Apparatus for supporting diagnosis of intraventricular hemorrhage and early death within a week in very low birth weight infants based on deep learning, and method thereof
A machine learning-based diagnostic support device for IVH in extremely low birth weight infants uses time-specific prediction models to enhance early detection and treatment planning, reducing brain damage and mortality.
Patent Information
- Application Number
- KR1020240080192
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2024-06-20
- Publication Date
- 2026-11-06
- Estimated Expiration
- 2044-06-20
AI Technical Summary
Current methods lack effective machine learning-based diagnostic tools for predicting intraventricular hemorrhage (IVH) and early death in extremely low birth weight infants, particularly focusing on time-specific diagnostic points before, during, and after birth, which are crucial for optimizing treatment plans.
A machine learning-based integrated diagnostic support device that collects and preprocesses demographic, maternal, delivery, and vital sign data to train pre-birth, birth, and post-birth prediction models using LR, RF, and XGB algorithms, enabling time-specific predictions and optimized treatment plans.
Enables accurate and timely diagnosis of IVH and early death risk, reducing brain damage and mortality by allowing for optimized treatment strategies at different stages post-birth.
Smart Images

Figure 112024066724465-PAT00001_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a machine learning-based integrated diagnostic support device for intraventricular hemorrhage and early death within one week in extremely low birth weight infants, and a method thereof, which enables simpler and faster diagnosis of the likelihood of intraventricular hemorrhage and the likelihood of early death within one week in extremely low birth weight infants. Background Technology
[0002] According to the 2019 Korean Network of Very Low Birth Weight Infants, 41.5% of very-low-birth-weight infants (VLBWI) experienced intraventricular hemorrhage (IVH), of which 63.7% showed severe IVH of grade 3 or higher.
[0003] IVH is associated with a poor prognosis and is a factor that increases the risk of death by up to three times. IVH in premature infants often occurs when multiple factors, including changes in cerebral blood flow, autoregulatory abnormalities, and fragility of the perivesticular budding matrix, coincide or overlap. This condition is associated with prenatal and postnatal clinical scenarios.
[0004] In other words, IVH in extremely low birth weight infants occurs most frequently immediately after birth, and treatment options capable of significantly improving the prognosis of already occurring hemorrhage remain limited. Therefore, it is important to properly assess and manage the infant's clinical condition before and after birth, and it is particularly important to focus on the prevention of IVH.
[0005] Meanwhile, the advancement of big data analysis has led to the use of artificial intelligence (AI), such as machine learning or deep learning, in various fields.
[0006] AI has been widely used in the medical field to develop predictive models for IVH based on imaging technologies such as brain ultrasound or brain magnetic resonance imaging, but most of these studies have been conducted on adults.
[0007] For example, in a clinical study on IVH based on initial data from the Korean Neonatal Network (KNN), a national-level cohort, traditional statistical methods such as logistic regression have been used to identify risk factors.
[0008] However, although KNN data was used to analyze clinical factors of early infants using AI technologies such as machine learning and deep learning, there is a limitation in that no research has been conducted to date that explicitly focuses on factors related to IVH. Prior art literature
[0009] Korean Patent Publication No. 10-2023-0083447 (Date of Publication: Dec. 03, 2021) The problem to be solved
[0010] Accordingly, in order to solve the aforementioned problems, the present invention aims to provide a machine learning-based integrated diagnostic support device for intraventricular hemorrhage and early death within one week in extremely low birth weight infants, and a method thereof, which utilizes machine learning technology to pre-diagnose the likelihood of occurrence of intraventricular hemorrhage and early death in extremely low birth weight infants.
[0011] In addition, we aim to provide a machine learning-based integrated diagnostic support device and method for intraventricular hemorrhage and early death within one week in extremely low birth weight infants, which enables diagnostic operations to be subdivided into three time points: before birth, immediately after birth, and one week after birth, thereby allowing for the acquisition and provision of an optimized treatment plan at the time of diagnosis.
