KR20190119198A - Method for monitoring cardiac impulse of fetus using artificial intelligence - Google Patents

Abstract

According to an embodiment of the present invention, a method for monitoring the cardiac impulse of a fetus using artificial intelligence can precisely interpret a vibration pattern using an algorithm (hereinafter, artificial intelligence) using big data and artificial intelligence technology. The artificial intelligence can include an algorithm capable of integrating a placental ultrasound image and fetal pathological image data and interpreting fetal heartbeat monitor patterns that have not been read until now.

Description

METHOOD FOR MONITORING CARDIAC IMPULSE OF FETUS USING ARTIFICIAL INTELLIGENCE}

The present invention is a technique for monitoring the heart rate of the fetus in a non-invasive method using artificial intelligence.

Conventional hospitals use non-invasive methods to detect fetal status, ultrasound and electronic fetal heartbeat monitoring. Ultrasound may only confirm the condition of the fetus at the moment the doctor contacts the ultrasound device with a part of the mother's body corresponding to the location of the fetus. The electronic fetal heartbeat monitoring test can confirm the condition of the fetus by analyzing the monitoring result paper outputting the vibration graph as the mother wears the test device. On the other hand, since the state of the fetus in the analgement process of the mother just before childbirth may cause problems for the mother or fetus, it is required to continuously monitor the state of the fetus during the labor process. However, maternal analgesia lasts for several hours. Ultrasonic examinations have practical limitations for doctors staying with their mothers for hours, and there are limitations to their use. Since the paper is printed indefinitely in proportion to the labor time, the amount of the resulting paper is so large that there is a limit to analyzing it efficiently.

The problem to be solved by the present invention is to propose a fetal heartbeat monitoring technology that can solve the limitations of the prior art as described above.

The problem to be solved by the present invention is to propose an artificial intelligence fetal monitoring system for precisely monitoring high-risk pregnant fetuses to overcome the lack of delivery infrastructure and obstetric specialists, reduce neonatal damage, and safely manage high-risk pregnant women.

However, the object of the present invention is not limited to the above-mentioned object, and although not mentioned, it may include an object that can be clearly understood by those skilled in the art from the following description. have.

According to an embodiment of the present invention, the fetal heartbeat may be monitored using artificial intelligence, and the monitoring result may be output as an image appearing in the form of a block.

According to an embodiment of the present invention, the fetal heartbeat test can be performed accurately and efficiently because artificial intelligence can be used to more accurately monitor fetal heartbeats, and the results can be expressed in the form of blocks to generate a single image. Will have the effect.

1 is a view for explaining the prior art of the present invention.
2 to 14 are views for explaining in more detail the method according to an embodiment of the present invention.

Advantages and features of the present invention and methods for achieving them will be apparent with reference to the embodiments described below in detail with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below, but can be implemented in various different forms, and only the embodiments make the disclosure of the present invention complete, and the general knowledge in the art to which the present invention belongs. It is provided to fully inform the person having the scope of the invention, which is defined only by the scope of the claims.

In describing the embodiments of the present invention, if it is determined that a detailed description of a known function or configuration may unnecessarily obscure the gist of the present invention, the detailed description thereof will be omitted. In addition, terms to be described below are terms defined in consideration of functions in the embodiments of the present invention, which may vary according to intentions or customs of users and operators. Therefore, the definition should be made based on the contents throughout the specification.

In recent years, births of multiple fetuses, premature babies, or underweight babies have increased, and the age of first marriage has also increased. The age at which women married first increased to 29.8 years in 2014, 30.0 years in 2015, and 30.1 years in 2016. The average maternal age of mothers increased to 32.0 years in 2014, 32.2 years in 2015, and 32.4 years in 2016. Here, multiple fetuses may refer to several fetuses born and raised simultaneously in one womb. As a result, high-risk pregnant women are increasing, and the mortality rate of high-risk pregnant women is increasing. On the other hand, there are very few obstetricians who can monitor fetuses of high-risk mothers, and there are many medical accidents in obstetrics and gynecologists.

