JP2024056292A - Diagnostic support information providing system, diagnostic support device, symptom index providing device, and symptom index providing program - Google Patents

Abstract

To provide a symptom index providing device for providing a symptom index on a vasomotor nerve symptom or a depressive symptom of a woman with menopause.SOLUTION: On the basis of brain wave data that changes with time on a sleep period of a subject acquired by an electroencephalograph 10, a sleep stage prediction unit 3110 of a symptom index providing device 3000 determines, with time, which of a plurality of sleep stages the sleep state of the subject is applicable at predetermined time intervals. After data on the sleep stage that changes with time is subjected to smoothing processing, and is converted to a frequency component, a depressive symptom prediction unit 3240 and a menopausal symptom prediction unit 3250 output a symptom index by a learned model generated by machine learning with the frequency component as an input.SELECTED DRAWING: Figure 4

Description

Application for application of Article 30, Paragraph 2 of the Patent Act has been submitted. 1) Published on October 30, 2021 in the Program and Abstracts of the 36th Academic Meeting of the Japanese Society of Women's Medicine, Vol. 29, No. 1, ISSN 2185-8861, p. 102. 2) Published at the Japanese Society of Women's Medicine Academic Meeting, hosted by the Japanese Society of Women's Medicine, on November 6, 2021, with the main theme "Women's Medicine: Comprehensive Gatekeeper of Women's Health."

The present invention relates to a diagnostic support device and diagnostic support information providing system that assists in distinguishing between vasomotor symptoms and depressive symptoms in menopausal women, and a symptom index providing device and symptom index providing program for providing symptom indexes related to menopausal symptoms.

"Menopause" refers to the five years before and after menopause in women, which for Japanese people is the period from about 45 to 55 years of age on average.

Here, "menopause" refers to a state in which ovarian activity gradually disappears and menstruation finally stops permanently. When menstruation has ceased for 12 months or more, menopause is considered to have occurred one year prior. The average age of menopause for Japanese women is about 50 years old, but there is a large degree of individual variation, with some women reaching menopause as early as their early 40s and others as late as their late 50s (see Non-Patent Document 1).

As mentioned above, the 10 years consisting of the 5 years before menopause and the 5 years after menopause are called the "menopause", and symptoms that appear during the menopause that are not caused by organic changes are called "menopausal symptoms", and among these symptoms, pathological conditions that interfere with daily life are defined as "menopausal disorders".

The main cause of menopausal symptoms is said to be the large fluctuations and decline of female hormones (estrogen) (endocrine factors), and it is believed that the onset of menopausal symptoms is due to a complex combination of physical factors such as aging, psychological factors such as upbringing and personality, and social factors such as interpersonal relationships at work and at home.

The social and psychological backgrounds of menopausal women include i) the independence of grown children (employment and marriage), ii) the death of a husband and retirement, iii) conflicts between spouses and with children, iv) feelings of alienation and loneliness from society and family, v) the loss of close relatives and friends, and vi) changes in appearance and features due to aging (see Non-Patent Document 2).

As a result, the state of feeling unwell physically or mentally without a clear cause is called "vague symptoms."

Menopausal symptoms are diverse and can be categorized into vasomotor symptoms such as hot flashes and sweating, mental symptoms such as insomnia, anxiety, and depression, and physical symptoms such as headaches, back pain, joint pain, and dry skin.

According to statistics, approximately 55% of the menopausal generation suffer from menopausal symptoms, and it has been reported that approximately 30% of the premenopausal generation are aware of menopausal-like symptoms and believe that they are in menopause (see Non-Patent Document 3).

And it is characteristic of Japanese people that they are more likely to have not only vasomotor symptoms (easily sweating, cold face and hands and feet, hot face), but also neurological symptoms (easily angered, irritable, difficulty falling asleep, shallow sleep, worrying, depression).

(Unmet needs in menopausal care)

Currently, in clinical practice, the diagnosis of menopausal disorders is made through interviews. For example, blood tests and current clinical tests are only useful for exclusion diagnosis.

In other words, there are currently few objective indicators for diagnosing menopausal disorders, and because blood estrogen and follicle-stimulating hormone levels fluctuate greatly up until about two years after menopause, their diagnostic usefulness has not been established. Furthermore, blood estradiol (E2) levels are generally low after complete menopause, but E2 levels do not correlate with the severity of menopausal symptoms, and diagnosis relies solely on questionnaires, i.e., the individual's self-reporting. For this reason, in actual clinical practice, it is particularly difficult to differentiate from mental disorders such as masked depression and anxiety disorders.

On the other hand, when obstetricians-gynecologists and psychiatrists performed detailed diagnoses on 2,210 patients who visited the hospital complaining of menopausal symptoms, it was reported that 25.9% of patients were thought to have estrogen deficiency disorder, 27.4% were thought to have mood disorders (13.1% were depressed), and 11.2% were thought to have anxiety disorders (see Non-Patent Document 4).

A detailed examination of patients who visited the hospital complaining of menopausal symptoms has shown that half of the patients had autonomic nervous system disorders, while the other half had mental disorders. Of these, mood disorders, primarily depression, and anxiety disorders were most prevalent.

If the symptoms are estrogen deficiency, hormone replacement therapy is effective, but if the symptoms are depression, psychotherapy is required, and establishing a method for differential diagnosis is extremely important for the health care of menopausal women.

(Electroencephalography as a biosignal and its application in the medical and health fields)

On the other hand, as a method for objectively measuring the state of a living organism, the brain waves and blood oxygen saturation of the living organism (human body) have been acquired as electrical signals to measure the state of the living organism. A device that can acquire brain waves and blood oxygen saturation by attaching it to the forehead has been proposed as a device for acquiring brain waves and blood oxygen saturation (see, for example, Patent Document 1).

Changes in electrical potential that occur with activity within the body, such as brain waves and electrocardiograms, can be measured through electrodes attached to the outside of the body. Changes in electrical potential measured in this way have traditionally been used as biosignals in hospitals, research institutions, and other settings to non-invasively determine the presence and severity of abnormalities within the body.

However, tests using conventional testing equipment do not place a significant physical burden on subjects if they are carried out for a relatively short period of time in specific locations such as those mentioned above. However, depending on the purpose of the test and the subject, the burden may be inappropriately large.

A specific example would be a test to obtain biosignals during sleep. The multiple electrodes that are attached, the ointment applied to the areas where the electrodes are attached, and the contact of the cables connecting the electrodes to the main unit with the skin can give the subject a strong feeling of discomfort or make them unable to fall asleep or affect the electrical activity occurring within the subject's body. As a result, it is not possible to obtain biosignals from the subject's normal sleeping state.

Patent Document 2 discloses a biosignal measuring device that reduces the feeling of discomfort and discomfort felt by the subject, and combines ease of measurement with high performance. The device is configured to acquire biosignals using multiple thin-film electrodes, which are exposed on one side of an attachment part that is an adhesive sheet, and both the electrodes for acquiring signals and the attachment part are flexible enough to stretch and bend freely, and when attached to the subject's body surface, they follow the shape and movement of the place where they are attached. Here, the acquired biosignals are transmitted to an external device as wireless signals.

Patent Document 3 discloses a device and program for evaluating the sleep state and verifying so-called sleep apnea syndrome by further providing an optical element that measures blood oxygen saturation in a sheet-type electroencephalograph like that disclosed in Patent Document 2.

The relationship between such EEG measurements and psychiatric and menopausal symptoms is outlined below.

In other words, vasomotor symptoms are thought to be caused by a maladaptation of the thermoregulatory function in the hypothalamus due to changes in the secretion of neurotransmitters such as norepinephrine and serotonin, and are known to affect sleep, suggesting that insomnia may be caused by changes in circadian rhythms and vasomotor symptoms.

However, due to the complexity of sleep EEG measurement, there have been very few reports of menopausal-specific changes. Focusing on sleep parameters, the only reports have shown that women with vasomotor symptoms have reduced total sleep time and increased frequency of awakenings during the night compared to women without vasomotor symptoms.

However, total sleep time and the number of awakenings vary greatly from person to person, and even within an individual, they are greatly influenced by the sleep environment, so it is difficult to determine whether a subject has vasomotor symptoms based on their individual sleep EEG using conventional sleep parameter analysis.

