KR102587466B1 - Predicting method for recovery of neutrophil counts and apparatus tereof - Google Patents

Abstract

The present invention relates to a method and device for predicting neutrophil count recovery. The method for predicting neutrophil count recovery performed in a computing device according to an embodiment of the present invention includes a data acquisition step of acquiring anticancer treatment data of a target patient, the anticancer treatment It includes a data preprocessing step of preprocessing the data, and a step of calculating the length of the risk period during which the absolute neutrophil count (ANC) after anticancer treatment is below a preset reference value using a first artificial intelligence model.

Description

Neutrophil count recovery prediction method and device {PREDICTING METHOD FOR RECOVERY OF NEUTROPHIL COUNTS AND APPARATUS TEREOF}

The present invention relates to a method and device for predicting neutrophil count recovery, and more specifically, to a method and device for more accurately predicting the risk period based on the value of the absolute neutrophil count after anticancer treatment using an artificial intelligence model. It's about.

After anticancer treatment, bone marrow suppression may occur due to the toxicity of anticancer drugs to hematopoietic stem cells. Bone marrow suppression is a temporary symptom and may recover naturally over time, but the degree of suppression and duration of bone marrow suppression may vary depending on various factors.

When the production of normal blood cells decreases due to this, the number of white blood cells in the blood decreases, and immunity to bacterial infections decreases. In particular, when the number of neutrophils among white blood cells decreases below 500/uL, the risk of developing sepsis due to bacterial infection greatly increases, which is one of the main causes of death due to toxicity of anticancer drug treatment.

It is difficult to predict the degree of toxicity of anticancer treatment for individual patients and the duration of symptoms due to this, as it varies from person to person. In addition, this information continuously changes in time series, and there are limitations in applying general statistical methods due to the interaction of various factors, such as treatment of the disease and accumulation of toxicity due to treatment. Therefore, an artificial intelligence model that can learn by collecting irregularly collected data is needed.

A method of learning by collecting irregularly collected data may be to rearrange the time intervals of the collected data to be consistent, but in this process, the value of the data may be damaged and data points may be reduced. Instead, you can set an interpolation function that includes parameters that the artificial intelligence model can learn, allowing you to select a value more suitable for the learning criteria. Also, without interpolating the values, you can input the collected data and the time at which the data was collected into the artificial intelligence model, allowing the model to recognize and learn that the time interval is irregular.

Korean Patent Registration No. 10-2017-0137212 (October 23, 2017) provides treatment prognoses tailored to the user's requirements for various types of cancer and treatment methods through data mining techniques, allowing users to receive the most appropriate cancer treatment. It relates to a method and system for providing cancer treatment prediction results that can identify methods, which includes the steps of collecting and data mining cancer treatment-related literature or case information to build a semantic database dictionary; When at least one of cancer type, treatment method, side effect type, and patient condition information is input as input information, constructing a treatment prediction model based on the input information and the semantic database dictionary; and providing result information corresponding to the input information based on the treatment prediction model.

However, when collecting and analyzing medical data recorded at irregular time intervals, it is difficult for artificial intelligence to properly learn and analyze, which may lead to inaccurate results.

1. Korean Patent Registration No. 10-2017-0137212 (2017.10.23)

The first technical problem that the present invention aims to solve is to provide a method for predicting neutrophil count recovery for predicting the risk period after anticancer treatment.

The second technical problem that the present invention aims to solve is to provide a neutrophil count recovery prediction device for predicting the risk period after anticancer treatment.

In order to solve the above-described technical problem, an embodiment of the present invention provides a method for predicting neutrophil count recovery performed on a computing device.

In one embodiment, the method for predicting neutrophil count recovery includes a data acquisition step of acquiring anticancer treatment data of a target patient, a data preprocessing step of preprocessing the anticancer treatment data, and a first artificial intelligence model, after anticancer treatment. It may include calculating the length of the risk period during which the absolute neutrophil count (ANC) is below a preset reference value.

In one embodiment, the step of calculating the length of the risk period includes calculating a relative date, which is a period of time elapsed based on the start date of the anticancer cycle, by the first artificial intelligence model, and calculating the relative date by the first artificial intelligence model. Based on this, calculating interpolation data of the anti-cancer treatment data according to a preset date interval, and calculating the length of the risk period by the first artificial intelligence model based on the anti-cancer treatment data and the interpolation data. may include.

In one embodiment, the step of calculating interpolation data includes setting, by the first artificial intelligence model, a standard value for each variable included in the anticancer treatment data, and interpolating where the first artificial intelligence model includes a first parameter. Setting a function, and calculating interpolation values at preset date intervals using the standard value and observed values included in the anticancer treatment data, wherein the first parameter is the first artificial intelligence This may be done by comparing the length of the risk period predicted by the model with the correct answer values, and updating it through repeated learning so that the comparison result exceeds the learning standard.

In one embodiment, the first artificial intelligence model may include the GRU-D algorithm.

The method for predicting neutrophil count recovery performed in a computing device according to an embodiment of the present invention includes a data acquisition step of acquiring anticancer treatment data of a target patient, a data preprocessing step of preprocessing the anticancer treatment data, and a second artificial intelligence model. Predicting the change pattern of the absolute neutrophil count during a preset period from the start date of the anticancer cycle and determining the absolute neutrophil count by mistake using the second artificial intelligence model, and using the second artificial intelligence model to predict the absolute neutrophil count during a preset period from the start date of the anticancer cycle It may include a step of determining whether the risk period is below the set standard value.

