing loss from each of a plurality of machine learning models included in the prognostic prediction model by inputting input data including an average hearing threshold value and categorized variable values ​​into the prognostic prediction model. The prognostic diagnosis device (120) can determine prognostic diagnosis data (130) for sudden hearing loss based on the prediction information on the prognosis of sudden hearing loss of a patient with sudden hearing loss obtained from each of the machine learning models included in the prognostic prediction model.

[0044] FIG. 2 is a flowchart illustrating a prognostic diagnostic method for predicting the prognosis of a patient with sudden hearing loss according to one embodiment.

[0045] Referring to FIG. 2, in step (210), the prognostic diagnosis device may collect user data and medical data regarding a patient with sudden hearing loss. The user data may include, for example, information regarding at least one of age, height, weight, and gender. Additionally, the user data may include demographic information including at least one of age, height, weight, and gender. The medical data may include at least one of health records, blood test results, onset records of sudden hearing loss, treatment records of sudden hearing loss, and pure tone audiometry (PTA) records. The health records may include eight variables related to body mass index, smoking intensity (packs / year), systolic blood pressure, diastolic blood pressure, smoking status, smoking cessation status, and the presence of a specific disease. Here, the specific disease may include hypertension, diabetes, stroke, dizziness, tinnitus, hyperlipidemia, chronic kidney disease, myocardial infarction, or syringomyelitis. Blood test results may include information on at least one of total cholesterol, low-density lipoprotein, triacylglycerol, hemoglobin, blood urea nitrogen, creatinine, white blood cell count, neutrophil count, lymphocyte count, neutrophil-to-lymphocyte ratio, platelet count, prothrombin time, and activated partial thromboplastin time. Onset history of sudden hearing loss, treatment history of sudden hearing loss may include at least one of the following variables: time from onset of sudden hearing loss to first treatment, time from onset of sudden hearing loss to first intratympanic dexamethasone injection (ITDI) treatment, whether hospitalized, affected ear (left or right ear), categorical variables for time from onset of sudden hearing loss to first ITDI treatment, month of onset of sudden hearing loss, length of frequency range of hearing loss, and presence or absence of treatment method.The presence / absence variables for treatment methods may include the presence / absence of systemic steroid treatment, the presence / absence of ITDI treatment, and the presence / absence of systemic steroid and intratympanic dexamethasone injection treatment. The PTA record may include at least one of the mean hearing threshold value including the mean value of the hearing threshold of the affected ear and the mean value of the hearing threshold of the non-affected ear in the frequency range of the hearing loss in patients with sudden hearing loss, a categorical variable value of each of the mean values ​​of the hearing threshold of the affected ear and the mean values ​​of the hearing threshold of the non-affected ear, and a binary variable for the type of audiogram in the PTA record measured at the first visit. The binary variable for the type of audiogram in the PTA record measured at the first visit may include the presence / absence of ascending type, the presence / absence of U-shaped type, the presence / absence of descending type, the presence / absence of flat type, and the presence / absence of hearing loss type. Among medical data, data related to hearing may be referred to as hearing data. Hearing data may include PTA records. The items included in the user data and medical data described herein are exemplary and are not limited to the descriptions herein. The user data and medical data may include at least one of the following formats: tabular files, image files, and audio files.

[0046] The sudden hearing loss patients described in this specification may correspond to patients diagnosed with sudden hearing loss and treated for sudden hearing loss at a hospital, and may also be referred to as patients in this specification.

[0047] In step (220), the prognostic diagnosis device can detect a frequency range of hearing loss in a patient with sudden hearing loss based on the patient's hearing data included in the medical data. The prognostic diagnosis device can detect a frequency range of hearing loss in a patient with sudden hearing loss based on a continuous frequency range in which the hearing threshold is shown to be above a threshold value among the patient's hearing data.

[0048] For example, a prognostic diagnostic device can detect the frequency range of hearing loss in a patient with sudden hearing loss based on the pure tone audiometry (PTA) record (hearing thresholds at hearing frequencies of 0.125 kHz, 0.25 kHz, 0.5 kHz, 1 kHz, 2 kHz, 3 kHz, 4 kHz, and 8 kHz) of the affected ear measured at the first visit of the patient with sudden hearing loss. Since sudden hearing loss is a sensorineural hearing loss in which a hearing loss of 30 dB or more occurs in three or more consecutive frequencies in a pure tone audiometry within three days (72 hours), the prognostic diagnostic device can detect three or more consecutive frequency ranges in which hearing thresholds of 30 dB or more occur among the frequency ranges as the hearing loss frequency range. For example, if the hearing threshold of a patient with sudden hearing loss is 30 dB or more in the 1 kHz, 2 kHz, and 3 kHz ranges, the prognostic diagnostic device can determine the 1 kHz, 2 kHz, and 3 kHz ranges as the hearing loss frequency ranges in the patient with sudden hearing loss. The hearing loss frequency range can be a reference frequency range that serves as a criterion for evaluating recovery in patients with sudden hearing loss.

[0049] As part of data preprocessing, the prognostic diagnostic device can detect the frequency range of hearing loss in patients with sudden hearing loss based on the patient's hearing data included in medical data. The data preprocessing may include detecting the frequency range of hearing loss in patients with sudden hearing loss, handling missing values, and scaling variables. A more detailed description of the data preprocessing can be found in Figure 4.