In the case of depression, the effect of the disease on sleep electroencephalograms has been reported, and there are several reports of diagnosing depression using electroencephalograms; however, even looking at the most recent reviews, none of these have been put to practical use and are on the market.
Furthermore, in relation to the diagnosis of menopausal symptoms as described above, a technology has been reported that uses artificial intelligence technology to analyze biological signals such as electroencephalograms as input, for example, to predict the probability of a subject experiencing "hot flashes." However, this technology does not attempt to use electroencephalogram data covering the subject's sleeping hours to provide auxiliary information at a level that would be useful in distinguishing menopausal symptoms (see Patent Document 4).

US Patent Publication No. 2016/0015281 International Publication No. WO2017-122379 JP 2018-164725 A JP 2020-014841 A

Japan Society of Obstetrics and Gynecology, Public Interest Foundation, Menopausal Disorders, Updated June 16, 2018 https://www.jsog.or.jp/modules/diseases/index.php?content_id=14 Knowing about menopause and staying true to yourself, Part 16: Let's take menopause as an opportunity to reassess our lifestyle, Yomiuri Shimbun, September 2, 2019 https://www.hisamitsu.co.jp/hrt/medicaltribune/medicaltribune_16.html Docomo Healthcare Co., Ltd. One in three pre-menopausal women are aware that "I might be in menopause..." ~ Survey on menopausal symptoms among working women ~ August 30, 2018 https://prtimes.jp/main/html/rd/p/000000107.000016519.html Naohisa Ushiroyama, Journal of the Japan Society of Obstetrics and Gynecology, Vol. 53, No. 182-6, 2012

In light of the above background, in order to provide appropriate treatment for menopausal disorders, it is extremely important that doctors are able to distinguish between estrogen deficiency symptoms and psychiatric disorders based on objective indicators.

The present invention aims to provide a diagnostic support device and a diagnostic support information providing system capable of providing information to assist doctors in distinguishing between vasomotor symptoms and depressive symptoms in menopausal women.
Another object of the present invention is to provide a symptom index providing device and a symptom index providing program for providing a symptom index relating to vasomotor symptoms or depressive symptoms of menopausal women.

According to one aspect of the present invention, a diagnostic auxiliary information providing system for providing diagnostic auxiliary information for menopausal symptoms in a subject includes an electroencephalogram sensor for acquiring an electroencephalogram signal from the subject, a storage means for storing time-course electroencephalogram data for the subject's sleep period based on the electroencephalogram signal from the electroencephalogram sensor, a sleep stage determination means for determining over time at predetermined time intervals which of a plurality of sleep stages the subject's sleep state corresponds to, using the time-course electroencephalogram data as an input, a smoothing processing means for performing a temporal smoothing process upon receiving a result of the time-course sleep stage determination, a frequency conversion means for receiving an output of the smoothing processing means and converting it into frequency components, and a symptom prediction means for outputting diagnostic auxiliary information for the subject based on the electroencephalogram signal from the subject, wherein the symptom prediction means includes a first trained model that is generated in advance by machine learning using frequency components of the electroencephalogram data measured for a plurality of training subjects and corresponding clinical indicators of menopausal symptoms as learning data, and that outputs diagnostic auxiliary information using frequency components converted by the frequency conversion means for the electroencephalogram data from the subject as input.
Preferably, the smoothing process is a moving triangle average process.
Preferably, the symptom prediction means includes a second trained model which is generated in advance by machine learning using frequency components of EEG data measured on a plurality of training subjects and corresponding clinical indicators of depressive symptoms as training data, and which outputs diagnostic auxiliary information using frequency components converted by a frequency conversion means for the EEG data from the subjects as input.
Preferably, the sleep stage determination means includes first and second convolutional neural networks having different filter sizes for receiving as input multiple channels of EEG data from electrodes placed at different positions on the subject's head in the EEG sensor and extracting features from the EEG data, and a bidirectional recurrent neural network for receiving as input outputs of the first and second convolutional neural networks and outputting the sleep stage.
According to another aspect of the present invention, there is provided a diagnostic support device for providing diagnostic support information for menopausal symptoms of a subject based on time-course electroencephalogram data regarding the sleep period of the subject acquired by an electroencephalogram sensor, the device comprising: a sleep stage determination means for receiving the electroencephalogram data acquired by the electroencephalogram sensor, and determining at a predetermined time interval which of a plurality of sleep stages the sleep state of the subject corresponds to over time, using the time-course electroencephalogram data as an input; a smoothing processing means for performing a temporal smoothing process upon receiving a result of the time-course sleep stage determination; a frequency conversion means for receiving an output of the smoothing processing means and converting it into frequency components; and a symptom prediction means for returning diagnostic support information to the subject based on the electroencephalogram signal from the subject, the symptom prediction means including a first trained model that is generated in advance by machine learning using frequency components of the electroencephalogram data measured for a plurality of training subjects and corresponding clinical indicators of menopausal symptoms as training data, and that outputs diagnostic support information using frequency components converted by the frequency conversion means for the electroencephalogram data from the subject as input.
Preferably, the smoothing process is a moving triangle average process.
Preferably, the symptom prediction means includes a second trained model which is generated in advance by machine learning using frequency components of EEG data measured on a plurality of training subjects and corresponding clinical indicators of depressive symptoms as training data, and which outputs diagnostic auxiliary information using frequency components converted by a frequency conversion means for the EEG data from the subjects as input.
According to yet another aspect of the present invention, there is provided a symptom index providing device for providing a symptom index for menopausal symptoms of a subject based on time-course electroencephalogram data about the sleep period of the subject acquired by an electroencephalogram sensor, the symptom index providing device comprising: sleep stage determination means for receiving the electroencephalogram data acquired by the electroencephalogram sensor, and determining at predetermined time intervals over time which of a plurality of sleep stages the sleep state of the subject corresponds to, using the time-course electroencephalogram data as an input; smoothing processing means for performing a temporal smoothing process upon receiving a result of the time-course sleep stage determination; frequency conversion means for receiving an output of the smoothing processing means and converting it into frequency components; and symptom prediction means for outputting a symptom index for the subject based on the electroencephalogram signal from the subject, the symptom prediction means including a first trained model that is generated in advance by machine learning using frequency components of the electroencephalogram data measured for a plurality of training subjects and corresponding clinical indicators of menopausal symptoms as training data, and that outputs a symptom index using frequency components converted by the frequency conversion means for the electroencephalogram data from the subject as input.
Preferably, the smoothing process is a moving triangle average process.
Preferably, the symptom prediction means includes a second trained model which is generated in advance by machine learning using frequency components of EEG data measured on a plurality of training subjects and corresponding clinical indicators of depressive symptoms as training data, and which outputs a symptom indicator using frequency components converted by a frequency conversion means for the EEG data from the subjects as input.
According to yet another aspect of the present invention, there is provided a symptom index providing program for causing a computer to function as the sleep stage determining means, the smoothing processing means, the frequency conversion means, and the symptom predicting means recited in claim 8.

According to the present invention, it is possible to obtain an objective symptom index for the menopausal symptoms of a subject based on electroencephalogram data covering the subject's sleeping hours.
In addition, based on EEG data from subjects exhibiting menopausal symptoms over their sleep time, it is possible to obtain objective symptom indicators that assist in distinguishing between depressive symptoms and vasomotor symptoms.

1 is a conceptual diagram for explaining a usage mode of a diagnostic auxiliary information providing system 1000 including an electrode sheet-type electroencephalograph 10', a measurement terminal 2000, and an information providing server 3000 according to the present embodiment. FIG. 2 is a diagram showing the appearance of an electrode sheet-type electroencephalograph 10'. 1 is a block diagram showing an outline of the configuration of an electrode sheet-type electroencephalograph 10' according to the present embodiment. FIG. FIG. 1 is a functional block diagram for explaining the configuration of a diagnostic assistance information providing system 1000. FIG. 2 is a block diagram for explaining the configuration of an artificial intelligence model in a sleep stage prediction unit. FIG. 13 is a diagram showing the results of sleep stage discrimination for each epoch for a certain subject using an electrode sheet-type electroencephalograph 10'. FIG. 13 is a diagram showing the results of smoothing the changes in sleep stages over time by smoothing processing (triangular moving average processing). A diagram for explaining the configuration of the learning model and the learning process in the depressive symptom prediction unit 3240 and the menopausal symptom prediction unit 3250. FIG. 2 is a block diagram for explaining the hardware configuration of the measurement terminal 2000. FIG. 5 is a block diagram for explaining a hardware configuration of an information providing server 3000 shown in FIG. 4. FIG. 1 shows a self-assessment questionnaire for calculating the Simplified Menopausal Index (SMI). FIG. 1 is a diagram showing a self-rating depression scale checklist. FIG. 1 is a flowchart showing the process of selecting cases and a diagram showing the background of the cases. FIG. 1 is a diagram showing the evaluation of sleep quality according to the SMI index. FIG. 1 shows the evaluation of sleep quality by the SDS index. A figure showing the final diagnostic accuracy of electroencephalogram measurements of menopausal women using an electrode sheet-type electroencephalograph 10' and its artificial intelligence analysis.