In one embodiment, the step of determining whether it is included in the risk period includes generating, by the second artificial intelligence model, a first derived variable based on the absolute neutrophil count, wherein the second artificial intelligence model, It may include the step of determining that the first derived variable is included in the risk period when it is less than a preset reference value, and determining that it is not included in the risk period when the first derived variable is more than a preset reference value. .

In one embodiment, the second artificial intelligence model may include a Transformer algorithm, an Attention algorithm, or a combination thereof.

In the method for predicting neutrophil count recovery performed on a computing device according to an embodiment of the present invention, a data acquisition step of acquiring anticancer treatment data of a plurality of patients, a data preprocessing step of preprocessing the anticancer treatment data, and the preprocessed It may include learning an artificial intelligence model using anticancer treatment data.

In one embodiment, the data preprocessing step includes de-identifying the anti-cancer treatment data by deleting personal information, classifying the de-identified anti-cancer treatment data into general anti-cancer treatment and high-dose anti-cancer treatment, and de-identifying the anti-cancer treatment data. Displaying an anti-cancer cycle of treatment data, inputting a relative date, which is a period of time elapsed from the start date of the anti-cancer cycle, to the anti-cancer treatment data, absolute neutrophil count (ANC) and absolute lymphocytes missing from the anti-cancer treatment data It may include calculating a count (Absolute Lymphocyte Count, ALC) from other variables, and removing variables with a count less than a preset number.

One embodiment according to the present invention provides a computer program stored in a computer-readable storage medium to execute the method for predicting neutrophil count recovery according to any of the above-described embodiments using a computer.

In order to solve the above-described technical problem, an embodiment of the present invention provides a memory for storing anti-cancer treatment data and a first artificial intelligence model, and pre-processes the anti-cancer treatment data and uses the first artificial intelligence model to prevent cancer. Provided is a neutrophil count recovery prediction device, including a processor, configured to calculate the length of the risk period during which the absolute neutrophil count after treatment is below a preset reference value.

In one embodiment, the memory further stores a second artificial intelligence model, and the processor predicts the change pattern of the absolute neutrophil count during a preset period from the start date of the anticancer cycle using the second artificial intelligence model. It can be configured to return a mistake.

In one embodiment, the memory further stores a second artificial intelligence model, and the processor uses the second artificial intelligence model to determine whether a preset period from the start date of the anticancer cycle is included in the risk period. It can be configured.

The present invention uses an artificial intelligence model that predicts the risk period through repeated learning to more accurately calculate the pattern of absolute neutrophil count changes in the risk period after anticancer treatment, whether a specific date includes the risk period, and the length of the risk period. there is.

Therefore, based on this, it is possible to prepare for patient infection, reduce the number of blood samples from patients, adjust the timing and duration of hospital visits in advance, and medical staff can adjust the overall treatment schedule by adjusting the intensity of anticancer drug treatment. You can. The technical effects of the present invention are not limited to those mentioned above, and other technical effects not mentioned will be clearly understood by those skilled in the art from the description below.

Figure 1 is a flowchart showing a method for predicting neutrophil count recovery according to an embodiment of the present invention.
FIG. 2 is a flow chart to explain in more detail the steps for calculating the length of the risk period shown in FIG. 1.
FIG. 3 is a flowchart for explaining in more detail the steps of calculating interpolation data shown in FIG. 2.
Figure 4 is a flowchart showing a method for predicting neutrophil count recovery according to an embodiment of the present invention.
FIG. 5 is a flowchart for explaining in more detail the steps of determining whether or not an item is included in the risk period shown in FIG. 4.
Figure 6 is a table showing the change in absolute neutrophil count over 30 days predicted using the second artificial intelligence model.
Figure 7 is a table indicating whether the risk period for 30 days predicted using the second artificial intelligence model is included.
Figure 8 is a flowchart showing the learning method of an artificial intelligence model.
FIG. 9 is a flow chart to explain the data preprocessing step shown in FIG. 8 in more detail.
Figure 10 is a block diagram showing a neutrophil count recovery prediction device according to an embodiment of the present invention.
Figure 11 is a conceptual diagram for explaining a first artificial intelligence model according to an embodiment of the present invention.
Figure 12 is a conceptual diagram for explaining a second artificial intelligence model according to an embodiment of the present invention.

While the invention is susceptible to various modifications and variations, specific embodiments thereof are illustrated in the drawings and will be described in detail below. However, it is not intended to limit the invention to the particular form disclosed, but rather, the invention is intended to cover all modifications, equivalents and substitutions consistent with the spirit of the invention as defined by the claims.

The invention may be represented by functional block configurations and various processing steps. These functional blocks may be implemented in various numbers of hardware or/and software configurations that execute specific functions. For example, the present invention provides integrated circuit configurations, such as memory, processing, logic, look-up tables, etc., that can execute various functions under the control of one or more microprocessors or other control devices. can be hired.

Similar to the fact that the components of the invention can be implemented as software programming or software elements, the invention also includes various algorithms implemented as combinations of data structures, processes, routines or other programming constructs, including C, C++, , may be implemented in a programming or scripting language such as Java, assembler, etc.