[0050] In step (230), the prognostic diagnosis device can determine a hearing threshold average value including an average value of the hearing threshold of the affected ear and an average value of the hearing threshold of the non-affected ear for the detected hearing impairment frequency range, and a variable value categorized by each of the average value of the hearing threshold of the affected ear and the average value of the hearing threshold of the non-affected ear. The categorized variable value can correspond to any one of, for example, hearing level 1 (mild sudden hearing loss) below 40 dB, hearing level 2 (moderate sudden hearing loss) between 40 dB and 60 dB, hearing level 3 (severe sudden hearing loss) between 60 dB and 80 dB, hearing level 4 (profound sudden hearing loss) between 80 dB and 100 dB, and hearing level 5 (severe hearing loss) above 100 dB. The variable values ​​categorized for each of the average values ​​of the hearing thresholds may be the variable values ​​assigned to each of the corresponding categories of the average values ​​of the hearing thresholds. In other words, the variable values ​​categorized for each of the average values ​​of the hearing thresholds of the affected ear and the hearing thresholds of the non-affected ear may be the variable values ​​assigned to each of the corresponding categories of the average values ​​of the hearing thresholds of the affected ear and the hearing thresholds of the non-affected ear. The prognostic diagnostic device may exclude from the medical data the hearing threshold variables for each hearing frequency outside the range corresponding to the variable values ​​categorized for each of the average values ​​of the hearing thresholds of the affected ear and the hearing thresholds of the non-affected ear, as well as the average hearing threshold variables in the high, middle, and low frequency regions.

[0051] In step (240), the prognosis diagnosis device may obtain prediction information on the prognosis of sudden hearing loss of a patient with sudden hearing loss from each of a plurality of machine learning models included in the prognosis prediction model by inputting input data including at least one of user data, medical data, an average value of hearing threshold, and a variable value categorized by each of the average value of hearing threshold of the affected ear and the average value of hearing threshold of the non-affected ear, into the prognosis prediction model. The input data may include at least one of user data on which data preprocessing has been performed, medical data on which data preprocessing has been performed, an average value of hearing threshold, and an average value of hearing threshold of the affected ear and the average value of hearing threshold of the non-affected ear. The prediction information on the prognosis of sudden hearing loss of a patient with sudden hearing loss may include a recovery probability for sudden hearing loss of the patient with sudden hearing loss. Each recovery probability may have a value of 0 to 1.

[0052] The prognostic model may be an ensemble model comprising multiple machine learning models. A more detailed description of the prognostic model can be found in the description of Figure 4.

[0053] In step (250), the prognostic diagnosis device may determine prognostic diagnosis data for sudden hearing loss based on prediction information about the prognosis of sudden hearing loss of a patient with sudden hearing loss obtained from each of the machine learning models included in the prognostic prediction model. The prognostic diagnosis device may determine prognostic diagnosis data for sudden hearing loss indicating recovery or non-recovery of sudden hearing loss of a patient with sudden hearing loss based on an average of the recovery probabilities of sudden hearing loss of a patient with sudden hearing loss included in the prediction information about the prognosis of sudden hearing loss of a patient with sudden hearing loss obtained from each of the machine learning models. For example, if the average of the recovery probabilities of sudden hearing loss of a patient with sudden hearing loss is less than 0.5, the prognostic diagnosis device may determine the prognostic diagnosis data for sudden hearing loss as 0 by determining that the patient's sudden hearing loss is non-recovery, and if it is 0.5 or more, the prognostic diagnosis data for sudden hearing loss may be determined as 1 by determining that the patient's sudden hearing loss is recovered.

[0054] Figure 3 is a flowchart illustrating a learning method for learning a prognostic prediction model according to one embodiment.

[0055] Referring to FIG. 3, the learning device may collect learning data including user data and medical data for each patient with sudden hearing loss in step (310). The user data and medical data may include user data and medical data corresponding to each patient with sudden hearing loss. The learning data may include user data and medical data for each patient with sudden hearing loss who visited the hospital during a specific period. The user data and medical data may include information such as the patient's age, weight, height, smoking, and blood pressure information, current disease and blood test result information, type of treatment technique, and treatment period, and at least one of information regarding hearing thresholds at frequencies of 0.125, 0.25, 0.5, 1, 2, 3, 4, and 8 kHz of the affected ear with sudden hearing loss and the unaffected ear without sudden hearing loss, as determined through audiometric tests on the patient at the first visit and one month after treatment. Patients with sudden hearing loss may have been administered steroids by oral or intravenous injection, intratympanic injection, or a combination of oral, intravenous, and intratympanic injection for a predetermined period. Patients with sudden hearing loss who have had surgery on the affected ear and those with chronic otitis media or inner ear abnormalities detected by MRI may be excluded. In addition, patients with sudden hearing loss who have missing audiometric data in the frequency ranges of 0.125, 0.25, 0.5, 1, 2, 3, 4, and 8 kHz measured on both the affected and unaffected ears at the first visit and one month after treatment, patients without a frequency range of hearing loss (patients whose hearing thresholds in each hearing frequency are not 30 dB or higher in at least three consecutive hearing frequencies in the hearing frequency range of the affected ear at the first visit), patients with sudden hearing loss on both sides, and patients with duplicated data may be excluded. The items of user data and medical data may be identical to the description of user data and medical data in the description of FIG. 2.In the medical data of patients with sudden hearing loss, variables with missing values ​​exceeding half of all patients with sudden hearing loss can be excluded from the variables in the training data, and only variables corresponding to the average hearing threshold value in the frequency range of hearing loss can be included. Among the variables included in the training data, categorical variables, rather than ordinal variables, can be encoded as binary variables. For example, the PTA audiogram type variable, where ascending type has a value of 1, U-shaped type has a value of 2, descending type has a value of 3, flat type has a value of 4, and hearing loss type has a value of 5, can be converted into five binary variables: ascending type, U-shaped type, descending type, flat type, and hearing loss type. To minimize the variables of hearing test records, detailed hearing records such as hearing threshold values ​​by frequency, average hearing threshold values ​​in low frequency, middle frequency, and high frequency regions are excluded from the data composition variables, and only the average hearing threshold value in the recovery evaluation reference frequency region from the audiometry records of the affected and non-affected ears and its categorical variables can be included in the learning data for the hearing test record variables. The categorical variable for hearing can be defined as hearing level 1 (mild) below 40 dB, hearing level 2 (moderate) between 40 dB and 60 dB, hearing level 3 (severe) between 60 dB and 80 dB, hearing level 4 (profound) between 80 dB and 100 dB, and hearing level 5 (severe) above 100 dB.