The following describes the embodiment with reference to the drawings.

The embodiments described below are all comprehensive or specific examples. The numerical values, shapes, materials, components, arrangement positions and connection forms of the components shown in the following embodiments are merely examples and are not intended to limit the present invention. Furthermore, among the components in the following embodiments, components that are not described in an independent claim that indicates a superordinate concept are described as optional components.

In addition, the biosignal measuring device and diagnostic auxiliary information providing system 1000 according to the following embodiments, or a program for operating them, are devices, systems, and programs capable of providing objective information to assist doctors in making diagnoses of biological diseases, more specifically, menopausal symptoms or disorders.
As will be described later, such a system and program are not only intended to provide information to assist a doctor's diagnosis, but can also function as a symptom index providing device and a symptom index providing program for providing symptom indexes that serve as reference information on menopausal symptoms in the so-called healthcare field.

Therefore, this does not exclude cases where such "supplementary objective information" is provided as a reference indicator that an individual can use as their own health information before receiving a doctor's diagnosis.

In each embodiment, the diagnostic assistance information providing system 1000 is composed of a biosignal measuring device 10, a measurement terminal 2000, and an information providing server 3000.

In addition, the biosignal measuring device 10 uses an electroencephalograph configured as an electrode sheet with electrodes that acquire signals from the living body, as described below.

Therefore, in the following, the biosignal measuring device 10 configured in this way will be referred to as an electrode sheet-type electroencephalograph 10' and will be described as an example of a device that measures the subject's brain waves.

However, the configuration of the device that measures the subject's brain waves and the configuration that provides the sleep state determination and supplementary information on menopausal disorders from the acquired brain wave data are not limited to such configurations. For example, the measurement terminal 2000 and the information provision server 3000 may be an integrated computer, and an electroencephalograph with a configuration other than an electrode sheet-type electroencephalograph may be used as the device that measures the brain waves. However, as described in the prior art, it is desirable to realize biosignal measurement that combines ease of measurement with high performance by using an electrode sheet-type electroencephalograph 10' to measure the user's brain waves while sleeping, while minimizing the feeling of wearing and discomfort that the subject feels.

Figure 1 is a conceptual diagram for explaining the usage of a diagnostic assistance information providing system 1000 including an electrode sheet-type electroencephalograph 10', a measurement terminal 2000, and an information providing server 3000 according to this embodiment.

As shown in FIG. 1, the electrode sheet-type electroencephalograph 10' measures the brain waves of the subject 2 during sleep. The electrode sheet-type electroencephalograph 10' is attached to the forehead of the subject 2 (human body) when in use. The electrode sheet-type electroencephalograph 10' is formed with a number of electrodes that acquire electrical signals based on the brain waves of the subject 2 (hereinafter simply referred to as "brain waves"). The electrode sheet-type electroencephalograph 10' also includes a communication unit (not shown) that can wirelessly communicate the brain waves acquired by the electrodes, and transmits the acquired brain wave data to the measurement terminal 2000.

The EEG data may be transferred from the electrode sheet-type EEG device 10' to the measurement terminal 2000 in parallel with the measurement, or the electrode sheet-type EEG device 10' may store and hold EEG data for a certain period, and transfer the EEG data for the specified period all at once to the measurement terminal 2000.

The measurement terminal 2000 is communicatively connected to the information providing server 3000 via the network 4, and the information providing server 3000 performs processing as described below based on the brainwave data received from the measurement terminal 2000 to calculate auxiliary information for the menopausal symptoms or menopausal disorders of the subject 2. The calculated auxiliary information is returned from the information providing server 3000 to the measurement terminal 2000 and displayed on the measurement terminal 2000.

In this embodiment, "acquiring brain waves" means that the electrode sheet-type electroencephalograph 10' measures brain wave signals from the subject 2 and converts them into time-dependent brain wave data.

The electrodes contact the forehead of the subject 2 to obtain the brain waves of the subject 2 in real time. This allows the electrodes to obtain the brain waves of the subject 2 over time. Note that multiple electrodes are provided to obtain brain waves at different positions on the forehead of the subject 2.

Figure 2 shows the external appearance of the electrode sheet-type electroencephalograph 10'.

As shown in FIG. 2(a), the electrode sheet-type electroencephalograph 10' is composed of a patch-type electroencephalograph 140 and a disposable sheet 110 with electrodes.

The disposable sheet with electrodes 110 is connected to the patch-type electroencephalograph 140 via a plug connector.

As shown in FIG. 2(b), the electrode sheet-type electroencephalograph 10' does not need to be worn by the subject like headgear, but can simply be attached to the user's forehead with double-sided tape. Electrodes 2-7 are evenly placed on the user's forehead, and electrode 1 is a reference electrode that is placed away from the forehead and can be placed, without any particular limitation, for example, behind the ear so as to be in electrical contact with the user's skin.

Although not limited to this, the electrode sheet-type EEG monitor 10' may be a patch-type EEG monitor HARU-1 manufactured by PGV Corporation or its improved version, Haru-2, which has been certified as a medical device (HARU-1: Class II, Certification Number: 302AFBZX00079000, Haru-2: Class II, Certification Number: 304AFBZX00012000).

(Configuration of electrode sheet type electroencephalograph 10')

The electrode sheet-type electroencephalograph 10' is also disclosed in International Publication No. WO2017-122379 (Patent Document 2), so its configuration will be briefly described below.

Figure 3 is a block diagram showing an outline of the configuration of the electrode sheet-type electroencephalograph 10' according to this embodiment.

The electrode sheet-type electroencephalograph 10' is a device that uses electrodes to measure changes in the electrical potential occurring in a living body over time, obtains the changes as biosignals, and outputs the data on these biosignals. The changes in the electrical potential of the living body referred to here are, for example, changes in weak electrical potential (tens of microvolts to tens of millivolts) caused by activity of the brain, heart, or other muscles. In the following description, the biosignals will be described as brain waves.

The electrode sheet-type electroencephalograph 10' includes an input section 100, a control section 200, a power supply section 300, and an output section 400.

The input unit 100 is provided on a disposable sheet 110 with electrodes, measures the potential inside the body that can be detected through the skin of a living body, and outputs a signal indicating the time series change of the measured potential to the control unit 200. The input unit 100 includes a signal acquisition unit 110, an attachment unit 115, an amplifier unit 120, and an analog-digital converter (referred to as an A/D converter in FIG. 1 and below) 130. Although not particularly limited, the amplifier unit 120 and the A/D conversion unit 130 may be configured to be provided inside the patch-type electroencephalograph 140, or may be configured to be provided in the plug connector.

The signal acquisition unit 110 is a plurality of thin-film electrodes, and is formed so as to be exposed on one side of the attachment unit 115, which is an adhesive sheet. In the following, the surface of the attachment unit 115 on which the signal acquisition unit 110 is exposed is referred to as the front side of the attachment unit 115, and the opposite side is also referred to as the back side. The attachment unit 115 is attached to the living body so that the front side is in contact with the skin of the living body. The signal acquisition unit 110 detects the potential in this state. The signal acquisition unit 110, which is a plurality of electrodes, can measure the potential respectively, and the biosignals acquired by measuring the potential with each electrode are input to the control unit 200 via the amplifier unit 120 and the A/D converter 130 through individual transmission paths (channels). In addition, these electrodes include one or more reference electrodes and one or more measurement electrodes, and the biosignals are acquired by measuring the potential difference between each measurement electrode and the reference electrode corresponding to the measurement electrode.
The number and arrangement of the electrodes can be changed depending on the purpose of measurement and the content of analysis, and are not necessarily limited to the configuration shown in FIG.

Both the signal acquisition unit 110 and the attachment unit 115 are flexible and stretchable, allowing them to bend freely. When attached to the subject's body surface, they follow the shape and movement of the area where they are attached.