[0012] The objectives of the present invention are not limited to those mentioned above, and other unmentioned objectives will be clearly understood by those skilled in the art to which the present invention pertains from the description below. means of solving the problem
[0013] As a means to solve the above problem, according to one embodiment of the present invention, a machine learning-based intraventricular hemorrhage prediction method for an intraventricular hemorrhage prediction device for an intraventricular hemorrhage of an extremely low birth weight infant comprises: a data collection step of collecting and storing demographic information, maternal information, delivery information, newborn information, disease information, vital signs at birth, vital signs for one week after birth, and intraventricular hemorrhage diagnosis results from a previously registered medical database; a data preprocessing step of generating a first feature value including demographic information and maternal information, a second feature value including delivery information, newborn information, and vital signs at birth additionally included in the first feature value, and a third feature value including vital signs for one week after birth and disease information additionally included in the second feature value; and a learning step of training a pre-birth prediction model based on the first feature value and intraventricular hemorrhage diagnosis results, training a birth prediction model based on the second feature value and intraventricular hemorrhage diagnosis results, and training a post-birth prediction model based on the third feature value and intraventricular hemorrhage diagnosis results. The present invention provides a machine learning-based integrated diagnostic support method for the intraventricular hemorrhage status and the likelihood of early death within one week in extremely low birth weight infants, comprising a diagnostic step in which, once the diagnostic subject and the diagnostic time are determined, one predictive model is selected based on the diagnostic time, and the medical information of the diagnostic subject is analyzed through the selected predictive model to predict and output a diagnosis result of intraventricular hemorrhage.
[0014] The above demographic information includes at least one of fetal sex and maternal age, the above maternal information includes at least one of number of pregnancies, whether in vitro fertilization occurred, maternal diabetes, maternal hypertension, and clinical chorioamnionitis, the above delivery information includes at least one of duration of premature rupture of membranes and mode of delivery, the above neonatal information includes at least one of oxygen saturation, electrocardiogram and whether resuscitation was performed during delivery, gestational age, weight, 1-minute and 5-minute Apgar scores, pH, and base excess index, the above disease information includes at least one of pulmonary hemorrhage, respiratory distress syndrome, and the presence or absence of hypotension requiring drug treatment, and the above vital signs include at least one of oxygen saturation and electrocardiogram.
[0015] The medical information of the subject of diagnosis is characterized by including demographic information and maternal information if the time of diagnosis is before birth, including demographic information, maternal information, delivery information, newborn information, and vital signs at birth if the time of birth, and including demographic information, maternal information, delivery information, newborn information, vital signs at birth, vital signs for one week after birth, and disease information if one week after birth.
[0016] Each of the above first to post-birth prediction models is characterized by being implemented as one of LR (Logistic Regression with Ridge Regulation), RF (Random Forest), and XGB (eXtreme Gradient Boosting).
[0017] As a means to solve the above problem, according to another embodiment of the present invention, a data collection unit that collects and stores demographic information, maternal information, delivery information, newborn information, disease information, vital signs at birth, vital signs for one week after birth, and intraventricular hemorrhage diagnosis results from a previously registered medical database; a data preprocessing unit that generates a first feature value including demographic information and maternal information, a second feature value including delivery information, newborn information, and vital signs at birth additionally included in the first feature value, and a third feature value including vital signs for one week after birth and disease information additionally included in the second feature value; and a prediction model learning unit that trains a pre-birth prediction model based on the first feature value and intraventricular hemorrhage diagnosis results, trains a birth prediction model based on the second feature value and intraventricular hemorrhage diagnosis results, and trains a post-birth prediction model based on the third feature value and intraventricular hemorrhage diagnosis results. A machine learning-based integrated diagnostic support device for intraventricular hemorrhage and premature death within one week in extremely low birth weight infants is provided, comprising a diagnostic unit that, once the diagnostic subject and the diagnostic time are determined, selects a prediction model based on the diagnostic time, analyzes the medical information of the diagnostic subject through the selected prediction model, and predicts and outputs a diagnosis result of intraventricular hemorrhage. Effects of the invention
[0018] The present invention enables the diagnosis of whether intraventricular hemorrhage has occurred in extremely low birth weight infants using machine learning technology, thereby potentially reducing the incidence of permanent brain damage and the possibility of early death caused by intraventricular hemorrhage in the early stages of birth.