Additionally, the present invention can employ conventional technologies for electronic environment settings, signal processing, and/or data processing. Terms such as “part,” “element,” “means,” and “configuration” may be used broadly, and the components of the present invention are not limited to mechanical and physical configurations. The term may include the meaning of a series of software routines in connection with a processor, etc.

It will be understood that when an element such as a layer, region or substrate is referred to as being “on” another element, it may be present directly on the other element or there may be intermediate elements in between. .

Although the terms first, second, etc. may be used to describe various elements, components, regions, layers and/or blocks, such elements, components, regions, layers and/or blocks It will be understood that we should not be limited by these terms.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the attached drawings. Hereinafter, the same reference numerals will be used for the same components in the drawings, and duplicate descriptions of the same components will be omitted.

Figure 1 is a flowchart showing a method for predicting neutrophil count recovery according to an embodiment of the present invention.

Referring to FIG. 1, the method for predicting neutrophil count recovery according to an embodiment of the present invention includes acquiring anticancer treatment data (S110), preprocessing the data (S120), and using a first artificial intelligence model. It may include calculating the length of the risk period (S130).

In the step of acquiring anti-cancer treatment data (S110), the neutrophil count recovery prediction device acquires the anti-cancer treatment data of the target patient. Here, the target patient's anticancer treatment data may include lab data, G-CSF administration data, and anticancer drug administration details.

The target patient's anticancer treatment data may be managed by an Electronic Medical Record (EMR) or Electronic Health Records (EHR) system.

The target patient's anticancer treatment data may be provided to the neutrophil count recovery prediction device by a person in charge of providing direct medical services, that is, a doctor or electronic health record manager. In another embodiment, the neutrophil count recovery prediction device may obtain anti-cancer treatment data of the target patient through a medical data system. Here, the medical data system includes the integrated medical information system, electronic medical record, laboratory information system, data warehouse, and clinical research support system. It may be a system necessary for medical services and research, such as an information system.

In the data preprocessing step (S120), the target patient's anticancer treatment data may be preprocessed into a form for input into the first artificial intelligence model.

The target patient's anticancer treatment data may include 1) the time at the time of recording, 2) the name of the variable or test item, and 3) the value of the variable or test item. In the preprocessing stage, the target patient's anticancer treatment data can be 1) indicated with an anticancer cycle, and 2) given a relative date, which is the period of time elapsed based on the start date of the anticancer cycle.

In addition, variables that are not used by the first artificial intelligence model among the target patient's anticancer treatment data are removed in advance, or variables that are used by the first artificial intelligence model are not included in the raw data of the anticancer treatment data. Variables can be calculated and entered from other variables.

For example, if the raw data of anticancer treatment data is missing the absolute neutrophil count for the relevant date, the absolute neutrophil count can be calculated and entered from band neutrophils and segmented neutrophils. As another example, if the absolute lymphocyte count for the relevant date is missing in the raw data of the anticancer treatment data, the absolute lymphocyte count can be calculated and entered from the lymphocyte measurement value and white blood cell (WBC) count. .

In the step of calculating the length of the risk period using the first artificial intelligence model (S130), the length of the risk period in which the absolute neutrophil count after anticancer treatment is less than or equal to a preset reference value can be calculated using the first artificial intelligence model. .

Here, the first artificial intelligence model may be a deep learning algorithm for receiving variables as input and classifying the resulting values into multiple classes (multiclass classification) and returning them. The first artificial intelligence model may include a RNN algorithm. As an example, the first artificial intelligence model may include the GRU-D algorithm.

Here, the risk period can be defined as an absolute neutrophil count (ANC) of less than 500 (units/mm ³ ) for more than 3 consecutive days.

The first artificial intelligence model can return the length of the expected future risk period as an integer based on the target patient's anticancer treatment data from immediately after the start of anticancer treatment until immediately after the absolute neutrophil count was first measured below 500.

A doctor or nurse providing medical services can determine the treatment schedule for a target patient based on the calculated length of the risk period. For example, a doctor or nurse can adjust the intensity and schedule of anticancer drug treatment for a target patient. Target patients or their guardians can determine in advance the timing and duration of hospital visits, adjust activities during the risk period when immunity decreases, and prepare for hospitalization or in-office treatment.

FIG. 2 is a flow chart to explain in more detail the step (S130) of calculating the length of the risk period shown in FIG. 1.

Referring to Figure 2, the step of calculating the length of the risk period (S130) includes the step of calculating the relative period (S131), the step of calculating interpolation data (S132), and the length of the risk period based on the relative period and interpolation data. It includes a step (S133) of calculating .

In the step of calculating the relative period (S131), the first artificial intelligence model calculates the relative date, which is the period elapsed based on the start date of the anticancer cycle. Here, the relative date may be a relative date given to the anticancer treatment data in the preprocessing step (S120). Here, anticancer treatment data may not have regular time intervals.

In the step of calculating interpolation data (S132), the first artificial intelligence model may calculate interpolation data to compensate for variables that do not exist according to a preset date interval based on the relative date.