As an adhesive part 115 having such characteristics, an adhesive base material made of a silicone material, which is used as a base material for compresses, etc., is used. An adhesive part 115 made of such a material has high elasticity and flexibility as well as moisture permeability, so that discomfort and sweating of the subject when worn can be suppressed. In addition, such a material has high biocompatibility, so the adhesive part 115 can be worn for long periods of time.

In addition, a stretchable conductor using metal particles can be used as the material of the signal acquisition unit 110, which is an electrode. Furthermore, this stretchable conductor is also used as the material of the wiring formed in the attachment unit 115, which transmits signals from the signal acquisition unit 110 to the outside of the attachment unit 115. For example, silver nanowires are used as the metal particles, which can provide a conductivity of about 10,000 S/cm. Therefore, the subject's biosignal is input to the amplifier unit 120 described below through the low-resistance and low-noise electrodes and wiring. In other words, the signal acquisition unit 110 and wiring made of a stretchable conductor have both stretchability, suitable transparency and conductivity, and do not interfere with the subject's movements due to their stretchability.

The signal acquisition unit 110 and the attachment unit 110 on which the wiring is formed are an adhesive sheet that is stretchable and bendable as a whole, and can be worn for long periods of time.

The amplifier 120 amplifies the biosignal input from the signal acquisition unit 110 and outputs it to the A/D converter 130. Such an amplifier 120 may be realized as an amplifier circuit including an operational amplifier on a substrate.

The A/D converter 130 converts the biosignal, which is an analog signal input from the amplifier 120, into a digital signal and outputs it to the control unit 200. In order to obtain the same accuracy (resolution) as conventional medical devices, the A/D converter 130 of the electrode sheet-type electroencephalograph 10' in this embodiment is realized by an A/D converter that converts an analog signal into a digital signal with a bit count of 20 or more bits. For example, a 24-bit ΔΣ type A/D converter is used.

The control unit 200 controls the overall operation of the electrode sheet-type electroencephalograph 10'. The control unit 200 includes a central processing unit (hereinafter also referred to as a CPU (central processing unit), and indicated as a CPU in FIG. 3) 210, a baseband digital signal processing unit (hereinafter also referred to as a baseband DSP (digital signal processor), and indicated as a DSP in FIG. 3)) 220, a contact state detection unit 230, a clock generation unit 240, a power management unit 250, and a power stabilization unit 260.

The CPU 210 reads out and executes a program from a storage unit (not shown), and manages and controls the operation of each of the components that make up the electrode sheet-type electroencephalograph 10'.
This storage unit can also be used to temporarily store electroencephalogram data measured over a predetermined period of time.

The baseband DSP 220 performs a predetermined process on the digital signal input from the input unit 100 and outputs it to the outside via the output unit 400.

The contact state detection unit 230 is an electronic circuit (contact state detection circuit) that detects the contact state between the signal acquisition unit 110, which is an electrode, and the living body. Specifically, the contact state detection unit 230 measures the contact impedance between the signal acquisition unit 110 and the skin of the living body, and judges whether the contact state is good or bad based on the magnitude of this contact impedance. For example, this judgment is performed when the electrode sheet type electroencephalograph 10' is worn and during measurement of the biological signal, thereby ensuring a proper wearing state of the electrode sheet type electroencephalograph 10'. The criteria for judging the contact state to be bad may be changed as appropriate depending on the settings. For example, when the signal input path provided in the input unit 100 has six channels, it may be judged to be bad if the electrode is removed from the subject (lead off) in at least one channel out of the six channels, or it may be judged to be bad if lead off is detected in the majority of the four channels.

The clock generation unit 240 is a clock generation circuit that generates the internal clock of the CPU 210.

The power management unit 250 and the power stabilization unit 260, together with the power supply unit 300 described below, form a power supply circuit, and supply the power input from the power supply unit 300 as power to drive the control unit 200 and other components. The power management unit 250 includes, for example, a DC-DC converter, and the input unit 100 and the output unit 400 are each supplied with power of different voltages transformed by the power management unit 250. The power stabilization unit 260 is a regulator that stabilizes the supplied power.

The power supply unit 300 supplies the power required for the operation of each component of the electrode sheet-type electroencephalograph 10' via the power supply management unit 250 and the power stabilization unit 260. The power supply unit 300 includes a charging circuit 310 and a rechargeable battery 320.

The charging circuit 310 is a circuit that takes in power from outside the electrode sheet-type electroencephalograph 10' and supplies the taken-in power to the rechargeable battery 320 by controlling the voltage and current in order to charge the rechargeable battery 320. The charging circuit 310 has a connection terminal 330 for taking in power. This connection terminal 330 may be, for example, a power receiving terminal for taking in power via a cable from a commercial power outlet, or may be a port such as a USB (universal serial bus) that can receive power from an external device that can supply power.

The rechargeable battery 320 is a rechargeable battery that stores power to be supplied to each component of the electrode sheet-type electroencephalograph 10'. The rechargeable battery 320 is one with a capacity appropriate for the application. For example, it may be one with a capacity that allows long continuous use so that the subject can be monitored while sleeping.

The output unit 400 is a wireless communication module capable of wirelessly transmitting data, and outputs the biosignal output from the baseband DSP 220 in real time to an external device of the electrode sheet-type electroencephalograph 10', such as a storage device, a display device such as a monitor, or a measurement terminal 2000 equipped with these. In the present embodiment, the output unit 400 uses a communication method based on a communication standard such as Bluetooth (registered trademark) or Wi-Fi (registered trademark). In addition, in order to extend the continuous use time of the electrode sheet-type electroencephalograph 10', a low-power consumption communication method such as Bluetooth (registered trademark) Low Energy (BLE) may be used. In addition, when outputting in real time, the amount of information (transmission rate) has an upper limit due to the bandwidth of each communication method, so the sampling rate of the analog signal by the A/D converter 130 is changed depending on the communication method and the number of measurement channels.

The electrode sheet-type electroencephalograph 10' according to the present embodiment having the above configuration is attached to a living body by the attachment unit 115 when used. When used in this manner, the electrode sheet-type electroencephalograph 10' equipped with the above power supply unit 300 and output unit 400 does not require a power cable for constantly supplying power and a communication cable for constantly outputting a biological signal. Therefore, the subject wearing the electrode sheet-type electroencephalograph 10' does not feel the wearing sensation caused by these cables or have their movements restricted. In addition, even if lead-off occurs due to the subject's movements or sweating, the electrode sheet-type electroencephalograph 10' has redundancy by using multiple electrodes, and it is possible to monitor the wearing situation based on lead-off detection. As a result, if the electrode sheet-type electroencephalograph 10' is used, it is possible to perform monitoring for a long period of time with high certainty, for example.

Figure 4 is a functional block diagram for explaining the configuration of the diagnostic support information providing system 1000.

Referring to FIG. 4, the diagnostic assistance information providing system 1000 includes an electrode sheet-type electroencephalograph 10' for acquiring electroencephalogram data from a subject 2, a measurement terminal 2000, and an information providing server 3000.

The measurement terminal 2000 includes, but is not limited to, an EEG data recording and transmitting unit 2001 for recording 8 hours of EEG data for three channels obtained from the electrode sheet-type EEG meter 10' at a sampling rate of 250 Hz and transmitting the data to the information providing server 3000, and a display device 2008 for displaying information for assisting diagnosis returned as a result of the analysis of the EEG data by the information providing server 3000.

Note that the 8-hour period means that brainwave data is obtained for the entire duration of subject 2's sleep, and is not limited to this period.

The information providing server 3000 includes a sleep stage estimation unit 3100. The sleep stage estimation unit 3100 includes a sleep stage prediction unit 3110 that predicts a sleep stage based on 30 seconds of electroencephalogram data using an artificial intelligence model described later based on the electroencephalogram data acquired from the subject 2 transmitted from the measurement terminal 2000, a sleep stage recording unit 3120 that records information on the sleep stage labels output from the sleep stage prediction unit 3110 as one stage label every 30 seconds, a logic unit 3130 that calculates electroencephalogram variables, and a sleep variable recording unit 3140 that records the sleep variables calculated by the logic unit 3130 over time.
Although not particularly limited, the sleep variable recording unit 3140 may be configured to return all or part of the sleep variables recorded to the measurement terminal 2000 .

In this specification, the 30-second period during which the sleep stages are calculated is called an "epoch." However, the time interval for predicting the sleep stages is not limited to this interval.