[0019] In addition, by enabling diagnosis by subdividing it into three time points—before birth, immediately after birth, and one week after birth—it is possible to more easily and accurately establish and carry out a treatment plan optimized for the time of diagnosis. Brief explanation of the drawing
[0020] FIG. 1 is a drawing illustrating a machine learning-based integrated diagnostic support method for intraventricular hemorrhage and early death within one week in extremely low birth weight infants according to an embodiment of the present invention. FIG. 2 is a drawing illustrating an example of medical information used for the learning and analysis of a prediction model according to an embodiment of the present invention. FIG. 3 is a drawing illustrating a machine learning-based integrated diagnostic support device for intraventricular hemorrhage and early death within one week in extremely low birth weight infants according to an embodiment of the present invention. Specific details for implementing the invention
[0021] Hereinafter, preferred embodiments are described in detail with reference to the attached drawings so that those skilled in the art can easily practice the present invention. However, in describing the preferred embodiments of the present invention in detail, if it is determined that a detailed description of related known functions or configurations may unnecessarily obscure the essence of the present invention, such detailed description is omitted. Additionally, the same reference numerals are used throughout the drawings for parts having similar functions and operations.
[0022] Additionally, throughout the specification, when a part is described as being 'connected' to another part, this includes not only cases where they are 'directly connected,' but also cases where they are 'indirectly connected' with other elements in between. Furthermore, the description of a component as 'including' means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.
[0023] FIG. 1 is a diagram illustrating a machine learning-based integrated diagnostic support method for the intraventricular hemorrhage status and the possibility of early death within one week in extremely low birth weight infants according to an embodiment of the present invention.
[0024] Referring to FIG. 1, the method of the present invention includes a data collection step (S1), a data preprocessing step (S2), a prediction model training step (S3), and a diagnosis execution step (S4), etc.
[0025] Data collection step (S1)
[0026] Among very low birth weight infants with a gestational age of 23 weeks or more and a birth weight of 500g or more but less than 1500g, a certain proportion of very low birth weight infants with mild intraventricular hemorrhage (IVH), very low birth weight infants with severe intraventricular hemorrhage, very low birth weight infants who died prematurely within one week, and very low birth weight infants in a normal state are selected.
[0027] Then, demographic information, maternal information, delivery information, newborn information, disease information, vital signs at birth, vital signs during the first week after birth, and diagnosis results of intraventricular hemorrhage of selected extremely low birth weight infants are collected and edited from a pre-established medical information database (DB) to create and store multiple data sets.
[0028] At this time, demographic information includes at least one of fetal sex and maternal age, maternal information includes at least one of number of pregnancies, whether in vitro fertilization occurred, maternal diabetes, maternal hypertension, and clinical chorioamnionitis, delivery information includes at least one of duration of premature rupture of membranes and mode of delivery, neonatal information includes at least one of oxygen saturation, electrocardiogram, and whether resuscitation was performed during delivery, gestational age, weight, 1-minute and 5-minute Apgar scores, pH, and base excess index, disease information includes at least one of pulmonary hemorrhage, respiratory distress syndrome, and the presence of hypotension requiring drug treatment, and vital signs may include at least one of oxygen saturation and electrocardiogram, but are not limited thereto.
[0029] In addition, the medical information database may be the Korean Neonatal Network (KNN) dataset, which was established in May 2013 with the support of the Korean Society of Neonatology and the Korea Centers for Disease Control and Prevention, and in which about 70 of the approximately 100 neonatal intensive care units (NICUs) in the country participate. The KNN prospectively registers about 2,000 to 2,400 very low birth weight infants (VLBWI) annually, and the data from the registered VLBWI is utilized as national big data for various research purposes.
[0030] Data preprocessing step (S2)
[0031] First, perform data preprocessing tasks on multiple data sets, such as handling missing values, handling outliers, handling duplicate data, and standardizing data formats.