[0056] In step (320), the learning device can detect a hearing damage frequency range of each of the sudden hearing loss patients based on the hearing data of each of the sudden hearing loss patients included in the medical data. The learning device can detect the hearing damage frequency range of each of the sudden hearing loss patients based on a continuous frequency range in which the hearing threshold is higher than a threshold value among the hearing data of each of the sudden hearing loss patients. For example, the learning device can detect the hearing damage frequency range of the sudden hearing loss patient based on a PTA record of the affected ear measured at the time of the first visit of the sudden hearing loss patient. Since sudden hearing loss is a sensorineural hearing loss in which a hearing loss of 30 dB or more occurs in three or more continuous frequencies within three days (72 hours) in a pure tone audiometry, the learning device can detect three or more continuous frequency ranges in which a hearing threshold of 30 dB or more occurs among the frequency ranges as a hearing damage frequency range. For example, if the hearing threshold of a patient with sudden hearing loss is 30 dB or higher in the 1 kHz, 2 kHz, and 3 kHz ranges, the learning device can determine the frequency ranges of hearing damage of the patient with sudden hearing loss as the 1 kHz, 2 kHz, and 3 kHz ranges.

[0057] As part of data preprocessing, the learning device can detect a frequency range of hearing loss for each patient with sudden hearing loss based on the hearing data of each patient with sudden hearing loss included in the medical data. The data preprocessing can include detecting a frequency range of hearing loss for each patient with sudden hearing loss, handling missing values ​​using MICE, and performing maximum-minimum variable scaling. The learning device can apply the MICE (Multiple Imputation by Chained Equations) model trained on the learning data to test data to compute and fill in missing values, and can apply maximum-minimum scaling to the corresponding variables of the test data using the maximum and minimum values ​​of each variable in the learning data so that the range of values ​​is between 0 and 1.

[0058] In step (330), the learning device can generate label data based on the detected hearing impairment frequency range and whether each of the sudden hearing loss patients has recovered.

[0059] The learning device can detect a frequency range of hearing impairment in a patient with sudden hearing loss based on pre-treatment hearing data, which is hearing data measured when the patient with sudden hearing loss first visits the hospital. The learning device can determine a mean hearing threshold value, which is an average of hearing thresholds of the affected ear in the frequency range of hearing impairment, based on the pre-treatment hearing data. The mean hearing threshold value, which is an average of hearing thresholds of the affected ear in the frequency range of hearing impairment determined based on the initial hearing data, can be referred to as a mean hearing threshold value before treatment. In addition, the learning device can determine a mean hearing threshold value, which is an average of hearing thresholds of the affected ear in the frequency range of hearing impairment, based on post-treatment hearing data, which is hearing data measured one month after the patient with sudden hearing loss receives treatment for the sudden hearing loss. The mean hearing threshold value, which is an average of hearing thresholds of the affected ear in the frequency range of hearing impairment determined based on the post-treatment hearing data, can be referred to as a mean hearing threshold value after treatment. Here, the patient with sudden hearing loss is one month after receiving treatment for sudden hearing loss, and is not limited to the description in this specification.

[0060] The learning device can determine whether a patient with sudden hearing loss has recovered by applying the average value of the hearing threshold before treatment and the average value of the hearing threshold after treatment to the Siegel criteria of Fig. 7, and can generate label data for the learning data by labeling whether the patient with sudden hearing loss has recovered to the learning data.

[0061] Referring back to FIG. 3, in step (340), the learning device can train a plurality of machine learning models included in the prognosis prediction model based on learning data and label data. Here, the plurality of machine learning models included in the prognosis prediction model may include, for example, logistic regression, decision tree, support vector machine, random forest, adaptive boosting, light gradient boosting machine, extreme gradient boosting, and K-nearest neighbors models, but are not limited to those described herein.