In the management of high-risk mothers, it is important to examine the condition of the fetus. Fetal endoscopy, computed tomography (CT), and magnetic resonance imaging (MRI) may be used to examine the condition of the fetus. Since the fetus is in a kind of water bag called amniotic membrane, it is very dangerous to monitor the fetus by invasive methods such as penetrating the amniotic membrane and examining the placenta and other tissues during pregnancy. In the case of CT, since it may be exposed to harmful components such as radiation, there is a problem that it is not good for the mother or the fetus. MRI requires the mother to lie still for the test, which may cause pressure on the inferior vena cava or uterus, which is not good for the mother or the fetus.

The fetal brain can control the fetal heart rate through the sympathetic and parasympathetic nerves. When the fetus is hypoxic, the balance of the sympathetic-parasympathetic nerves may be broken, resulting in different patterns of fetal heartbeat changes. According to one embodiment of the present invention, in order to observe such fetal heart rate changes, a monitor for checking the fetal heart rate and a pressure transducer for monitoring the maternal uterine muscle contraction may be attached to the abdomen of the mother to record a signal. .

According to an embodiment of the present invention, the fetal heartbeat monitoring method may precisely interpret the vibration pattern using an algorithm (hereinafter, artificial intelligence) using big data and artificial intelligence technology. The artificial intelligence may include an algorithm capable of integrating placental ultrasound images and fetal pathological image data to interpret fetal heartbeat monitor patterns that have not been read until now.

Artificial intelligence can automatically determine the data. Artificial intelligence can learn preprocessed data and readout information and use it to predict fetal heart rate patterns. Artificial intelligence can include one-dimensional ResNet. Artificial intelligence may include various algorithms. For example, artificial intelligence may include a neural network algorithm.

The present invention can add fetal heartbeat monitoring readings by adding placental ultrasound and placental pathology information. Monitoring readings can be performed based on pre-processing algorithms for heart rate data interpretation. The present invention can automatically quantify fetal heartbeat image data. Placental ultrasound images and placental pathology can be combined with fetal heartbeat reading algorithms to compensate for data loss on fetal heartbeats.

In analyzing images of placental pathology, the placenta can be examined under a microscope to determine the presence, severity, and cause of fetal disease using morphology and immunohistochemical staining techniques. However, because the placenta can be obtained after the birth of a pregnant woman, it is difficult to use for the diagnosis of the fetus during pregnancy. Even if some placental tissues are removed and subjected to biopsy, they cannot be used because some placental tissues cannot represent the entire placenta.

The placenta's echotexture is heterogeneous, making it difficult to discern with the human eye, and yet, when the image of placenta ultrasonography is still present, the actual pathology of the placenta is Very little research is available to analyze how it looks. This is because there is no understanding of placental pathology of obstetrics and gynecologists who perform placental ultrasound, and the placental pathologists do not understand ultrasound.

Using a heart beat device, some information may be lost when obtaining a graph for the heart rate. That is, the result may be derived without displaying a part of the graph. In this case, the present invention may infer a part of the graph that is not displayed by using artificial intelligence, placental pathology information, placental ultrasound information, and the like. As inferred, portions which are not partially displayed may be displayed. According to the present invention, even when the fetal heartbeat data is not measured, the unmeasured portion may be estimated using the learned data based on the pathopathological information and the ultrasound ultrasound information.

The artificial intelligence, which is not sufficiently limited, may select a placental pathology image mapped thereto when the placenta region is separated and input from the ultrasound image. It is not necessary to input separately into the first area corresponding to the placenta, and in some cases, the entire ultrasound image may be input, and the machine learning apparatus may separate the area corresponding to the placenta by itself.

An ultrasound image of the fetus may also be considered by inputting a first region corresponding to the placenta and a second region corresponding to the fetus. Similarly, it is not necessary to input the first area and the second area, which are separated by the input of the machine learning device, and when the entire ultrasound image is input, the machine learning device separates and maps the areas corresponding to the placenta and the fetus. Pathological images can also be selected.

Artificial intelligence learns placental pathology images and placental ultrasound images. Analyze the correlation between placental ultrasound and placental pathological images, and learn to find corresponding images.

After the learning is completed, the artificial intelligence receives an image similar to the placental detachment ultrasound image and may find and analyze a pathological image mapped thereto. Image generation is performed based on the learned machine learning apparatus.