In the case of artificial intelligence models including RNN algorithms, there is a problem that the model's prediction results become inaccurate if the input data does not have a certain time interval. Therefore, a method of artificially dividing the input data in the preprocessing step or generating and inputting interpolated data is used. However, in the method for predicting neutrophil count recovery according to an embodiment of the present invention, anticancer treatment data and relative dates can be input together into the first artificial intelligence, so that the first artificial intelligence can calculate appropriate interpolation data on its own. In other words, when anticancer treatment data corresponding to a preset date interval is not input, the first artificial intelligence calculates and generates interpolation data on its own.

In the step of calculating the length of the risk period (S133), the first artificial intelligence model calculates the length of the risk period based on the anticancer treatment data and interpolation data.

That is, the first artificial intelligence model can estimate and select one value among the lengths of possible risk periods by using anticancer treatment data and interpolation data as input. In one embodiment, the length of the possible risk period can be divided into 14 classes, from 1 to 14 days, at 1-day intervals, and the first artificial intelligence determines the expected risk period of the target patient based on the input data. You can calculate and return which class the length belongs to among the 14 classes.

FIG. 3 is a flowchart for explaining in more detail the step (S132) of calculating interpolation data shown in FIG. 2.

Referring to FIG. 3, the step of calculating interpolation data (S132) includes setting a standard value for each variable (S141), setting an interpolation function (S142), calculating interpolation values (S143), and 1 It includes a step of updating parameters (S144).

In the step of setting standard values for each variable (S141), the first artificial intelligence model may set standard values for each variable included in the anticancer treatment data. In one embodiment, the first artificial intelligence model may set the average value of the variable values of a plurality of patients on the corresponding relative date as the standard value for the variable included in the anticancer treatment data. In another embodiment, the first artificial intelligence model may set the median value of the variable values of a plurality of patients on the corresponding relative date as the standard value for the variables included in the anticancer treatment data.

In the step of setting the interpolation function (S142), the first artificial intelligence model may set an interpolation function including the first parameter using a standard value.

In one embodiment, the first artificial intelligence model may be set by considering the interpolation function to decay exponentially with time from the last observed value of the variable included in the anticancer treatment data toward the standard value.

In the step of calculating interpolation values (S143), the first artificial intelligence model inputs observed values of variables included in the anticancer treatment data into a set interpolation function, calculates interpolation values at preset date intervals, and includes these. Interpolation data can be generated.

In the step of updating the first parameter (S144), the length of the risk period predicted by the first artificial intelligence model is compared with the correct answer values, and the comparison can be repeated through repeated learning so that it exceeds the learning standard. The step of updating the first parameter (S144) may be included in the step of training the first artificial intelligence model using the training data set, but is not limited to this. In one embodiment, the first artificial intelligence model calculates interpolation values from the target patient's anti-cancer treatment data, and when anti-cancer treatment data of the corresponding date is generated through the patient's treatment, the interpolation values and variables included in the anti-cancer treatment data The first parameter can be updated by comparing the actual observed values of .

In one embodiment, the first artificial intelligence model operates when a variable has an observed value x _t at time t, there is no observed value from time t to time t+2, and the standard value of this variable is m. , a function that interpolates the values of the variable up to time t+2 It can be defined as: Here, r may be the first parameter that the first artificial intelligence updates through repeated learning to find the optimal interpolation function.

Therefore, in the preprocessing process, interpolation values can be determined according to predetermined criteria, or interpolation values more suitable than the conventional technique of arbitrarily dividing data can be determined and used as input data. Therefore, the method for predicting neutrophil count recovery according to an embodiment of the present invention may have higher prediction performance than conventional artificial intelligence models.

Figure 4 is a flowchart showing a method for predicting neutrophil count recovery according to an embodiment of the present invention.

Referring to FIG. 4, the method for predicting neutrophil count recovery according to an embodiment of the present invention includes obtaining anticancer treatment data (S210), preprocessing the data (S220), and using a second artificial intelligence model to determine absolute It may include a step of determining the change pattern of the neutrophil count (S230) and a step of determining whether the date is included in the risk period using a second artificial intelligence model (S240).

The step of acquiring anti-cancer treatment data (S210) may be a data acquisition step in which the neutrophil count recovery prediction device acquires anti-cancer treatment data of the target patient. Here, the target patient's anticancer treatment data may include lab data, G-CSF administration data, and anticancer drug administration details.

The target patient's anticancer treatment data may be managed by an Electronic Medical Record (EMR) or Electronic Health Records (EHR) system.

The target patient's anticancer treatment data may be provided to the neutrophil count recovery prediction device by a person in charge of providing direct medical services, that is, a doctor or electronic health record manager. In another embodiment, the neutrophil count recovery prediction device may obtain anti-cancer treatment data of the target patient through a medical data system. Here, the medical data system includes the integrated medical information system, electronic medical record, laboratory information system, data warehouse, and clinical research support system. It may be a system necessary for medical services and research, such as an information system.

The data preprocessing step (S220) may be a step of preprocessing the anticancer treatment data obtained by the neutrophil count recovery prediction device into a form for inputting it into a second artificial intelligence model.

The target patient's anticancer treatment data may include 1) the time at the time of recording, 2) the name of the variable or test item, and 3) the value of the variable or test item. In the preprocessing stage, the target patient's anticancer treatment data can be 1) indicated with an anticancer cycle, and 2) given a relative date, which is the period of time elapsed based on the start date of the anticancer cycle.