[0062] A learning device can obtain prediction information on the prognosis of sudden hearing loss for each patient with sudden hearing loss from each of the multiple machine learning models included in the prognosis prediction model by inputting learning data into the prognosis prediction model. The learning device can train the multiple machine learning models included in the prognosis prediction model based on label data and the prediction information on the prognosis of sudden hearing loss for each patient with sudden hearing loss. The learning data can be preprocessed and input into the prognosis prediction model in the form of a table. The learning device can perform supervised machine learning on the prognosis prediction model, i.e., on the multiple machine learning models included in the prognosis prediction model, based on the learning data and the label data. The prognosis prediction model can output prediction information on the prognosis of sudden hearing loss for a patient with sudden hearing loss corresponding to the learning data based on the learning data. The prediction information on the prognosis of sudden hearing loss for a patient with sudden hearing loss can include a recovery probability for sudden hearing loss for the patient with sudden hearing loss output by each of the machine learning models included in the prognosis prediction model. If the recovery probability of sudden hearing loss in a patient with sudden hearing loss does not correspond to or match the label data, the learning device can train the prognostic prediction model by adjusting the parameters of multiple machine learning models included in the prognostic prediction model. Exceptionally, the multiple machine learning models included in the prognostic prediction model are not limited to models trained based on the label data and prediction information on the prognosis of sudden hearing loss in each patient with sudden hearing loss, and may also include models that do not undergo an actual training process and make predictions based on stored training data, such as the KNN (K-Nearest Neighbors) model.

[0063] In one example, among the patients with sudden hearing loss corresponding to the training data, 80% of the training data, which has been preprocessed to maintain the ratio of patients who recovered from sudden hearing loss and patients who did not recover from sudden hearing loss, can be used to train a prognostic model, and 20% of the training data can be used to test the prognostic model. The learning device can improve the performance degradation caused by data imbalance by setting the "class_weight" parameter to "balanced" in models that support the parameter in Python APIs such as Scikit-learn and LightGBM, thereby giving a higher prediction weight to a minority patient group than to a majority patient group. Models that set the "class_weight" parameter to "balanced" can include logistic regression, support vector machine, decision tree, random forest, and light gradient boosting models. The learning device can apply grid search, a model parameter optimization technique, to find the optimal parameters of the model through stratified K-fold cross-validation, and train a model with the corresponding parameters introduced to the entire learning data.

[0064] The prognostic model may be a soft voting ensemble model comprising multiple machine learning models. The learning device may construct a soft voting ensemble model by configuring the multiple trained machine learning models. For example, the number of machine learning models may be eight, but this is not limited to the description herein.

[0065] In one embodiment, the prognostic model may be trained using a machine learning process. The machine learning process uses given data to enable the prognostic model to recognize patterns in input data and make predictions on its own. This machine learning process may include (1) data preparation, (2) parameter optimization of each machine learning model constituting the prognostic model, and (3) training each machine learning model constituting the prognostic model.

[0066] The data preparation process involves collecting and preprocessing the training data that will be used by the multiple machine learning models that comprise the prognostic model. Preprocessing involves refining the training data and, if necessary, performing standardization, normalization, and feature selection to transform the training data into a format suitable for the multiple machine learning models that comprise the prognostic model.

[0067] The parameter optimization process for each machine learning model that constitutes the prognostic prediction model is the process of adjusting the model's parameters so that the predictive performance of multiple machine learning models included in the prognostic prediction model is optimized. For example, the "C" parameter of the logistic regression model in Python's Scikit-learn API can be set to 0.01, 0.1, 1, 10, and 100, and a grid search algorithm that sequentially applies all the set "C" values ​​can be used to select the C value that produces the highest AUROC (Area Under the Receiver Operative Characteristic) predictive performance evaluation index. The parameter optimization process for a machine learning model can also be performed with multiple parameters, not just a single parameter.

[0068] Performance evaluation during the parameter optimization process for each machine learning model that constitutes the prognostic model can be performed using K-fold cross-validation, which divides the entire training data into multiple combinations of training and validation data, and then performs the evaluation on each partition. For example, the process of training a model with specific parameter values ​​set to training data and evaluating the performance of the trained model through validation data can include calculating the average performance across multiple combinations of training and validation data sets. The parameters of the model with the highest measured performance indicator can then be selected.

[0069] The training process for each machine learning model that constitutes the prognostic model can be a process of training a machine learning model with optimized parameters using the entire training data set, which is a combination of the split training data set and the validation data set. By training the model based on the entire training data set, which includes a larger number of patients, more generalized predictions can be achieved.

[0070] Machine learning models that construct prognostic models calculate a loss function by comparing the model's predicted probability or classification results with label data, and are trained to minimize this loss function. For example, a logistic regression model updates the model's weights and biases to minimize the cross-entropy loss function by comparing the model's predicted probability with label data. However, as an exception, a K-nearest neighbor (KNN) model can be included. Instead of training to minimize the loss function, it compares stored data with new input data to find the K nearest neighbors and performs predictions through majority vote.

[0071] The evaluation of the predictive performance of a prognostic model and each of the multiple machine learning models that comprise the prognostic model can include three main steps: 1) separation of training and test data sets, 2) validation through K-fold internal cross-validation and a test data set, and 3) selection of a high-performing model. First, in the separation of training and test data sets, the training data can be split into an 80% training data set and a 20% test data set. Then, in the validation through K-fold internal cross-validation and a test data set, the evaluation of the predictive performance of each machine learning model can be performed through internal validation using K-fold cross-validation using the training data and prediction performance evaluation using the test data. In the selection of a high-performing model, the model with the highest AUROC index among the machine learning models can be selected as the best machine learning model. As a result of measuring the AUROC of the machine learning model, the soft voting ensemble model showed the best performance among the models built in both cases, with 0.775 (95% CI 0.659-0.887) in the internal validation prediction performance evaluation and 0.864 (95% CI 0.801-0.927) in the prediction performance evaluation conducted with test data.