The information providing server 3000 further includes a smoothing processing unit 3200 that performs smoothing processing on the temporal brain wave data.

The smoothing method is not particularly limited, but here we use the "Triangular Moving Average (TMA)". The Triangular Moving Average is an English name for the Triangular Moving Average, abbreviated as TMA, and is a moving average of moving averages, that is, it is calculated by first calculating a simple moving average and then calculating the moving average of that value.

The calculation method for the "Triangular Moving Average (TMA)" is as follows.
SMA = N-period Simple Moving Average Triangular Moving Average (TMA) = N-period Simple Moving Average of SMA

The "Triangular Moving Average (TMA)" is calculated by taking a double simple moving average in the above-mentioned way. In other words, it is a moving average line calculated by doubly smoothing, and has the characteristic that a smoother line is obtained than a simple moving average or an exponentially smoothed moving average line.
Although not particularly limited, for example, as a specific application example of this embodiment as described later, the results show that the highest discrimination accuracy was obtained when the window size of the first simple moving average in SMI was 15 and the size of the second simple moving average was also 15 as parameters of the triangular moving average. On the other hand, the results show that the highest discrimination accuracy was obtained when the window size of the first simple moving average in SDS was 10 and the size of the second simple moving average was also 10. From this viewpoint, the window size is preferably, for example, about 10 to 20. Note that the window size here is expressed in units of "epoch period".

The information providing server 3000 further includes a continuous waveform acquisition and recording unit 3210 that acquires and records the discrete data of the sleep stages as continuous waveform data after smoothing processing in the smoothing processing unit 3200, a frequency conversion unit 3220 that converts the continuous waveform data recorded by the continuous waveform acquisition and recording unit 3210 into frequency components, and a feature extraction unit 3230 that extracts features from the frequency data from the frequency conversion unit 3220.

The feature extraction unit 3230 is not particularly limited, but for example, extracts the intensity distribution of continuous waveform data for each predetermined frequency range.
For example, but not limited to, in the specific example described below, in order to easily see the frequency spectrum of the result, the sleep stage calculated with 1 point corresponding to 30 s (1 epoch period) is calculated with 1 point corresponding to 1 s. Therefore, it is equivalent to the original sampling rate being changed from 1/30 Hz to 1 Hz.
The sampling frequency was set to 1 Hz, and frequency intensities were calculated for each 0.008 Hz region in the range of 0 Hz to 0.08 Hz for both SMI and SDS.

Based on the features from the feature extraction unit 3230, the depressive symptom prediction unit 3240 calculates an index indicating a depressive tendency, and the menopausal symptom prediction unit 3250 calculates an index indicating menopausal symptoms.

The indices output from the depression symptom prediction unit 3240 and the menopausal symptom prediction unit 3250 are returned to the measurement terminal 2000 and displayed on the display device 2008 of the measurement terminal 2000.

In particular, but not limited to, the indicator may be information indicating only the presence or absence of a symptom tendency, such as "depression symptoms: yes/no" and "menopausal symptoms: yes/no." However, the indicator is not limited to this, and may be, for example, an indicator indicating the degree of each symptom (e.g., three levels, high, medium, low, etc.) on the premise of classification into multiple classes, rather than classification into two classes.
Alternatively, depending on the configuration of the classifier, the severity of each symptom may be presented as a numerical value representing a probability.
In this specification, the "symptom index" refers to an index that objectively indicates the degree of the subject's menopausal symptoms or depression symptoms, output by artificial intelligence that has undergone machine learning based on clinical data. In particular, when the information is used as a medical device to assist a doctor in making a diagnosis, it is called a "symptom index as diagnostic auxiliary information," and when the information is used as information indicating a health condition, particularly in the application in the healthcare field, it is called a "symptom index as reference information."

(Configuration of the sleep stage prediction section)

Figure 5 is a block diagram to explain the configuration of the artificial intelligence model in the sleep stage prediction unit.

An example will be described in which the deep neural network architecture DSN disclosed in the following document is adopted as the learning model.
Public literature 1: A. Supratak, H. Dong, C. Wu, and Y. Guo, “DeepSleepNet: A model
for automatic sleep stage scoring based on raw single-channel EEG”, IEEE Trans. Neural Syst. Rehabil. Eng., vol. 25, no. 11, pp. 1998-2008,Nov. 2017.

The learning model disclosed in the above document is called "DeepSleepNet (DSN)."

In the above paper, an algorithm is developed to extract features from EEG, and a model for automatic sleep stage scoring based on raw single-channel EEG is introduced. By utilizing the feature extraction capabilities of deep learning, the manual process of feature extraction is automated.

The features of this "DeepSleepNet" are as follows:

i) The model architecture uses two 1D convolutional neural networks (CNNs) with different filter sizes and a bidirectional long short term memory (BidLSTM) in the first layer. The CNN can learn filters to extract time-invariant features from raw single-channel EEG, while the bidirectional LSTM can be learned to encode temporal information, such as sleep stage transition rules, into the model.

ii) We implement a two-stage learning algorithm that can effectively train the model end-to-end using backpropagation while avoiding the class imbalance problem (i.e., learning to classify only the majority of sleep stages) that is difficult to train in large sleep datasets.

iii) Without modifying the model architecture and learning algorithm, the model can automatically learn features for sleep stage scoring from different raw single-channel EEG data sets with different characteristics (e.g., sampling rate) and scoring criteria (AASM and R&K) without using any manually designed features.
Here, AASM stands for the American Academy of Sleep Medicine, which defines the specifications for sleep tests, and R&K is a standardized test technique for sleep polysomnography (PSG) and sleep stage determination that was standardized by Rechtscaffen and Kales in 1968.

In other words, the DSN used in this embodiment was originally trained end-to-end based on single-channel sleep EEG and does not rely on manually created features. The DSN is composed of two main networks: a feature network using CNN that extracts time-invariant features, and a sequence network using bidirectional LSTM that determines the transition rules for sleep stages.

In the above-mentioned prior art document 1, the network is designed to input electroencephalograms from a single channel, but in the present embodiment, in order to construct a more robust model, electroencephalograms from multiple channels are input.
In addition, the LSTM is a model that improves on a simple recurrent neural network (RNN), and other bidirectional recurrent neural network configurations may be used as long as they are suitable for processing time-series (ordered) data.

The configuration and operation of such a sleep stage prediction unit are disclosed in the following literature:

Publication 2: SHOYA MATSUMORI, KOSEI TERAMOTO, HIROYA IYORI, TAKANORI SODA, SHUSUKE YOSHIMOTO, AND HARUO MIZUTANI, “HARU Sleep: A Deep Learning-Based Sleep Scoring System With Wearable Sheet-Type Frontal EEG Sensors”, IEEE Access, Volume 10, 2022

Referring to Figure 5, the feature network that constitutes the upper part of the model has two streams of convolutional networks, CNN1 and CNN2, that receive the same input but have different filter settings. Each convolutional layer stream starts with a 1D convolutional layer, followed by a batch normalization layer, a max pooling layer, and a dropout layer.

After that, three 1D convolutional layers and a max pooling layer are used, and finally the two output streams are flattened, concatenated, and fed into a dropout layer. Different filter settings, large and small, are applied to each stream, CNN1 and CNN2, to capture different levels of features.

The sequence network that constitutes the lower part of the model has a mainstream bidirectional LSTM layer that is bypassed by skip connections with a fully connected layer. The LSTM layer (recurrent architecture) is introduced to capture the temporal information present in different sleep stages, such as transition rules that sleep experts can take into account in the analysis. Bidirectional LSTM is an extension of LSTM that includes backpropagation in addition to forward propagation, allowing us to consider not only the relationship between past and current stages, but also future stages.

The model is trained in two phases: a pre-training phase, which optimizes the feature network and the sequence network, and a fine-tuning phase.

In the pre-training stage, a feature network is optimized using supervised learning for a simple classification problem, and a softmax layer is added to the end of the network to optimize the feature network. In the fine-tuning phase, a sequence network is added to the feature network trained in the pre-training stage, replacing the softmax layer, and both networks are optimized.

Although not particularly limited, the learning model can be configured as follows, in accordance with the above-mentioned publicly known document 1:

In other words, in the learning model, each CNN consists of four convolutional layers and two max pooling layers.

Each convolutional layer performs three operations in sequence: one-dimensional convolution with a filter, batch regularization, and activation function processing with a rectified linear unit (ReLU).