In addition, variables that are not used by the second artificial intelligence model among the target patient's anticancer treatment data are removed in advance, or variables that are used by the second artificial intelligence model are not included in the raw data of the anticancer treatment data. Variables can be calculated and entered from other variables.

For example, if the raw data of anticancer treatment data is missing the absolute neutrophil count for the relevant date, the absolute neutrophil count can be calculated and entered from band neutrophils and segmented neutrophils. As another example, if the absolute lymphocyte count for the relevant date is missing in the raw data of the anticancer treatment data, the absolute lymphocyte count can be calculated and entered from the lymphocyte measurement value and white blood cell (WBC) count. .

According to an embodiment of the present invention, input data of the first artificial intelligence model and the second artificial intelligence model may include the same variables. As an example, anticancer treatment data used as input data for the first artificial intelligence model and the second artificial intelligence model may include the following variables obtained through a blood test.

WBC Basophile Protein, Total R.B.C. Lymphocyte Albumin Hemoglobin Monocyte Globulin Hematocrit Atypical Lymphocyte A/G ratio MCV Immature cell Cholesterol MCH Plasma cell AST MCHC Nucleated RBCs ALT Platelet ANC ALP Blast ALC Glucose, Fasting Promyelocyte Abnormal Lymphoid cell BUN Myelocyte MPV Creatinine Metamyelocyte Reticulocyte BUN&Cr. ratio Band neutrophil RDW Estimated GFR Segmented neutrophil PDW Bilirubin, Total Eosinophil IRF Uric Acid Ca Potassium (K) Na P Cl CRP, Quantitative (High Sensitivity)

As an example, the anticancer treatment data may include the patient's age, gender, anticancer drug administration information, G-CSF administration information, and platelet transfusion information. At this time, anticancer drug administration information may be entered using the derived variable intensity, which does not reflect the type of anticancer drug. Here, the derived variable may be a calculation of the actual administered dose compared to the planned administered dose. If several types of anticancer drugs are administered on the same day, the product of the actual administered dose compared to the planned administered dose of each anticancer drug can be used as the derived variable. G-CSF administration information and platelet transfusion information do not reflect the administration dose and may only include administration status as a binary value. In one embodiment, the input data may have a form such as the table below.

relative date variable name variable value 0 WBC 4.88 0 Monocyte 10.2 0 Myelocyte 0 One WBC 3.28 One Hemoglobin 9.7 One Bilirubin, Total 0.5 One R.B.C. 3 2 WBC 3.2 2 Band neutrophil 2 2 Na 140 3 WBC 8.66 3 Metamyelocyte 0 3 ANC 7.794

At this time, since the test contents performed on the target patient may be different for each relative date, all of the above variables may not exist simultaneously in the anticancer treatment data for each relative date. Absolute neutrophil count using the second artificial intelligence model In the step of determining the change pattern (S230), using the second artificial intelligence model, the change pattern of the absolute neutrophil count is predicted at 1-day intervals for a preset period from the start date of the anti-cancer cycle and returned as a real value. This may be a step.

Here, the second artificial intelligence model may be a deep learning algorithm for reducing errors and estimating more accurate values by comparing the estimated value with the actual value. The second artificial intelligence model may include an algorithm that assigns importance to and processes specific parts even when the length of the input data is variable. As an example, the second artificial intelligence model may include a transformer algorithm.

The second artificial intelligence model may directly track the absolute neutrophil count, predict the pattern of change, and return the absolute neutrophil count value for that date as a real number. In another embodiment, the second artificial intelligence model may generate a derived variable based on the absolute neutrophil count, predict the change pattern of the derived variable, and return the value of the derived variable for the relevant date as a real number.

The step of determining whether the date is included in the risk period using the second artificial intelligence model (S240) uses the second artificial intelligence model to determine the absolute neutrophil count during a preset period from the start date of the anticancer cycle to a preset reference value. This may be the step of determining whether or not it is included in the risk period below or not as a binary value and returning it.

In one embodiment, the second artificial intelligence model may return 1 if it determines that the relevant date is included in the risk period, and may return 0 if it determines that the relevant date is not included in the risk period.

The method for predicting neutrophil count recovery according to an embodiment of the present invention includes calculating the length of the risk period using the first artificial intelligence model described above, and using the second artificial intelligence model to calculate the length of the risk period from the start date of the anticancer cycle. At least one of the steps of predicting the change pattern of the absolute neutrophil count during the period and making a mistake decision and using a second artificial intelligence model to determine whether the date is included in the risk period can be included as needed. there is. In one embodiment, the target patient or guardian is provided with the length of the risk period and whether the date is included in the risk period, and can adjust the activity schedule based on this.

However, it is not limited to this embodiment, and the method for predicting neutrophil count recovery according to an embodiment of the present invention includes the steps of calculating the length of the risk period using the first artificial intelligence model described above, and the second artificial intelligence model. A step of predicting the change pattern of absolute neutrophil count during a preset period from the start date of the anticancer cycle and making a mistake decision using a second artificial intelligence model, and a step of determining whether the date is included in the risk period using a second artificial intelligence model. It may also be included.

FIG. 5 is a flowchart to explain in more detail the step (S240) of determining whether the time period shown in FIG. 4 is included in the risk period.

Referring to FIG. 5, the step of determining whether it is included in the risk period using the second artificial intelligence model (S240) includes the step of generating a first derived variable (S241) and comparing the first derived variable and the reference value. Thus, a step may be included to determine whether the date is included in the risk period (S243) or not (S245).