[0072] FIG. 4 is a diagram for explaining a prognosis prediction model according to another embodiment.

[0073] Referring to FIG. 4, at least one of the prognostic diagnosis device and the learning device may include a preprocessing unit (420), a model unit (430), and an output unit (440).

[0074] The preprocessing unit (420) can generate input data including user data on which data preprocessing has been performed and medical data on which data preprocessing has been performed by performing data preprocessing on user data and medical data (410). The preprocessing unit (420) can include a patient-specific hearing loss frequency domain detection unit (421), a missing value processing unit (422), and a variable scaling unit (423). The patient-specific hearing loss frequency domain detection unit (421) can detect a frequency domain of hearing loss of a patient with sudden hearing loss based on hearing data of the patient with sudden hearing loss included in the medical data. The patient-specific hearing loss frequency domain detection unit (421) can detect a frequency domain of hearing loss of a patient with sudden hearing loss based on hearing data of the patient with sudden hearing loss included in the medical data during a learning process.

[0075] The missing value processing unit (422) can calculate and replace missing values ​​contained in at least one of medical data and user data using MICE. MICE can be trained based on training data for training a prognostic prediction model.

[0076] The variable scaling unit (423) can scale the corresponding variable to a range of 0 to 1 based on the maximum and minimum values ​​of each variable included in at least one of the user data and the medical data. The preprocessing unit (420) can perform data preprocessing by detecting a hearing loss frequency range for at least one of the user data and the medical data, filling in missing values, and scaling variables, and can generate input data. The input data can also include variable values ​​categorized by the average value of the hearing threshold, the average value of the hearing threshold of the affected ear, and the average value of the hearing threshold of the non-affected ear, respectively.

[0077] The learning device can also perform data preprocessing on learning data including user data and medical data through the above process.

[0078] The model unit (430) may correspond to the prognostic model described herein. The model unit (430) may be an ensemble model including multiple machine learning models. The model unit (430) may include, for example, logistic regression, decision trees, random forests, adaptive boosting, extreme gradient boosting, support vector machines, light gradient boosting models, and K-nearest neighbor models, but is not limited to those described herein.

[0079] A plurality of machine learning models can output prediction information on the prognosis of sudden hearing loss in a patient with sudden hearing loss based on input data. The prediction information on the prognosis of sudden hearing loss in a patient with sudden hearing loss can include a recovery probability for sudden hearing loss in the patient with sudden hearing loss. The model unit (430) can calculate an average of the recovery probabilities for sudden hearing loss in the patient with sudden hearing loss output from each of the machine learning models. If the average of the recovery probabilities for sudden hearing loss in the patient with sudden hearing loss is less than 0.5, the prognosis diagnosis data (450) for sudden hearing loss can be determined as 0 because the patient's sudden hearing loss is determined to be unrecoverable, and if it is 0.5 or more, the prognosis diagnosis data (450) for sudden hearing loss can be determined as 1 because the patient's sudden hearing loss is determined to be recoverable, but is not limited to the description in the present specification. The output unit (440) can output determined prognostic diagnostic data (450).

[0080] Figures 5 to 7 are drawings for explaining a process of generating label data according to one embodiment.

[0081] The learning device can detect three or more consecutive frequency regions in which a hearing threshold of 30 dB or more occurs in the hearing data of a patient with sudden hearing loss as a hearing loss frequency region.

[0082] Referring to Fig. 5, it can be seen that the hearing threshold was measured to be 30 dB at 0.125 kHz, 0.25 kHz, 0.5 kHz, 1 kHz, 4 kHz, and 8 kHz. In the embodiment of Fig. 5, the learning device can detect the hearing damage frequency range (520) of a sudden hearing loss patient as 0.125 kHz, 0.25 kHz, 0.5 kHz, 1 kHz, 4 kHz, and 8 kHz. Three or more consecutive frequency ranges among the hearing damage frequency ranges (520) can be referred to as recovery evaluation reference frequency ranges (530) for evaluating recovery from sudden hearing loss in a sudden hearing loss patient.

[0083] Referring to FIG. 6, the learning device can detect the hearing impairment frequency range (615) as 0.125 kHz, 0.25 kHz, 0.5 kHz, and 1 kHz based on the learning data in step (610).

[0084] The learning device can calculate an average hearing threshold value (625) at the time of the first examination, which is an average of the hearing thresholds of the affected ear in the hearing damage frequency range, based on the hearing data measured at the time of the first examination after the patient with sudden hearing loss visits the hospital in step (620). Here, the time of the first examination can correspond to before treatment or at the time of the visit. The hearing data measured at the time of the first examination after the patient with sudden hearing loss visits the hospital can correspond to the hearing data before treatment described herein or the hearing data measured at the time of the visit. The average hearing threshold value (625) at the time of the first examination can be 55 dB based on (70+60+50+40) dB / 4.