Each pooling layer downsamples the input using the max operation.

In the feature network, the parameters of CNN1 and CNN2 are selected to capture time and frequency information from the EEG. For example, in Figure 5, the filter size of the first Conv1D layer of CNN1 is set to Fs/2 (i.e., half the sampling rate (Fs)) and its stride size is set to Fs/16 so that it can detect when a specific EEG pattern appears.

On the other hand, the filter size of the conv1D layer of CNN2 is set to Fs×4 to better capture the frequency components from the EEG. Also, since fine convolution is not necessary to extract the frequency components, the stride size is set to Fs/2, which is larger than that of the conv1 layer of CNN1.

The filter size and stride size of the following convolutional layer conv1D are fixed. Although not particularly limited, for example, for CNN1, the settings can be 128 filters for three layers, (filter kernel size, stride size, activation function) = (8, 1, ReLU), and for CNN2, the settings can be 64 filters for three layers, (filter kernel size, stride size, activation function) = (6, 1, ReLU).

By using multiple convolutional layers with small filter sizes instead of a single convolutional layer with a large filter, it is possible to reduce the number of parameters and computational costs while still achieving the same level of model expressiveness.

In the sequence network, the parameters of the bidirectional LSTM layer and the fc layer (fully connected layer) were set to 1024, which is smaller than the output of the feature network. This is to restrict the model to select and combine only important features to prevent overfitting.

(Experimental example of learning of sleep stage prediction unit 3110)
(Dataset preparation)

To evaluate the performance of the sleep stage prediction unit 3110, a dataset was created according to the following conditions.

Here, the performance will be compared against a standard known as polysomnography (PSG) testing, which has traditionally been used to evaluate sleep states.

The basic EEG electrode positions required for PSG are the frontal (F3, F4), central (C3, C4) and occipital (O1, O2) regions, and their placement follows the International Electroencephalography Society standard method (ten-twenty electrode system; 10/20 method). For the electrooculogram (EOG), two leads (unipolar leads) are used, connecting the electrodes attached 1 cm below the left lateral canthus (E1) and 1 cm above the right lateral canthus (E2) to the right mastoid process (M2).

The learning dataset was created using all possible channel patterns, from 1 to 3 channels. Because the number and position of electrodes differ between the electrode sheet-type EEG 10' and the PSG, a direct comparison under the same conditions cannot be made.

Therefore, we used the closest and same number of channels, namely the F3-M2 and F4-M1 channels, as the comparison subjects. Also, to investigate the effect of the number of channels on the performance of the device, we used the four-channel models F3-M2, F4-M1, C3-M2, and C4-M1, as well as the six-channel models E1-M2 and E2-M1 in addition to the four channels mentioned above.

(DSN training method)

The network was trained for a total of 250 epochs per cross-validation using the ADAM optimizer with a learning rate of 0.0001, with 150 epochs for pre-training and 100 epochs for fine-tuning.

Gradient clipping was used in pre-training, and gradients were clipped to a maximum norm of 10.0. The training dataset was divided into training and validation datasets using cross-validation. Generally, cross-validation is performed in two approaches, within-patient and between-patient, based on the characteristics of the datasets. In within-subject cross-validation, both the training and validation data include data from the same subject, while in between-subject cross-validation, data is used for each subject.

Sleep recordings are highly patient-dependent, i.e., different patients have similar EEG patterns. Therefore, we adopted between-patient cross-validation to validate the model, using the leave-one-out cross-validation method. In leave-one-out cross-validation, the training dataset includes all subjects except one subject for validation.

The sleep stages are divided into the following stages:
・Awakening (Stage W)
・Non-REM sleep Sleep stage N1 (Stage 1)
Sleep stage N2 (Stage 2)
Sleep stage N3
Sleep Stage 3
Sleep Stage 4
・REM sleep (Stage R)

In general, the distribution of sleep stages is not uniform, with N2 stage accounting for up to 50% of the total, while N1 stage accounts for only 7%. Machine learning approaches essentially optimize data distribution, so the imbalance between classes significantly affects prediction accuracy. This problem is often called the class imbalance problem, and various studies have developed methods to address this issue. Here, we use a class weight method to solve the class imbalance problem. This method adjusts the loss calculation according to the class label frequency to equalize the contribution of each class, from those with a small number of samples to those with a large number of samples.

All models were evaluated using precision, recall, F1 score, etc.

(Evaluation of artificial intelligence models)

Accuracy ignores the class distribution in the dataset, so in the case of an unbalanced dataset, the evaluation of accuracy is not sufficient. Therefore, in order to accurately evaluate the performance of the proposed system, we used three metrics: recall, F1 score, and accuracy.

The overall accuracy of the electrode sheet-type electroencephalograph 10' in this embodiment is 75% to 78%, and the overall accuracy of the PSG is 78% to 79%, so it can be said that the accuracy of discrimination by the electrode sheet-type electroencephalograph 10' is sufficiently high.

The highest score for the electrode sheet-type electroencephalograph 10' was 78.6% under three-channel conditions using the F4-M1, while the highest score for the PSG device was 79.6% under one-channel conditions. These results show that this system achieved performance equivalent to that of a PSG device.

(Construction of depression symptom prediction model and menopausal symptom prediction model)
FIG. 6 is a diagram showing the results of sleep stage discrimination for each epoch for a certain subject using the electrode sheet-type electroencephalograph 10'.

The results show that stages of sleep transition between non-REM and REM over time.

Figure 7 shows the results of smoothing the changes in sleep stages over time determined as shown in Figure 6 using a smoothing process (triangular moving average process).

As explained in Figure 4, the time-dependent change curve with the sleep stage changes smoothed in this way is then frequency converted to extract features as frequency components.

In this way, the input to the depression symptom prediction unit 3240 and the menopausal symptom prediction unit 3250 is converted into features in frequency space, rather than into features of the change over time itself. This is equivalent to compressing the amount of data as preprocessing for the input data for "sleep assessment," which usually has a measurement time of about 8 hours.

Figure 8 is a diagram to explain the configuration and learning process of the learning model in the depressive symptom prediction unit 3240 and the menopausal symptom prediction unit 3250.

Although not limited to this, the depressive symptom prediction unit 3240 and the menopausal symptom prediction unit 3250 will be described as using a random forest model, which is a type of ensemble learning, as a model for supervised learning in which correct answer labels exist.

Referring to FIG. 8, first, the computing device generates N sets of data sets "Data 1"..."Data N" from the original training data by random sampling with bootstrap and re-extraction, and stores them in the storage device (STEP 1).

Next, for each dataset, a decision tree is trained using supervised learning with the Random Forest method, and N decision tree models are created (STEP 2).

Prediction processing is performed for each decision tree (STEP 3), and the final prediction result is generated by majority voting of the best results of the N decision tree models (STEP 4).

However, the configuration of the learning model for class classification is not limited to this configuration, and it is also possible to use bagging, boosting, or other methods as ensemble learning methods. Also, other configurations can be adopted as long as the learning model can be used as a classifier.
The depressive symptom prediction unit 3240 and the menopausal symptom prediction unit 3250 output symptom indicators according to the input subject data using a trained model generated by such a learning process.

(Hardware configuration of diagnostic support information provision system 1000)

Figure 9 is a block diagram illustrating the hardware configuration of the measurement terminal 2000.

Although not particularly limited, the measurement terminal 2000 can be configured to use a tablet terminal or a smartphone terminal.

Referring to FIG. 9, it corresponds to the EEG data recording and transmitting unit 2002 described in FIG. 4 and includes a computing device 2003 such as an MPU (Micro Processing Unit) or a CPU (Central Processing Unit), a storage device consisting of a RAM 2004, a ROM 2005, etc., and controls each part and creates a native platform environment and application execution environment in the software configuration by executing a predetermined basic OS, middleware, etc.

A liquid crystal panel or an organic EL panel is used as the display device 2008, and the operation device 2007 may be a touch panel integrated with the display panel in addition to key input, or may be a voice recognition device.

In the brainwave data recording and transmission unit 2002, the storage device includes, for example, semiconductor memory such as RAM 2004 as a temporary storage device and flash memory 2005 as a non-volatile storage device. This non-volatile storage device stores driver programs, operating system programs, application programs, data, etc. used for processing in each unit. Furthermore, in the brainwave data recording and transmission unit 2002, the brainwave data is stored in this non-volatile storage device.