The first region is extracted from the received ultrasound image and input to the learned machine learning apparatus. The learned machine learning apparatus selects a first matching pathological image mapped to a first region. An image processing method for providing event information corresponding to the first matching pathological image is performed. The event information may be a disease classification code corresponding to a first matching pathology image or an appropriate birth time, but is not limited thereto. The event information may include medical related information corresponding to the first matching pathology image. Choose. An image processing method for providing event information corresponding to the first matching pathological image is performed. The event information may be a disease classification code corresponding to a first matching pathology image or an appropriate birth time, but is not limited thereto. The event information may include medical related information corresponding to the first matching pathology image. The predicted model is generated through machine learning method by post processing on the found features. At this time, find the best combination from the various features and verify the prediction model through 10-fold or Leave one out cross validation. In the case where the second region corresponding to the fetus is utilized in addition to the first region corresponding to the placenta in the ultrasound image, variables of the fetal part are combined in addition to the features obtained through the CNN method. More specifically, the variable is a clinical variable measured in the image information. For example, the variable may be the height of the fetus, the head circumference, the length of the nape, the presence or absence of a nasal bone, and the like. By combining the variables of the fetal part and analyzing the correlation, the accuracy of the prediction model can be improved. In addition to placental imaging, a variety of information is collected to utilize the fetal image. Basically, pregnant women data, biometric data, and fetal measurement data using ultrasound are collected. The pregnant woman data may be information about age, the start of the last menstruation, drug administration, past medical history, natural pregnancy, pre-pregnancy hormonal status, quad test, abnormal vision, headache, and the like. The biometric data may be seismic, fetal heart rate, uterine contraction monitoring, prenatal genetic test results, and the like. The fetal measurement information using the ultrasound may be a predicted weight, leg length, head circumference, abdominal circumference, double head bone diameter, and the like. An ultrasound image and a pathological image are correlated with the pregnant woman data, biometric data, and fetal measurement data using ultrasound, and an algorithm is generated by applying a high risk pregnancy. In addition, categorized by disease occurring in the fetus is performed.

In the correlation analysis between ultrasound image and placental pathology image, we need to find the best parameters from the extracted features to make an accurate prediction model using cross-validation method. For example, the parameter may be the number of hidden layer layers, but the parameter is not limited thereto and may be a parameter that can be reflected in the prediction model.

When the ultrasound image of the placenta is input to the input of the learned machine learning apparatus, the primary output may be "presence of pregnancy addiction", and in some cases, placenta detachment and fetal infection may also be applied. Since the data are generated regularly at several weeks intervals, which is the advantage of prenatal ultrasound data, it is possible to use the Recurrent Neural Network (RNN) method to suggest childbirth recommendations.

The medical history listening information may be information about a pregnancy method, a past medical history, parity, abortion, drug use, abdominal pain, and vaginal bleeding. The biometric information may be information on blood pressure, weight, height, proteinuria, nutrition evaluation, etc. The choking ultrasound information may be information on the presence or absence of a pregnancy bag, the number of fetuses, fetal length, fetal heartbeat, and yolk evaluation. All of the above information is synthesized to determine whether there are any abnormal symptoms, and accordingly, a high risk pregnancy algorithm or a low risk pregnancy algorithm is applied.

High risk and low risk Cases of chorionic deformity, chorionic hemorrhage, acute inflammatory infections, chronic inflammation, maternal rejection of fetal fetuses, and rare diseases should be determined. Accordingly, the secondary evaluation is performed considering the benefits and risks of drug treatment, genetic testing, and absolute stability of immunosuppressants, anticoagulants, and anti-hyperlipidemic drugs.

In accordance with the above benefits and risks, a combination of the most risk to benefit benefits is delivered to the physician. Doctors can use the information to make decisions about whether to perform treatment or follow-up. Since the data are generated regularly at several weeks intervals, which is the advantage of prenatal ultrasound data, it is possible to use the Recurrent Neural Network (RNN) method to suggest childbirth recommendations.

The medical history listening information may be information about a pregnancy method, a past medical history, parity, abortion, drug use, abdominal pain, and vaginal bleeding. The biometric information may be information on blood pressure, weight, height, proteinuria, nutrition evaluation, etc. The choking ultrasound information may be information on the presence or absence of a pregnancy bag, the number of fetuses, fetal length, fetal heartbeat, and yolk evaluation. All of the above information is synthesized to determine whether there are any abnormal symptoms, and accordingly, a high risk pregnancy algorithm or a low risk pregnancy algorithm is applied.