First, in the step of generating the first derived variable (S241), the second artificial intelligence model may generate the first derived variable based on the absolute neutrophil count. By using the first derived variable, it is possible to remove noise that may occur in each test and more accurately determine whether it is included in the risk period.

The second artificial intelligence model may compare the first derived variable and the reference value, and if the first derived variable is less than the reference value on the relevant date, it may be determined that the relevant date is included in the risk period (S243). In this case, the second artificial intelligence model may return 1 as the result value for that date.

The second artificial intelligence model compares the first derived variable and the reference value, and if the first derived variable is greater than the reference value on the relevant date, it can be determined that the relevant date is not included in the risk period (S235). In this case, the second artificial intelligence model may return 0 as the result value for that date.

Figure 6 is a table showing the change in absolute neutrophil count over 30 days predicted using the second artificial intelligence model.

Referring to Figure 6, the second artificial intelligence model can determine and return the change pattern of the absolute neutrophil count according to the relative date as a real value at 1-day intervals from the start date of the anti-cancer cycle.

Figure 7 is a table indicating whether the risk period for 30 days predicted using the second artificial intelligence model is included.

Referring to Figure 7, the second artificial intelligence model can determine whether the date is included in the risk period at intervals of 1 day from the start date of the anticancer cycle and return the value as 1 or 0. That is, if the value of the first derived variable created based on the absolute neutrophil count is less than the standard value, it is judged to be a risk period and 1 is returned, and if the value of the first derived variable is greater than the standard value, it is judged not to be a risk period. It can return 0.

Figure 8 is a flowchart showing the learning method of an artificial intelligence model.

Referring to FIG. 8, the method of learning an artificial intelligence model according to an embodiment of the present invention includes acquiring anticancer treatment data of a plurality of patients (S310), preprocessing the anticancer treatment data (S320), and artificial intelligence model. It includes a step of learning (S330) and a step of updating the artificial intelligence model by comparing the result value of the learned artificial intelligence model with the correct answer value (S340).

In the step of acquiring anti-cancer treatment data of a plurality of patients (S310), the neutrophil count recovery prediction device acquires anti-cancer treatment data of a plurality of patients. Here, the anticancer treatment data of multiple patients may include lab data, G-CSF administration data, and anticancer drug administration details.

Anticancer treatment data for multiple patients may be provided by an Electronic Medical Record (EMR) or Electronic Health Records (EHR) system.

In the step of preprocessing the anticancer treatment data (S320), the neutrophil count recovery prediction device may process the anticancer treatment data of a plurality of patients into a form that can be input into the first artificial intelligence model or the second artificial intelligence model.

In the step of learning the artificial intelligence model (S330), the artificial intelligence model may perform repeated learning using preprocessed anticancer treatment data. As an example, anticancer treatment data may have the form of a learning data set with input variables and correct answer values. An artificial intelligence model can derive result values using input variables included in the learning data set.

In the step of updating the artificial intelligence model by comparing the result value of the learned artificial intelligence model with the correct answer value (S340), the artificial intelligence model uses the result value derived using the input variables included in the learning data set and the learning data By comparing the correct answer values included in the set, each parameter of the artificial intelligence model can be updated so that the comparison result is higher than a preset standard value.

The first artificial intelligence model and the second artificial intelligence model used in the method for predicting neutrophil count recovery according to an embodiment of the present invention may be learned in advance through the above-described artificial intelligence model learning method, but are not limited thereto. In one embodiment, the first artificial intelligence model or the second artificial intelligence model receives the target patient's anticancer treatment data, calculates a predicted value based on this, and compares the calculated predicted value with the target patient's new medical record. You can learn by updating the artificial intelligence model.

FIG. 9 is a flow chart to explain the data preprocessing step shown in FIG. 8 in more detail.

Referring to Figure 9, the data preprocessing step (S320) includes a de-identification step (S321), a step of classifying general anti-cancer treatment and high-dose anti-cancer treatment (S322), a step of displaying the anti-cancer cycle (S323), and a step of classifying the anti-cancer cycle (S323). It includes a step of displaying the elapsed period from the start date (S324), a step of calculating missing ANC and ALC (S325), and a step of removing unnecessary variables (S326).

In the de-identification step (S321), the neutrophil count recovery device can de-identify the cancer treatment data of multiple patients by deleting personal information that can identify the patient.

In the step of classifying general anti-cancer treatment and high-dose anti-cancer treatment (S322), the neutrophil count recovery device can classify and display general anti-cancer treatment and high-dose anti-cancer treatment. Compared to the large number of general anti-cancer treatment data, the high-dose anti-cancer treatment data is not only small in number, but also has different distributions of risk periods, so it is necessary to mix the two data and use them as training and validation data to learn an artificial intelligence model. .

In the step of displaying the anti-cancer cycle (S323) and the step of displaying the elapsed period from the start date of the anti-cancer cycle (S324), the anti-cancer cycle of the anti-cancer treatment data is displayed and the relative date, which is the period of time elapsed from the start date of the anti-cancer cycle, is entered. .

In the step of calculating missing ANC and ALC (S325) and the step of removing unnecessary variables (S326), variables that are not used by the first artificial intelligence model or the second artificial intelligence model among the anticancer treatment data of multiple patients are removed in advance. Alternatively, variables that are not included in the raw data of the anticancer treatment data can be calculated from other variables and input.