For example, the non-volatile storage device stores driver programs such as a communication driver program that executes a wireless communication method conforming to the IEEE 802.11 standard, a short-range communication method conforming to the Bluetooth (registered trademark) standard, and a wireless communication method for mobile communication (cellular communication), an input device driver program that controls the operation device 2007, and an output device driver program that controls the display device 2008.

The non-volatile storage device also stores operating system programs, such as basic OSs such as Android (registered trademark) OS and iOS (registered trademark), and connection control programs that perform authentication in wireless communication methods such as the IEEE 802.11 standard, the Bluetooth (registered trademark) standard short-range communication method, and mobile communication (cellular communication) wireless communication method.

The communication interface 2006 has the functionality to perform wireless LAN communication, short-range communication based on the Bluetooth (registered trademark) standard, and, if necessary, mobile communication via a base station (not shown) of a cellular mobile communication network.

Figure 10 is a block diagram illustrating the hardware configuration of the information providing server 3000 shown in Figure 4.

As described above, the information providing server 3000 may be configured such that a computing device (CPU: Central Processing Unit) within its own housing executes the arithmetic processing, or a part of the program processing may be executed on another server. In the following description, it is assumed that a computing device within its own housing executes the arithmetic processing.

Referring to FIG. 10, the information providing server 3000 includes a computer device 3010, a network communication unit 3300 for communicating with a network, and a recording medium (e.g., a memory card) 3210 for recording data from the outside and providing the data to the computer device 3010.

For example, the recording medium 3210 may be a USB memory, a memory card, or an external storage device. The network communication unit 3300 may be a wired LAN router or a wireless LAN access point.

As shown in FIG. 10, the computer main body constituting this computer device 3010 includes, in addition to the disk drive 3030 and memory drive 3020, a CPU (Central Processing Unit) 3040 connected to a bus 3050, memory including a ROM (Read Only Memory) 3030 and a RAM (Random Access Memory) 3070, a non-volatile rewritable storage device such as an SSD (Solid State Drive) 3080, and an input/output interface 3090 for communicating via a network and sending and receiving data to and from the outside. An optical disk can be attached to the disk 3030. A memory card 3210 can be attached to the memory drive 3020.

As will be explained later, when the programs of the computer device 3010 are operating, the data and programs that store the information that is the basis for the operation of the computer are stored in the SSD 3080.

In FIG. 10, the medium capable of recording information such as a program to be installed in the computer main body may be, for example, a DVD-ROM (Digital Versatile Disc), a memory card, or a USB memory. In such cases, the computer main body is provided with a drive device (memory drive 3020, disk drive 3030) capable of reading these media.

The main components of the computer device 3010 are computer hardware and software executed by the CPU 3040. Generally, such software is stored in a storage medium and distributed or distributed via a network, and is acquired via the disk drive 3030 or the network communication unit 3300 and temporarily stored in the SSD 3080. It is then read from the SSD 3080 to the RAM 3070 in the memory and executed by the CPU 3040. Note that, when connected to a network, the software may be directly loaded into the RAM and executed without being stored in the SSD 3080.

When distributed, a program for functioning as computer device 3010 does not necessarily need to include an operating system (OS) that causes computer main body 3010 to execute functions such as an information processing device. The program only needs to include instructions that call appropriate functions (modules) in a controlled manner to obtain the desired results. How computer system 3010 operates is well known, and a detailed explanation will be omitted.

Furthermore, the CPU 3040 may also be a one-core processor or a multiple-core processor. That is, it may be a single-core processor or a multi-core processor.

In the above explanation, the measurement terminal 2000 and the information providing server 3000 are described as separate hardware devices, but for example, the measurement terminal 2000 and the information providing server 3000 may be realized by a single computer device.

(Simplified Menopausal Index (SMI))

Figure 11 shows a questionnaire for self-assessment to calculate the Simplified Menopausal Index (SMI).

Menopause brings about a multitude of changes to both the mind and body in a short period of time, with symptoms ranging in severity from barely noticeable to those that interfere with daily life. The Simplified Menopausal Index (SMI) is a simple questionnaire used to determine the severity of symptoms, and is tailored to suit the symptoms of Japanese people.

Since around 1990, when Dr. Koyama Takao and his colleagues were practicing at Tokyo Medical and Dental University, they had accumulated data on several hundred cases, and in 1992 developed the Simplified Menopause Index (SMI) with the aim of easily understanding vague symptoms using a common scale and linking it to diagnosis and treatment.

The total score is evaluated on a 5-point scale, providing a guide to whether or not a woman is suffering from menopausal symptoms and whether or not she should undergo treatment.

It is a questionnaire in which the patient marks the symptoms according to their severity and then enters a score. If any one symptom is severe, the patient is supposed to mark it as strong.

A check will be done based on the total score of the questionnaire, and the self-assessment method of the menopausal index will be as follows:

0-25 points: no abnormalities.

26-50 points: Pay attention to diet, exercise, etc.

51-65 points: Visit a menopause/menopause clinic. See a doctor, receive lifestyle advice, counseling, and drug therapy.

66-80 points: Planned treatment over a long period (six months or more) is required.

81-100 points: After thorough medical examinations in each department, if the only symptom is menopausal symptoms, long-term treatment by a specialist is required.

In the following explanation, as "correct answer data" for supervised learning, an SMI score of 51 or higher is defined as the presence of menopausal symptoms.

(Self-Rating Depression Scale)

Figure 12 shows the self-assessment depression scale checklist.

Zung's (1965) Self-Rating Depression Scale (SDS) is a questionnaire-type depression scale developed to easily measure the degree of depression.

The SDS consists of 20 questions about depressive symptoms. Based on their content, the questions are classified into main depressive feelings (2 items), physical symptoms (8 items), and mental symptoms (10 items). Based on the wording of the questions, they are further classified into 10 items written in negative terms, as when patients complain about their symptoms, and 10 items written in positive terms (reversed items). Respondents are asked to select the frequency of symptoms from four options, with each option assigned a score from 1 to 4 points. The correspondence between options and scores is reversed for negatively worded items and reversed items. The sum of the 20 points obtained by the respondent is called the SDS score, and is used as an index of the severity of the depressive state.

The cut-off point for depressive symptoms is 40 points, and they are classified as follows:
40-47 points: mild 48-55 points: moderate 56 points and above: severe

Treatment is required for moderate to severe cases.

In the following explanation, as "correct answer data" for supervised learning, an SDS score of 48 or higher is defined as the presence of depressive symptoms.

[Experimental result]

Below, we explain the results of a clinical study conducted on female patients in their 30s to 50s who were admitted to the gynecology ward at Osaka University Hospital.

As for the experimental subjects, taking into consideration the care they would need to provide, it was thought that it would be best to begin measurements on hospitalized patients, and in order to examine the possibility of differentiating symptoms, it would be best to begin measurements on bilateral oophorectomy patients with severe menopausal symptoms. Measurements began in June 2019 after approval from the ethical committee (a total of 153 cases were measured).

Including these patients and control cases, a symptom index was obtained as follows.
i) Measurement of brain waves during sleep using a new patch-type measurement sheet ii) Kupperman Menopausal Index (KSSI), Simplified Menopausal Index (SMI),
Zung Self-Rating Depression Scale (SDS) and Quick Inventory of Depressive Symptoms (QIDS-J) questionnaire

Figure 13 shows a flowchart showing the case selection process and the background of the cases.

Figure 13 (a) explains the process for selecting study subjects. 174 consenting cases (14 healthy subjects) were included in the study, and 233 EEG measurements (35 healthy subjects) were performed.

Of these, 133 cases and 174 measurements were ultimately included in the study as a result of exclusion due to factors such as being within the learning curve, meeting the exclusion criteria, measurement failure, and withdrawal of consent.

Figure 13(b) also shows a summary of the subjects' ages, BMI values, ovarian function status, and symptom indicators.

Figure 14 shows the assessment of sleep quality using the SMI index.

The following sleep variables are obtained to evaluate sleep quality:
- "Total sleep time" is the time excluding the time spent awake during sleep
- "Sleep time" is the time from falling asleep to the final awakening
SE: Sleep efficiency (total sleep time/total time in bed)
SOL: Sleep latency (time taken from lights out to first sleep onset)
Deep sleep latency (time from falling asleep to the first deep sleep (stage 3))
・Number of awakenings during sleep and duration ・WASO/SPT: Occurrence rate of awakenings during sleep ・SOREMP: REM sleep latency (time from falling asleep to the first REM sleep)
・REM/SPT: REM sleep occurrence rate during sleep time

Figure 14 shows that N3 (deep sleep) tends to decrease in patients with moderate to severe SMI.