[0032] Then, a first feature value is generated that includes demographic information and maternal information, a second feature value is generated that additionally includes delivery information, newborn information and vital signs at birth to the first feature value, and a third feature value is generated that additionally includes information about one week after birth and disease information to the second feature value.
[0033] That is, the present invention collects all medical information available from the mother's birth preparation period to the first week after birth, and classifies the collected data into three time points—the situation immediately before birth, the situation within the first hour after birth, and the first week after birth—to enable the training and analysis of a prediction model.
[0034] Predictive model training step (S3)
[0035] Multiple first training data sets are generated with a first feature value as an input condition and an intraventricular hemorrhage diagnosis result as an output condition, multiple second training data sets are generated with a second feature value as an input condition and an intraventricular hemorrhage diagnosis result as an output condition, and multiple third training data sets are generated with a third feature value as an input condition and an intraventricular hemorrhage diagnosis result as an output condition.
[0036] Then, a pre-birth prediction model is machine-trained using multiple first training data sets, a birth time prediction model is machine-trained using multiple second training data sets, and a post-birth prediction model is machine-trained using multiple third training data sets.
[0037] That is, three models are created—a pre-birth prediction model, a birth time prediction model, and a post-birth prediction model—having different learning conditions as shown in Fig. 2, and through these, an optimized prediction operation at the current time of diagnosis is performed.
[0038] More specifically, based on a plurality of data sets generated through the data collection step (S1), a plurality of first training data sets are generated with prenatal medical information of extremely low birth weight infants as a first input condition and a diagnosis result of intraventricular hemorrhage as an output condition, and then divided into training and validation sets in an 8:2 ratio according to the numerical distribution.
[0039] The dataset is split into a training set and a validation set using hyperparameters within the splitting function while maintaining a constant ratio of severe IVH / early death.
[0040] The validation set is used for model training and hyperparameter selection, and is used to evaluate the overall performance of the model. To compare the reference characteristics of extremely low birth weight infants in the validation set, a bivariate t-test is used for continuous variables and a chi-square test is used for categorical variables, and the significance of group differences is evaluated using p-values and standardized mean differences.
[0041] Continuous predictor variables are normalized using the min-max scaling method before being input into the model. This adjusts the range of numeric variable data points to between 0 and 1 while maintaining relative differences.
[0042] Synthetic prime oversampling technique (SMOTE) is utilized to address imbalances in the target variable, and this method promotes a balanced and effective learning process by augmenting the dataset by generating additional samples from the prime class within the derived data.
[0043] Training and tuning of the prediction model are performed using derived data through stratified k-fold cross-validation, and cross-validation of the training model is based on four iterations of the receiver operating characteristic (AUROC) region.
[0044] In addition, the first to post-birth prediction model of the present invention may be constructed using classification machine learning algorithms such as LR (Logistic Regression with Ridge Regulation), RF (Random Forest), and XGB (eXtreme Gradient Boosting), but is not limited thereto.
[0045] Diagnosis execution step (S4)
[0046] When machine learning of the diagnostic model is completed, and an unborn fetus or extremely low birth weight infant is selected as the diagnostic subject and the diagnostic time is determined, one of the pre-trained predictive models corresponding to the diagnostic time is selected.
[0047] After acquiring the medical information of the subject, the result of the diagnosis of intraventricular hemorrhage is predicted through analysis using a selected prediction model, and this is provided to the user audiovisually or to a pre-configured external device.
[0048] In this case, the diagnostician's medical information may also vary depending on the time of diagnosis. If the time of diagnosis is before birth, only demographic and maternal information is included; if it is immediately after birth, delivery information, newborn information, and vital signs at birth are additionally included in addition to demographic and maternal information; and if it is one week after birth, a third feature value is generated that additionally includes demographic information, maternal information, delivery information, newborn information, vital signs at birth, vital signs during the first week after birth, and disease information.
[0049] Thus, the present invention enables tracking and monitoring of the condition of extremely low birth weight infants over three time points—before birth, immediately after birth, and one week after birth—and allows for the pre-training of a time-specific prediction model by reflecting the results.