For example, if the raw data of anticancer treatment data is missing the absolute neutrophil count for the relevant date, the absolute neutrophil count can be calculated and entered from band neutrophils and segmented neutrophils. As another example, if the absolute lymphocyte count for the relevant date is missing in the raw data of the anticancer treatment data, the absolute lymphocyte count can be calculated and entered from the lymphocyte measurement value and white blood cell (WBC) count. .

In one embodiment, in order to increase the number of learning data, each patient's anticancer treatment data may be segmented into anticancer treatment cycle units.

In one embodiment, the anticancer treatment data may be refined by removing variables whose total number is less than a preset number among the variables included in the anticancer treatment data.

Figure 10 is a block diagram showing a neutrophil count recovery prediction device according to an embodiment of the present invention.

Referring to FIG. 10, the neutrophil count recovery prediction device 100 according to an embodiment of the present invention may include a communication unit 110, a memory 120, and a processor 130.

The communication unit 110 may be a wired communication device or a wireless communication device for transmitting anti-cancer treatment data, artificial intelligence models, and data necessary for operation of the neutrophil count recovery prediction device to the neutrophil count recovery prediction device 100. When the communication unit 110 uses a Wi-Fi chip or a Bluetooth chip, various connection information such as SSID and session key are first transmitted and received, and various information can be transmitted and received after establishing a communication connection using this. A wireless communication chip refers to a chip that performs communication according to various communication standards such as IEEE, Zigbee, 3G (3rd Generation), 3GPP (3rd Generation Partnership Project), and LTE (Long Term Evolution). An NFC chip refers to a chip that operates in the NFC (Near Field Communication) method using the 13.56MHz band among various RF-ID frequency bands such as 135kHz, 13.56MHz, 433MHz, 860~960MHz, 2.45GHz, etc.

The memory 120 may store anti-cancer treatment data, artificial intelligence models, and data necessary for driving a neutrophil count recovery prediction device. The memory 120 may further store various data for the overall operation of the neutrophil count recovery prediction device 100, such as a program for processing or controlling the processor 130. As an example, the OS, data, and commands for driving the neutrophil count recovery prediction device 100 may be stored. At least some of these data and applications may be downloaded from an external server through the communication unit.

The memory 120 may be implemented as internal memory such as ROM or RAM included in the processor 130, or may be implemented as a device separate from the processor 130.

The processor 130 is configured to generally control the neutrophil count recovery prediction device 100. Specifically, the processor 130 controls the overall operation of the neutrophil count recovery prediction device 100 using various programs stored in the memory 120 of the neutrophil count recovery prediction device 100. For example, the processor 130 may include a CPU, RAM, ROM, and a system bus. According to an embodiment of the present invention, the processor 130 may be implemented as a digital signal processor (DSP), a microprocessor, or a time controller (TCON) that processes digital signals. However, It is not limited to, but is not limited to, a central processing unit (CPU), a micro controller unit (MCU), a micro processing unit (MPU), a controller, an application processor (AP), or a communication processor ( It may include one or more of a communication processor (CP), an ARM processor, or be defined by the corresponding term. In addition, the processor 130 may be a system on chip (SoC) with a built-in processing algorithm, and a large scale integration (LSI). ), or it may be implemented in the form of an FPGA (Field Programmable Gate Array).

The processor 130 may be configured to pre-process the anti-cancer treatment data and use a first artificial intelligence model to calculate the length of the risk period during which the absolute neutrophil count after anti-cancer treatment is less than or equal to a preset reference value. In one embodiment, the processor 130 may be configured to predict the change pattern of the absolute neutrophil count during a preset period from the start date of the anticancer cycle using a second artificial intelligence model and return the result as a mistake. In one embodiment, the processor 130 may be configured to determine whether a preset period from the start date of the anticancer cycle is included in the risk period using a second artificial intelligence model.

The neutrophil count recovery prediction device 100 may acquire anticancer treatment data from the external device 200, analyze the data, and transmit the derived result to the external device 200. Here, the external device 200 may be a personal computer, a personal terminal, and a medical data storage of a hospital or research institute, but is not limited thereto.

Figure 11 is a conceptual diagram for explaining a first artificial intelligence model according to an embodiment of the present invention.

Referring to FIG. 11, the first artificial intelligence model may include a RNN algorithm. In one embodiment, the first artificial intelligence model may include the GRU-D algorithm to use medical data with irregular input intervals as input values.

Therefore, the first artificial intelligence model does not perform the process of re-sampling the input data with irregular input intervals by dividing them at regular intervals through a separate preprocessing process, but updates the interpolation function in the first artificial intelligence model, By calculating interpolation values, more accurate results can be predicted without damaging the original data.

Figure 12 is a conceptual diagram for explaining a second artificial intelligence model according to an embodiment of the present invention.

Referring to FIG. 12, the second artificial intelligence model may include a transformer algorithm to utilize medical data having irregular length.

In one embodiment, the second artificial intelligence model may include an attention algorithm or a combination of a transformer algorithm and an attention algorithm. However, the present invention is not limited to this, and the second artificial intelligence model may include the latest known deep learning algorithm suitable for analyzing data with irregular length without limitation.