We look at each case more specifically. In the case of gestational addiction, the possibility of development is scored and the severity of gestational addiction is predicted. If the pregnancy is continued until each week, the risk of death of the pregnant woman is calculated and the risk of death of the fetus is likewise calculated. It is recommended to start the week of gestation and to monitor the fetus continuously. Finally, the optimal time of delivery is suggested.

According to another embodiment similarly applicable to the case of gestational diabetes.

Scoring the risk of developing gestational diabetes and quantifying the risk of malformations associated with gestational diabetes. Calculation of the risk of death of the fetus at the end of each gestation until week is suggested and the insulin dose at continuous blood glucose input. Finally, it is recommended how many weeks of gestation you need to monitor your fetus and suggest an optimal delivery time.

In the case of intrauterine infection prediction, it is possible to predict a specific infection in which the shape of the placenta changes characteristically. Examples include syphilis, cytomegalovirus, parvovirus. In this case, it is also recommended to calculate the risk of death of the fetus when pregnancy continues until the end of each week, and recommend that antibiotics continue when entering body temperature, blood CBC, and CRP. Likewise, it is recommended to finally monitor the number of gestational weeks from which the fetus needs continuous monitoring and suggest an optimal delivery time.

In these various cases the optimal timing of delivery is recommended and treatment or follow-up is performed accordingly. As the optimal delivery time arrives, ultrasound images of the placenta during pregnancy and microscopic images of the actual placenta obtained by delivery can be compared, and the deep learning algorithm can be fed back and modified.

Algorithms exist for emergency room visits, emergency situations during hospitalization, and shortly before delivery. The algorithms applied to machine learning devices can provide obstetricians with intelligent assistance at the pathologic level.

In the above algorithms for various cases, non stress test (NST) and Tocomonitoring are also interpreted as deep learning and included in risk assessment. The NST refers to a test that looks at the relationship between fetal movements and a heartbeat in the absence of stimulation, and the Toco monitoring refers to a test to see a relationship between fetal heartbeats associated with uterine contractions.

According to another exemplary embodiment, the image processing method may determine a recommendation date of twins. In twins with growth differences, the small fetus must be delivered quickly, and the large fetus is born prematurely because of the small fetus. However, in the case of a single chorional twin, if the pregnancy continues for a large fetus, the large fetus that survives the death of the small fetus is very likely to suffer severe brain damage. Therefore, it is necessary to maintain pregnancy as long as the small fetus does not die. The placenta ultrasound image may be utilized in determining a recommendation date for the delivery of the twins. Scoring the maturity of the placenta allows the machine learning device to determine the optimal delivery time. In the above, the placenta image was exemplarily described for only a few diseases, but the present invention is not limited thereto. Application is possible.

Using the present invention, a graph representing heart rate can be represented as one image. An image may appear as a plurality of blocks representing a unit of time. The present invention can obtain a graph representing the heartbeat of the fetus and analyze the graph. For example, you can look at the shape of the graph and predict the level of risk. The danger level may be indicated in color in the image according to the invention. For example, each of the plurality of blocks included in the image may appear in a different color depending on the level of danger. Therefore, even if only one image is viewed once, it is possible to determine the overall risk of the fetus.

The plurality of blocks included in the image represent inspection results within a predetermined time unit. For example, one block may correspond to a heartbeat graph for about one minute. When the color of one block is red, it may mean that a risk is detected in the graph of 1 minute. Blocks can be arranged in a variety of ways. For example, they may be arranged in a chronological order from top to bottom, then from left to right. If the fetus is not at risk, certain colored blocks (eg red) may not be displayed in the image, and if the fetus is detected as dangerous, certain colored blocks may be displayed within the image.

The machine learning apparatus performs learning using a plurality of placental ultrasound images and a plurality of placental pathological images. After the machine learning apparatus performs both machine learning with the plurality of placental ultrasound images and placental pathology images, the machine learning apparatus may extract placental pathology images corresponding to the ultrasound images. For example, the machine learning apparatus selects a placental pathology image having the highest correlation with the placental ultrasonography image when a cyst occurs. Event information corresponding to the placental pathology image selected according to the placental ultrasonography image when the cyst occurs may be extracted. The event information may be information about a disease classification code of a disease that can be grasped in a placental pathology image or a period during which a corresponding safe pregnancy can be maintained and an appropriate time of birth.