[0085] The learning device can calculate a mean hearing threshold value (635) after one month of treatment, which is an average of the hearing thresholds of the affected ear in the hearing impairment frequency range, based on the hearing data measured after one month of treatment of the sudden hearing loss patient in step (630). Here, the hearing data measured after one month of treatment may correspond to the hearing data after treatment described herein. The mean hearing threshold value (635) after one month of treatment may be 30 dB. The improved mean hearing threshold value (637) may represent a change in the mean hearing threshold value in the hearing impairment frequency range (615), and may be the difference between the mean hearing threshold value (625) at the time of the initial examination and the mean hearing threshold value (635) after one month of treatment. The improved mean hearing threshold value (637) may be 25 dB, which is (55-30) dB. In step (640), the learning device can generate label data by applying the average hearing threshold value (635) and the average improved hearing threshold value (637) after one month of treatment to the Siegel criteria. The learning device can determine the recovery stage of the sudden hearing loss patient corresponding to the learning data as partial recovery based on the fact that the average hearing threshold value after one month of treatment is 45 dB or less and the average improved hearing threshold value (637) is greater than 15 dB. In addition, in step (650), the learning device can generate label data by labeling the learning data with 1, which indicates that the sudden hearing loss patient has recovered from the sudden hearing loss.

[0086] Figure 7 may be a diagram for explaining the Siegel criterion. If the recent average hearing threshold value is 25 dB or less, the recovery stage may be complete recovery, and whether or not recovery may be recovery, and the corresponding training data may be labeled as recovery. Here, the recent average hearing threshold value may be based on the post-treatment hearing data described herein and the hearing data measured one month after the treatment, and may correspond to the average hearing threshold value one month after the treatment. If the recent average hearing threshold value is 45 dB or less and the degree of improvement in the average hearing threshold value is greater than 15 dB, the recovery stage may be partial recovery, and whether or not recovery may be recovery, and the corresponding training data may be labeled as recovery. Here, the degree of improvement in the average hearing threshold value may correspond to the improved average hearing threshold value described herein. If the recent average hearing threshold is less than 75 dB and the degree of improvement in the average hearing threshold is greater than 15 dB, the recovery stage may be weak recovery, the recovery may be incomplete, and the corresponding training data may be labeled as incomplete. In addition, if the recent average hearing threshold is greater than 75 dB or the degree of improvement in the average hearing threshold is less than 15 dB, the recovery stage may be incomplete recovery, the recovery may be incomplete, and the corresponding training data may be labeled as incomplete.

[0087] FIG. 8 is a diagram illustrating a configuration of a prognostic diagnostic device according to one embodiment.

[0088] Referring to FIG. 8, the prognostic diagnosis device (800) may include a processor (810) and a memory (820). The prognostic diagnosis device (800) may correspond to the prognostic diagnosis device described herein.

[0089] The memory (820) is coupled to the processor (810) and can store instructions executable by the processor (810), data to be calculated by the processor (810), or data processed by the processor (810). The memory (820) can include a non-transitory computer-readable medium, such as a high-speed random access memory and / or a non-volatile computer-readable storage medium (e.g., one or more disk storage devices, flash memory devices, or other non-volatile solid-state memory devices).

[0090] The processor (810) may perform one or more operations related to the operation of the prognostic diagnosis device described herein. For example, the processor (810) may control the prognostic diagnosis device (800) to collect user data and medical data regarding a patient with sudden hearing loss, and to detect a frequency range of hearing loss in the patient with sudden hearing loss based on the hearing data of the patient with sudden hearing loss included in the medical data.

[0091] The processor (810) can control the prognostic diagnosis device (800) to detect a hearing damage frequency range of a sudden hearing loss patient based on a continuous frequency range in which the hearing threshold of the sudden hearing loss patient is shown to be above a threshold value among the hearing data of the sudden hearing loss patient.

[0092] The processor (810) can control the prognosis diagnosis device (800) to obtain prediction information on the prognosis of sudden hearing loss of a patient with sudden hearing loss from each of a plurality of machine learning models included in the prognosis prediction model by determining a hearing threshold average value including an average value of the hearing threshold of the affected ear and an average value of the hearing threshold of the non-affected ear for a detected hearing loss frequency range, and a variable value categorizing each of the average value of the hearing threshold of the affected ear and the average value of the hearing threshold of the non-affected ear, and inputting input data including the hearing threshold average value and the variable value categorizing each of the average value of the hearing threshold of the affected ear and the average value of the hearing threshold of the non-affected ear into the prognosis prediction model.

[0093] The processor (810) can control the prognostic diagnosis device (800) to determine prognostic diagnosis data for sudden hearing loss based on the prediction information for the prognosis of sudden hearing loss of a patient with sudden hearing loss obtained from each of the machine learning models included in the prognostic prediction model. The processor (810) can control the prognostic diagnosis device (800) to determine prognostic diagnosis data for sudden hearing loss indicating recovery or non-recovery of sudden hearing loss of a patient with sudden hearing loss based on the average of the recovery probability for sudden hearing loss of a patient with sudden hearing loss included in the prediction information for the prognosis of sudden hearing loss of a patient with sudden hearing loss obtained from each of the machine learning models.

[0094] Fig. 9 is a diagram illustrating the configuration of a learning device according to one embodiment.

[0095] Referring to FIG. 9, the learning device (900) may include a processor (910) and a memory (920). The learning device (900) may correspond to the learning device described herein.