The second artificial intelligence model assigns weight to specific parts of the input data and can predict more accurate results even if the length of the input data is variable. As an example, the second artificial intelligence model may include time information in input data through a time-embedding process. Known techniques can be applied to the time-embedding process without particular limitations. As an example, when performing time-embedding using a D-dimensional vector, for time t, sequentially for each dimension, or can be applied. At this time, may be set to optimal values for variables given as learnable parameters.

As described above, although the embodiments have been described using limited examples and drawings, various modifications and variations can be made by those skilled in the art from the above description. For example, the described techniques are performed in a different order than the described method, and/or components of the described system, structure, device, circuit, etc. are combined or combined in a different form than the described method, or other components are used. Alternatively, appropriate results may be achieved even if substituted or substituted by an equivalent.

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

100: Method for predicting neutrophil count recovery
110: Communication unit 120: Memory
130: processor 200: external device

Claims (13)

In a method for predicting neutrophil count recovery performed on a computing device,
A data acquisition step of acquiring anticancer treatment data of the target patient;
A data preprocessing step of preprocessing the anticancer treatment data;
A first artificial intelligence model calculating a relative date, which is a period of time elapsed based on the start date of the anticancer cycle;
Calculating, by the first artificial intelligence model, interpolation data of the anti-cancer treatment data according to a preset date interval based on the relative date; and
Comprising: calculating, by the first artificial intelligence model, the length of a risk period in which the absolute neutrophil count (ANC) after anticancer treatment is below a preset reference value based on the anticancer treatment data and interpolation data; Method for predicting neutrophil count recovery. delete According to paragraph 1,
The step of calculating interpolation data is,
Setting, by the first artificial intelligence model, standard values for each variable included in the anticancer treatment data;
Setting an interpolation function in the first artificial intelligence model including a first parameter; and
Comprising: calculating interpolation values at preset date intervals using the standard value and the observed value included in the anticancer treatment data,
The first parameter is a method for predicting neutrophil count recovery, wherein the length of the risk period predicted by the first artificial intelligence model is compared with the correct answer values, and the compared result is repeatedly learned and updated so that it exceeds the learning standard. According to paragraph 1,
The first artificial intelligence model is a neutrophil count recovery prediction method including the GRU-D algorithm. In a method for predicting neutrophil count recovery performed on a computing device,
A data acquisition step of acquiring anticancer treatment data of the target patient;
A data preprocessing step of preprocessing the anticancer treatment data;
Predicting the change pattern of the absolute neutrophil count during a preset period from the start date of the anticancer cycle using a second artificial intelligence model and making a mistake decision; and
A method for predicting neutrophil count recovery, comprising: using the second artificial intelligence model to determine whether the absolute neutrophil count is included in a risk period below a preset reference value during a preset period from the start date of the anticancer cycle. According to clause 5,
The step of determining whether it is included in the risk period is,
generating, by the second artificial intelligence model, a first derived variable based on the absolute neutrophil count;
The second artificial intelligence model determines that the first derived variable is included in the risk period when it is less than a preset reference value, and does not include the risk period when the first derived variable is more than a preset reference value. A method for predicting neutrophil count recovery, including the step of determining. According to clause 5,
The second artificial intelligence model is a method for predicting neutrophil count recovery, including a Transformer algorithm, an Attention algorithm, or a combination thereof. In a method for predicting neutrophil count recovery performed on a computing device,
A data acquisition step of acquiring anticancer treatment data of a plurality of patients;
A data preprocessing step of preprocessing the anticancer treatment data; and
Comprising: learning an artificial intelligence model using the preprocessed anticancer treatment data,
The data preprocessing step is,
de-identifying the anti-cancer treatment data by deleting personal information;
Classifying the de-identified anti-cancer treatment data into general anti-cancer treatment and high-dose anti-cancer treatment;
displaying anticancer cycles of the anticancer treatment data;
Entering a relative date, which is a period of time elapsed from the start date of the anticancer cycle, into the anticancer treatment data;
Calculating the missing absolute neutrophil count (ANC) and absolute lymphocyte count (ALC) among the anticancer treatment data from other variables; and
Method for predicting neutrophil count recovery, comprising: removing variables having a number less than a preset number. delete A computer program stored in a computer-readable storage medium for executing the method for predicting neutrophil count recovery according to any one of claims 1 to 8 using a computer. Memory for storing anti-cancer treatment data and a first artificial intelligence model; and
Preprocess the anticancer treatment data,
The first artificial intelligence model calculates a relative date, which is a period of time elapsed based on the anticancer cycle start date, calculates interpolation data of the anticancer treatment data according to a preset date interval based on the relative date, and calculates the anticancer treatment data and a processor configured to calculate the length of a risk period in which the absolute neutrophil count is less than or equal to a preset reference value based on the interpolated data. According to clause 11,
The memory further stores a second artificial intelligence model,
The processor,
A neutrophil count recovery prediction device configured to predict the change pattern of the absolute neutrophil count during a preset period from the start date of the anticancer cycle using the second artificial intelligence model and return the change in error. According to clause 11,
The memory further stores a second artificial intelligence model,
The processor,
A neutrophil count recovery prediction device configured to determine whether the neutrophil count recovery period is included in the risk period during a preset period from the start date of the anticancer cycle using the second artificial intelligence model.

Images