Figure 15 shows the evaluation of sleep quality using the SDS index.

The sleep variables used to evaluate sleep quality are the same as those in Figure 14.

Figure 15 shows that, within the limits of human perception, no sleep patterns characteristic of moderate to severe SDS could be detected.

Therefore, we investigated the detection of sleep EEG patterns characteristic of cases with depressive tendencies and vasomotor symptoms by measuring EEG during sleep using the configuration of the diagnostic support information providing system 1000 described above.

In other words, by using artificial intelligence technology, we attempted to automatically classify patients who exhibit menopausal symptoms into depression type and hot flash type.

Figure 16 shows the final diagnostic accuracy of electroencephalogram measurements of menopausal women using an electrode sheet-type electroencephalograph 10' and its artificial intelligence analysis.

As an aid in diagnosing menopausal symptoms, both the sensitivity and negative predictive value are over 90%, indicating that it may be an effective screening test for menopausal women.

If we consider menopausal women aged 45-55 and pre-menopausal women aged 35-44, at least half of them are complaining of some vague complaints (menopausal symptoms). Although they have symptoms, it seems that there are a great many women who endure without visiting a medical institution because the hurdle of visiting a medical institution is high.
Among them, a considerable number of women suffer from these symptoms and are forced to quit their jobs. For such women, the symptom indicator providing device of this embodiment provides an opportunity to easily diagnose at home whether the symptoms are related to menopausal disorder caused by estrogen deficiency or mental symptoms caused by other reasons. This can increase women's health literacy regarding menopause and reduce the number of women who have not been able to take appropriate measures such as treatment.

In the above description, the present invention has been described as being capable of providing information to assist in the diagnosis of menopausal disorders and providing health information regarding symptoms of menopausal disorders. However, the present invention is not necessarily limited to such cases, and can also be applied to other diseases and health conditions when the state of sleep is related to the diagnosis of diseases or health conditions.
The embodiments disclosed herein are merely examples of configurations for specifically implementing the present invention, and do not limit the technical scope of the present invention. The technical scope of the present invention is indicated by the claims, not by the description of the embodiments, and is intended to include modifications within the literal scope of the claims and within the scope of equivalent meanings.

2 Subject, 4 Network, 10 Biosignal Measuring Device, 10' Electrode Sheet Type Electroencephalograph, 1000 Diagnostic Support Information Providing System, 2000 Measurement Terminal, 2002 Electroencephalogram Data Recording and Transmission Unit, 2008 Display Device, 3000 Information Providing Server, 3100 Sleep Stage Estimation Unit, 3110 Sleep Stage Prediction Unit, 3120 Sleep Stage Recording Unit, 3130 Logic Unit, 3140 Sleep Variable Recording Unit, 3200 Smoothing Processing Unit, 3210 Continuous Waveform Acquisition and Recording Unit, 3220 Frequency Conversion Unit, 3230 Feature Extraction Unit, 3240 Depression Symptom Prediction Unit, 3250 Menopausal Symptom Prediction Unit 3250.

Claims (11)

A diagnostic auxiliary information providing system for providing diagnostic auxiliary information for menopausal symptoms of a subject, comprising:
an electroencephalogram sensor for acquiring an electroencephalogram signal from a subject;
a storage means for storing time-course electroencephalogram data regarding the subject's sleep period based on the electroencephalogram signal from the electroencephalogram sensor;
a sleep stage determination means for determining, over time, at a predetermined time interval, which of a plurality of sleep stages the sleep state of the subject corresponds to, using the time-varying electroencephalogram data as an input;
a smoothing processing means for performing a time-smoothing process in response to the sleep stage determination results over time;
a frequency conversion means for receiving an output of the smoothing processing means and converting it into a frequency component;
a symptom prediction means for outputting auxiliary diagnostic information for the subject based on the electroencephalogram signal from the subject,
The symptom prediction means of this diagnostic auxiliary information providing system includes a first trained model that is generated in advance by machine learning using frequency components of electroencephalogram data measured on a plurality of training subjects and corresponding clinical indicators of menopausal symptoms as training data, and that outputs the diagnostic auxiliary information using frequency components converted by the frequency conversion means for the electroencephalogram data from the subjects as input. The diagnostic support information providing system according to claim 1, wherein the smoothing process is a moving triangle average process. The diagnostic auxiliary information providing system according to claim 1 or 2, wherein the symptom prediction means includes a second trained model that is generated in advance by machine learning using frequency components of electroencephalogram data measured for a plurality of training subjects and corresponding clinical indicators of depressive symptoms as training data, and that outputs the diagnostic auxiliary information using frequency components converted by the frequency conversion means for the electroencephalogram data from the subjects as input. The sleep stage determination means includes:
a first convolutional neural network and a second convolutional neural network, each having a different filter size, for receiving as an input the electroencephalogram data of a plurality of channels from electrodes arranged at different positions on the subject's head in the electroencephalogram sensor and extracting features from the electroencephalogram data;
a bidirectional recurrent neural network that receives outputs of the first and second convolutional neural networks as inputs and outputs the sleep stage. A diagnostic support device for providing diagnostic support information for menopausal symptoms of a subject based on time-dependent electroencephalogram data on the sleep period of the subject acquired by an electroencephalogram sensor, comprising:
a sleep stage determination means for receiving the electroencephalogram data acquired by the electroencephalogram sensor, and determining, at a predetermined time interval, which of a plurality of sleep stages the sleep state of the subject corresponds to, using the electroencephalogram data over time as an input;
a smoothing processing means for performing a time-smoothing process in response to the sleep stage determination results over time;
a frequency conversion means for receiving an output of the smoothing processing means and converting it into a frequency component;
a symptom prediction means for returning diagnostic auxiliary information to the subject based on the electroencephalogram signal from the subject,
The symptom prediction means is a diagnostic support device that includes a first trained model that is generated in advance by machine learning using frequency components of electroencephalogram data measured on a plurality of training subjects and corresponding clinical indicators of menopausal symptoms as training data, and that outputs the diagnostic support information using frequency components converted by the frequency conversion means for the electroencephalogram data from the subjects as input. The diagnostic support device according to claim 5, wherein the smoothing process is a moving triangle average process. The diagnostic support device according to claim 5 or 6, wherein the symptom prediction means includes a second trained model that is generated in advance by machine learning using frequency components of electroencephalogram data measured for a plurality of training subjects and corresponding clinical indicators of depressive symptoms as training data, and that outputs the diagnostic support information using frequency components converted by the frequency conversion means for the electroencephalogram data from the subjects as input. 1. A symptom index providing device for providing a symptom index for menopausal symptoms of a subject based on time-varying electroencephalogram data on a sleep period of the subject acquired by an electroencephalogram sensor, comprising:
a sleep stage determination means for receiving the electroencephalogram data acquired by the electroencephalogram sensor, and determining, at a predetermined time interval, which of a plurality of sleep stages the sleep state of the subject corresponds to, using the electroencephalogram data over time as an input;
a smoothing processing means for performing a time-smoothing process in response to the sleep stage determination results over time;
a frequency conversion means for receiving an output of the smoothing processing means and converting it into a frequency component;
a symptom prediction means for outputting a symptom index for the subject based on an electroencephalogram signal from the subject,
The symptom prediction means includes a first trained model that is generated in advance by machine learning using frequency components of electroencephalogram data measured on a plurality of training subjects and corresponding clinical indicators of menopausal symptoms as training data, and that outputs the symptom indicator using frequency components converted by the frequency conversion means for the electroencephalogram data from the subjects as input. The symptom index providing device according to claim 5, wherein the smoothing process is a moving triangle average process. The symptom index providing device according to claim 8 or 9, wherein the symptom prediction means includes a second trained model that is generated in advance by machine learning using frequency components of electroencephalogram data measured for a plurality of training subjects and corresponding clinical indices of depressive symptoms as training data, and that outputs the symptom index using frequency components converted by the frequency conversion means for the electroencephalogram data from the subjects as input. A symptom index providing program for causing a computer to function as the sleep stage determination means, the smoothing processing means, the frequency conversion means, and the symptom prediction means described in claim 8.

Images