[0096] The memory (920) is coupled to the processor (910) and can store instructions executable by the processor (910), data to be calculated by the processor (910), or data processed by the processor (910). The memory (920) can include a non-transitory computer-readable medium, such as a high-speed random access memory and / or a non-volatile computer-readable storage medium (e.g., one or more disk storage devices, flash memory devices, or other non-volatile solid-state memory devices).

[0097] The processor (910) may perform one or more operations related to the operation of the learning device described herein. For example, the processor (910) may control the learning device (900) to collect learning data including user data and medical data for each patient with sudden hearing loss, and to detect a hearing loss frequency range for each patient with sudden hearing loss based on hearing data for each patient with sudden hearing loss included in the medical data.

[0098] The processor (910) can control the learning device (900) to detect a hearing damage frequency range of each patient with sudden hearing loss based on a continuous frequency range in which the hearing threshold of each patient with sudden hearing loss is shown to be above a threshold value among the hearing data of each patient with sudden hearing loss.

[0099] The processor (910) can control the learning device (900) to generate label data based on the detected hearing damage frequency range and whether each of the sudden hearing loss patients has recovered, and to train a plurality of machine learning models included in the prognosis prediction model based on the learning data and the label data. The processor (910) can control the learning device (900) to obtain prediction information on the prognosis of each of the sudden hearing loss patients from each of the plurality of machine learning models included in the prognosis prediction model by inputting the learning data into the prognosis prediction model, and to train the plurality of machine learning models included in the prognosis prediction model based on the label data and the prediction information on the prognosis of each of the sudden hearing loss patients.

[0100] The embodiments described above may be implemented using hardware components, software components, and / or a combination of hardware components and software components. For example, the devices, methods, and components described in the embodiments may be implemented using a general-purpose computer or a special-purpose computer, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing instructions and responding to them. The processing device may execute an operating system (OS) and software applications running on the operating system. Furthermore, the processing device may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing device is sometimes described as being used alone; however, one of ordinary skill in the art will recognize that the processing device may include multiple processing elements and / or multiple types of processing elements. For example, a processing unit may include multiple processors, or a processor and a controller. Other processing configurations, such as parallel processors, are also possible.

[0101] Software may include computer programs, codes, instructions, or a combination of one or more of these, which may configure a processing device to perform a desired operation or may, independently or collectively, command the processing device. The software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium, or device for interpretation by the processing device or for providing instructions or data to the processing device. The software may also be distributed over networked computer systems and stored or executed in a distributed manner. The software and data may be stored on a computer-readable recording medium.

[0102] The method according to the embodiment may be implemented in the form of program commands that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may store program commands, data files, data structures, etc., alone or in combination, and the program commands recorded on the medium may be those specially designed and configured for the embodiment or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks, and magnetic tapes, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, and hardware devices specially configured to store and execute program commands, such as ROMs, RAMs, and flash memories. Examples of program commands include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter, etc.

[0103] The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

[0104] Although the embodiments described above have been described with limited drawings, those skilled in the art will appreciate that various technical modifications and variations can be applied based on the described embodiments. For example, appropriate results can still be achieved even if the described techniques are performed in a different order than described, and / or components of the described systems, structures, devices, circuits, etc. are combined or combined in a different manner than described, or are replaced or substituted with other components or equivalents.

[0105] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims described below.

Claims

1. In a prognostic diagnostic method for predicting the prognosis of patients with sudden hearing loss, Step of collecting user data and medical data on patients with sudden hearing loss; A step of detecting a hearing damage frequency range of a patient with sudden hearing loss based on hearing data of the patient with sudden hearing loss included in the medical data; A step of determining a hearing threshold average value including an average value of the hearing threshold of the affected ear and an average value of the hearing threshold of the non-affected ear for the detected hearing impairment frequency range, and variable values ​​categorized into each of the average value of the hearing threshold of the affected ear and the average value of the hearing threshold of the non-affected ear; A step of obtaining prediction information on the prognosis of sudden hearing loss of the patient with sudden hearing loss from each of a plurality of machine learning models included in the prognosis prediction model by inputting input data including the average value of the hearing threshold and the categorized variable values ​​into the prognosis prediction model; and A step of determining prognostic diagnostic data for sudden hearing loss based on prediction information on the prognosis of sudden hearing loss of the patient with sudden hearing loss obtained from each of the machine learning models included in the prognostic prediction model. Including, How to diagnose prognosis.

2. In paragraph 1, The step of detecting the hearing loss frequency range of the above sudden hearing loss patient is, A step of detecting a hearing damage frequency range of a patient with sudden hearing loss based on a continuous frequency range in which a hearing threshold is shown to be above a threshold value among the hearing data of the patient with sudden hearing loss, How to diagnose prognosis.

3. In paragraph 1, Predictive information on the prognosis of sudden hearing loss in the above patients with sudden hearing loss is as follows: Including the probability of recovery from sudden hearing loss in the above sudden hearing loss patient, How to diagnose prognosis.

4. In paragraph 1, The above prognosis prediction model is, Characterized in that it is an ensemble model including the above multiple machine learning models, How to diagnose prognosis.

5. In paragraph 1, The step of determining the above prognostic diagnostic data is: A step of determining prognostic diagnostic data for sudden hearing loss indicating recovery or non-recovery of sudden hearing loss of the patient based on the average of the recovery probability for sudden hearing loss of the patient included in the prediction information for the prognosis of sudden hearing loss of the patient obtained from each of the machine learning models. Including, How to diagnose prognosis.