The present invention improves the accuracy of fetal heartbeat diagnosis and develops a basic technology for establishing and reading a baseline database for diagnosing placental ultrasound with data for a time shorter than the existing fetal heartbeat monitoring time (20-40 minutes). Interpretation and placental ultrasound analyzes allow the fetus to be monitored precisely.

Since the present invention can be applied to all mothers, the market can be very wide when commercialized. That is, the fetal heartbeat monitoring analysis system according to the present invention can be spread worldwide.

The present invention is the accuracy of the obstetrician's abnormalities when the fetus heart rate monitoring in the delivery-vulnerable area without the obstetrician, no obstetrics and gynecology hospital during the night shift, or when the midwives, nurses have to interpret the fetal heart rate monitoring Can provide an interpretation. In addition, during fetal heartbeat monitoring in a delivery-vulnerable hospital, when the fetus is read as an emergency, data can be transmitted to a hospital capable of providing emergency delivery of a high-risk mother at a recent distance, thereby enabling an emergency delivery system.

The present invention can allow doctors at night time, in rural areas, or on island islands to keep pregnancy safe for women with complications. Doctors using the present invention can reduce fetal damage due to rapid first aid in situations where the mother or fetus is at risk during childbirth. It can also reduce anxiety about women's pregnancy and delivery.

According to one embodiment, a computer-implemented image processing method comprising: extracting a first area corresponding to the placenta from a received ultrasound image; Receiving a first matching pathology image from a machine learning engine by using the first region as an input; And providing at least one event information corresponding to the first matching pathological image.

In another exemplary embodiment, an image processing method including the disease information code mapped to the first matching pathological image is also disclosed.

In addition, the event information may be an image processing method that includes a calculated estimated estimated birth date mapped to the first matching pathological image.

According to another embodiment, the machine learning engine may be an image processing method including an engine trained using a plurality of placental ultrasound images and placental pathology images.

According to another embodiment, an image processing method further comprising a preprocessing step of removing noise of the received ultrasound image.

According to one side, a computer-implemented image processing method comprising: extracting a first area corresponding to the placenta from a received ultrasound image; Extracting a second region corresponding to the fetus from the received ultrasound image; Receiving a first matching pathological image from a machine learning engine using the first region and the second region as inputs; And providing at least one event information corresponding to the first matching pathological image.

The apparatus described above may be implemented as a hardware component, a software component, and / or a combination of hardware components and software components. For example, the devices and components described in the embodiments may be, for example, processors, controllers, arithmetic logic units (ALUs), digital signal processors, microcomputers, field programmable arrays (FPAs), It may be implemented using one or more general purpose or special purpose computers, such as a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to the execution of the software.

For convenience of explanation, one processing device may be described as being used, but one of ordinary skill in the art will appreciate that the processing device includes a plurality of processing elements and / or a plurality of types of processing elements. It can be seen that it may include. For example, the processing device may include a plurality of processors or one processor and one controller. In addition, other processing configurations are possible, such as parallel processors.

The software may include a computer program, code, instructions, or a combination of one or more of the above, and configure the processing device to operate as desired, or process it independently or collectively. You can command the device. Software and / or data may be any type of machine, component, physical device, virtual equipment, computer storage medium or device in order to be interpreted by or to provide instructions or data to the processing device. Or may be permanently or temporarily embodied in a signal wave to be transmitted. The software may be distributed over networked computer systems so that they may be stored or executed in a distributed manner. Software and data may be stored on one or more computer readable recording media.

The method according to the embodiment may be embodied in the form of program instructions that can be executed by various computer means and recorded in a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the media may be those specially designed and constructed for the purposes of the embodiments, or they may be of the kind well-known and available to those having skill in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tape, optical media such as CD-ROMs, DVDs, and magnetic disks, such as floppy disks. Magneto-optical media, and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine code generated by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the embodiments have been described with reference to the accompanying drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques may be performed in a different order than the described method, and / or components of the described systems, structures, devices, circuits, etc. may be combined or combined in a different manner than the described method, or other components. Or even if replaced or substituted by equivalents, an appropriate result can be achieved. Therefore, other implementations, other embodiments, and equivalents to the claims are within the scope of the claims that follow.

Claims (1)

Way.

Images