[0050] In addition, prenatal prediction models utilize prenatal data to predict premature birth or early death before delivery, helping medical staff prepare appropriate interventions in advance; birth prediction models utilize data up to one hour after delivery to predict whether intensive care unit management is necessary for extremely low birth weight infants; and birth prediction models utilize data up to one week after birth to prevent severe IVH or early death.
[0051] In other words, by using a time-based prediction model to accurately predict the condition of extremely low birth weight infants at the time of diagnosis and to establish an optimized treatment plan in advance based on this, it is possible to prevent problems such as permanent brain damage or death of extremely low birth weight infants due to IVH.
[0052] FIG. 3 is a diagram illustrating a machine learning-based integrated diagnostic support device for intraventricular hemorrhage and early death within one week in extremely low birth weight infants according to an embodiment of the present invention.
[0053] Referring to FIG. 3, the device of the present invention comprises: a data collection unit (10) that collects and stores demographic information, maternal information, delivery information, newborn information, disease information, vital signs at birth, vital signs after birth, and intraventricular hemorrhage diagnosis results from a previously registered medical database; a data preprocessing unit (20) that generates a first feature value including demographic information and maternal information, a second feature value including delivery information, newborn information, and vital signs at birth additionally included in the first feature value, and a third feature value including post-birth and disease information additionally included in the second feature value; a prediction model learning unit (30) that learns a pre-birth prediction model (M1) based on the first feature value and intraventricular hemorrhage diagnosis results, learns a birth prediction model (M2) based on the second feature value and intraventricular hemorrhage diagnosis results, and learns a post-birth prediction model (M3) based on the third feature value and intraventricular hemorrhage diagnosis results; and when a diagnosis subject and a diagnosis time are determined, one prediction model is selected based on the diagnosis time, and medical information of the diagnosis subject is analyzed through the selected prediction model. It includes a diagnostic unit (40) that predicts and outputs the diagnosis result of intraventricular hemorrhage, and a monitoring unit (50) that directly provides audiovisual guidance on the diagnosis result of intraventricular hemorrhage or provides it to a registered external device.
[0054] As described above, although the present invention has been explained by limited embodiments and drawings, the present invention is not limited to the above embodiments, and various modifications and variations are possible from this description by those skilled in the art to which the present invention belongs. Accordingly, the concept of the present invention should be understood only by the claims set forth below, and all equivalent or analogous variations thereof shall be considered to be within the scope of the concept of the present invention.
[0055] In addition, the apparatus and method according to the present invention can be implemented as computer-readable code on a computer-readable recording medium. A computer-readable recording medium includes all types of recording devices in which data that can be read by a computer system is stored. Examples of recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, hard disk, flash drive, etc., and also include implementation in the form of a carrier wave (e.g., transmission over the Internet). In addition, the computer-readable recording medium may be distributed across networked computer systems, and computer-readable code may be stored and executed in a distributed manner.
Claims
Claim 1 A machine learning-based integrated diagnostic support method for intraventricular hemorrhage and premature death within one week in extremely low birth weight infants, performed by an integrated diagnostic support device for the condition of extremely low birth weight infants, comprising: a data collection step of selecting extremely low birth weight infants with mild intraventricular hemorrhage, extremely low birth weight infants with severe intraventricular hemorrhage, extremely low birth weight infants who died prematurely within one week, and extremely low birth weight infants in a normal state according to a preset ratio, and then collecting and storing demographic information, maternal information, delivery information, newborn information, disease information, vital signs at birth, vital signs for one week after birth, and intraventricular hemorrhage diagnostic results for each of the selected extremely low birth weight infants from a preset medical database; a first feature value reflecting the situation immediately before birth including demographic information and maternal information; a second feature value reflecting the situation within the first hour after birth including delivery information, newborn information, and vital signs at birth in addition to the first feature value; and a birth value including vital signs for one week after birth and disease information in addition to the second feature value. A machine learning-based integrated diagnostic support method for intraventricular hemorrhage and early death within one week in extremely low birth weight infants, comprising: a data preprocessing step for generating a third feature value that reflects the situation of the first week thereafter; a learning step for training a pre-birth prediction model based on the first feature value and the intraventricular hemorrhage diagnosis result, training a birth prediction model based on the second feature value and the intraventricular hemorrhage diagnosis result, and training a post-birth prediction model based on the third feature value and the intraventricular hemorrhage diagnosis result; and a diagnosis step for selecting one prediction model based on the diagnosis time when the diagnosis subject and the diagnosis time are determined, and then analyzing the medical information of the diagnosis subject through the selected prediction model to predict and output an intraventricular hemorrhage diagnosis result, wherein the intraventricular hemorrhage diagnosis result includes at least one of the following: whether premature birth occurred before delivery, whether intensive care unit management is required, whether severe IVH is present, and whether early death occurred. Claim 2 A machine learning-based integrated diagnostic support method for intraventricular hemorrhage and premature death within one week in extremely low birth weight infants, characterized in that, in claim 1, the demographic information includes at least one of fetal sex and maternal age, the maternal information includes at least one of number of pregnancies, whether in vitro fertilization occurred, maternal diabetes, maternal hypertension, and clinical chorioamnionitis, the delivery information includes at least one of duration of premature rupture of membranes and mode of delivery, the neonatal information includes at least one of oxygen saturation, electrocardiogram, whether resuscitation was performed during delivery, gestational age, weight, 1-minute and 5-minute Apgar scores, pH, and base excess index, the disease information includes at least one of pulmonary hemorrhage, respiratory distress syndrome, and the presence or absence of hypotension requiring drug treatment, and the vital signs include at least one of oxygen saturation and electrocardiogram. Claim 3 A machine learning-based integrated diagnostic support method for intraventricular hemorrhage and premature death within one week in extremely low birth weight infants, wherein, in claim 1, the learning step divides the learning data for training the pre-birth prediction model, the birth time prediction model, and the post-birth prediction model into a training set and a validation set according to a pre-set ratio of severe IVH and premature death, and performs training and tuning of each prediction model through stratified k-fold cross-validation and receiver operation characteristic (AUROC)-based model optimization. Claim 4 A machine learning-based integrated diagnostic support method for intraventricular hemorrhage and early death within one week in extremely low birth weight infants, characterized in that, in claim 1, each of the first to post-birth prediction models is implemented as any one of LR (Logistic Regression with Ridge Regulation), RF (Random Forest), and XGB (eXtreme Gradient Boosting). Claim 5 A data collection and storage unit that selects infants with mild intraventricular hemorrhage, infants with severe intraventricular hemorrhage, infants who died prematurely within one week, and infants in a normal state according to a preset ratio, and then collects and stores demographic information, maternal information, delivery information, newborn information, disease information, vital signs at birth, vital signs for one week after birth, and intraventricular hemorrhage diagnosis results for each of the selected infants from a pre-registered medical database; a data preprocessing unit that generates a first feature value reflecting the situation immediately before birth including demographic information and maternal information, a second feature value reflecting the situation within the first hour after birth including delivery information, newborn information, and vital signs at birth in addition to the first feature value, and a third feature value reflecting the situation during the first week after birth including vital signs for one week after birth and disease information in addition to the second feature value; and a pre-birth prediction model that trains the first feature value and intraventricular hemorrhage diagnosis results, and the second A machine learning-based integrated diagnostic support device for the intraventricular hemorrhage status and the possibility of early death within one week of an extremely low birth weight infant, comprising: a prediction model learning unit that trains a prediction model at birth based on a feature value and an intraventricular hemorrhage diagnosis result, and trains a prediction model after birth based on the third feature value and an intraventricular hemorrhage diagnosis result; and a diagnosis unit that, when a diagnosis subject and a diagnosis time are determined, selects one prediction model based on the diagnosis time, analyzes the medical information of the diagnosis subject through the selected prediction model, and predicts and outputs an intraventricular hemorrhage diagnosis result, wherein the intraventricular hemorrhage diagnosis result includes at least one piece of information among whether premature birth occurred before delivery, whether intensive care unit management is required, whether severe IVH is present, and whether early death occurs.
Citation Information
Patent Citations
Systems and methods to assess neonatal health risk and uses thereof
WO2023164308A2