6. In a learning method for learning a prognostic prediction model, A step of collecting learning data including user data and medical data for each patient with sudden hearing loss; A step of detecting a hearing damage frequency range of each of the sudden hearing loss patients based on the hearing data of each of the sudden hearing loss patients included in the medical data; A step of generating label data based on the detected hearing impairment frequency range and whether each of the sudden hearing loss patients has recovered; and A step of training multiple machine learning models included in the prognosis prediction model based on the above learning data and the above label data. Including, How to learn.

7. In paragraph 6, The above prognosis prediction model is, Characterized by being an ensemble model including multiple machine learning models, How to learn.

8. In paragraph 6, The step of detecting the hearing loss frequency range of each of the above sudden hearing loss patients is as follows. A step of detecting a hearing damage frequency range of each of the sudden hearing loss patients based on a continuous frequency range in which the hearing threshold is shown to be above a threshold value among the hearing data of each of the sudden hearing loss patients, How to learn.

9. In paragraph 6, The above learning steps are: A step of obtaining prediction information on the prognosis of each of the sudden hearing loss patients from each of the plurality of machine learning models included in the prognosis prediction model by inputting the learning data into the prognosis prediction model; and A step of training multiple machine learning models included in the prognosis prediction model based on the label data and the prediction information on the prognosis of each of the sudden hearing loss patients. Including, How to learn.

10. A computer program stored on a computer-readable recording medium to execute the method of claim 1 by being combined with hardware.

11. In a prognostic diagnostic device for predicting the prognosis of patients with sudden hearing loss, Including memory and processor, The above memory stores instructions executable by the processor, When the above instructions are executed by the processor, the processor causes the prognostic diagnostic device to: Collect user data and medical data on patients with sudden hearing loss, Detecting a hearing damage frequency range of a patient with sudden hearing loss based on the hearing data of the patient with sudden hearing loss included in the medical data, The average value of hearing threshold including the average value of hearing threshold of the affected ear and the average value of hearing threshold of the non-affected ear for the detected hearing impairment frequency range, and the variable value categorized by each of the average value of hearing threshold of the affected ear and the average value of hearing threshold of the non-affected ear, are determined. By inputting input data including the above hearing threshold average value and the above categorized variable values ​​into a prognosis prediction model, prediction information on the prognosis of the sudden hearing loss patient is obtained from each of a plurality of machine learning models included in the prognosis prediction model, Controlling the prognostic diagnosis device to determine prognostic diagnosis data for the sudden hearing loss based on the prediction information on the prognosis of the sudden hearing loss patient obtained from each of the machine learning models included in the prognostic prediction model. Prognostic diagnostic device.

12. In paragraph 11, The above processor is the prognostic diagnostic device, Controlling the prognostic diagnosis device to detect a hearing damage frequency range of the sudden hearing loss patient based on a continuous frequency range in which the hearing threshold of the sudden hearing loss patient is shown to be above a threshold value among the hearing data of the sudden hearing loss patient. Prognostic diagnostic device.

13. In paragraph 11, Predictive information on the prognosis of sudden hearing loss in the above patients with sudden hearing loss is as follows: Including the probability of recovery from sudden hearing loss in the above sudden hearing loss patient, Prognostic diagnostic device.

14. In paragraph 11, The above prognosis prediction model is, Characterized in that it is an ensemble model including the above multiple machine learning models, Prognostic diagnostic device.

15. In paragraph 11, The above processor is the prognostic diagnostic device, Controlling the prognostic diagnosis device to determine prognostic diagnosis data for sudden hearing loss indicating recovery or non-recovery of sudden hearing loss of the sudden hearing loss patient based on the average of the recovery probability for sudden hearing loss of the sudden hearing loss patient included in the prediction information for the prognosis of sudden hearing loss of the sudden hearing loss patient obtained from each of the machine learning models; Prognostic diagnostic device.

16. In a learning device that learns a prognostic prediction model, Including memory and processor, The above memory stores instructions executable by the processor, When the above instructions are executed by the processor, the processor causes the learning device to: Collect training data including user data and medical data for each patient with sudden hearing loss, Detecting the hearing damage frequency range of each of the sudden hearing loss patients based on the hearing data of each of the sudden hearing loss patients included in the above medical data, Generate label data based on the detected hearing loss frequency range and whether each of the patients with sudden hearing loss recovered, Controlling the learning device to train a plurality of machine learning models included in the prognosis prediction model based on the learning data and the label data; Learning device.

17. In paragraph 16, The above prognosis prediction model is, Characterized by being an ensemble model including multiple machine learning models, Learning device.

18. In paragraph 16, The above processor is the learning device, Controlling the learning device to detect a hearing damage frequency range of each of the sudden hearing loss patients based on a continuous frequency range in which the hearing threshold is shown to be above a threshold value among the hearing data of each of the sudden hearing loss patients. Learning device.

19. In Article 16, The above processor is the learning device, By inputting the above learning data into the above prognosis prediction model, prediction information on the prognosis of each of the sudden hearing loss patients is obtained from each of the multiple machine learning models included in the above prognosis prediction model, Controlling the learning device to train a plurality of machine learning models included in the prognosis prediction model based on the label data and the prediction information on the prognosis of each of the sudden hearing loss patients. Learning device.

Images