KR20220109527A - Method and system for predicting adenoma related information based on machine-leaned model - Google Patents

Abstract

The present invention relates to a method and system for predicting adenoma-related information based on a machine-leaned model and, a method and system for predicting adenoma-related information based on a machine-leaned model, which can derive a risk of adenoma that may cause colon cancer by using a model trained by data on lifestyles of people acquired through health check results to be used by a user to predict an occurrence of colon cancer based on big data on the basis of the lifestyle information.

Description

Method and system for predicting adenoma related information based on machine-leaned model}

The present invention relates to a method and system for predicting adenoma-related information based on a machine learning model, and more particularly, to a method and system for predicting adenoma-related information that can cause colorectal cancer using a model learned by data on people's lifestyle obtained through health examination results. By deriving the possibility of occurrence of adenoma, the user relates to a method and system for predicting adenoma-related information based on a machine learning model, which can be used as an index to predict the onset of colorectal cancer based on big data based on lifestyle information.

Colorectal cancer refers to a malignant tumor that occurs in the colon and rectum, and the incidence of colorectal cancer is rapidly increasing worldwide in recent years. Among them, the incidence of colorectal cancer in Korea was 51.7 cases per 100,000 people in 2012, ranking third among all cancers. With the recent increase in the incidence of colorectal cancer in Korea, interest in early detection of colorectal cancer has increased, and colonoscopy and polypectomy are being actively performed. Adenomatous polyps occurring in the colon are known as precursor lesions of cancer, and by removing these polyps, it has been reported that the incidence of colorectal cancer can be reduced. The National Cancer Center currently performs more than 8,000 diagnostic and therapeutic colonoscopy every year, and this is expected to increase further due to the demand from patients in the future. The incidence rate of colorectal cancer is low until the age of 30, but the incidence rate starts to increase after the age of 40, and the incidence rate shows a large increase from the age of 50. In addition, the incidence rate is higher in men than in women, and in the case of colorectal cancer, compared to other cancers, the risk factors for cancer are relatively clear, so many studies focusing on the prevention of cancer have been preceded.

Based on this, there is a need for a technology that can provide an indicator for predicting colorectal cancer by deriving the possibility of colorectal polyps and colorectal cancer according to individual characteristics and lifestyle, but such prior art is nonexistent.

On the other hand, a polyp corresponds to a part of the mucous membrane protruding from the surrounding surface and protruding like a lump. It can occur in any organ with digestive tract or mucous membranes in our body, but in particular, the human organ where polyps occur the most is the large intestine. Most colon polyps do not show any specific symptoms, so it is difficult to detect them until colonoscopy is performed.

Adenoma refers to a neoplastic polyp that can develop into colorectal cancer among polyps. Most polyps are benign, but adenomas can progress to colorectal cancer over time, requiring special attention. More than 80% of colorectal cancer proceeds from adenoma, and if the adenoma itself is removed, the incidence of colorectal cancer can be lowered by that much, so it is necessary to predict adenoma.

However, since the discovery of such adenomas can be discovered through a procedure that requires a kind of anesthesia called colonoscopy to date, the discovery is delayed and metastasis to cancer is frequent.

The present invention derives the possibility of occurrence of an adenoma that can cause colorectal cancer using a model learned by data on people's lifestyle obtained through the health checkup result, so that the user can use the big data-based large-database based on lifestyle information. An object of the present invention is to provide a method and system for predicting adenoma-related information based on a machine learning model, which can be used as an index to predict the onset of cancer.

In order to solve the above problems, the present invention is a method for predicting adenoma-related information based on a machine learning model, performed by a server system, and lifestyle information for a plurality of first items for each individual collected from a medical institution, and a colonoscopy Using the first inference model statistically learned by the first learning data including the adenoma information including the information on the presence or absence of adenoma according to the result, the prediction of the lifestyle information for the plurality of first items for each individual is predicted. a second learning data deriving step of deriving Zen type information and deriving second learning data including lifestyle information and Zen type information for the plurality of first items for each individual; Generates first compressed learning data including lifestyle information and Zen type information for a second item in which some of the plurality of first items are removed from the first learning data, and in the second learning data, the plurality of first items A compressed data generating step of generating second compressed learning data including lifestyle information and Zen type information for a second item from which some of the first items are removed; and a second inference model for inferring the possibility of information on the presence or absence of an adenoma with respect to the lifestyle information for the inputted second item, including an artificial neural network, based on the first compressed data and the second compressed data. It provides a method for predicting ship type-related information based on a machine learning model, including; inference model learning step.

In some embodiments of the present invention, the second learning data derivation step includes adenoma information including lifestyle information on a plurality of first items for each individual collected from a medical institution, and information on the presence or absence of adenoma according to the results of colonoscopy. Learning information receiving step of receiving the first learning data including; The first learning data is clustered with adenomatous information according to the specific lifestyle information according to the first item, and based on the clustered information, the probability of adenomatosis for a group having the specific lifestyle information according to the first item is derived. possibility extraction stage; a predictive data set deriving step of deriving a predictive data set including the adenomatous probability of a group having the specific lifestyle information according to a plurality of the first items; and a prediction result derivation step of deriving a result on the possibility of adenomatosis of a person having the inputted lifestyle information based on the prediction data set with respect to the inputted lifestyle information for a plurality of first items for each individual; may include

In some embodiments of the present invention, in the step of deriving the prediction result, the degree of similarity is derived according to the degree of matching between the inputted lifestyle information for the plurality of first items for each individual and the lifestyle information included in the prediction data set. and deriving a plurality of similarity group prediction data by determining whether the degree of similarity meets a preset criterion; And by using a representative value of numerical information on the information on the presence or absence of adenoma in the plurality of similar group prediction data, the information on the presence or absence of adenoma of a person having lifestyle information for a plurality of first items for each individual inputted above It may include; deriving probability information about the information.

In some embodiments of the present invention, the information on the presence or absence of a line type of the first learning data is,

By extracting a keyword related to a polyp or adenoma from the colonoscopy result information and the biopsy result information collected from the medical institution, the first polyp data including at least one of the polyp location, polyp size, and polyp type for the individual is derived, and , by determining the information on the presence or absence of the adenoma from the first polyp data can be derived in the form of normalized information.

In some embodiments of the present invention, the first item and the second item include age, sex, BMI, exercise habit, drinking habit, smoking habit, family medical history and personal medical history, and the second item is the first Some of the items may be excluded.

In some embodiments of the present invention, in the second learning data derivation step, the server system receives lifestyle information for a plurality of first items from the first user requesting his/her predicted line type information, and the received The first service provided to the first user is performed by deriving the predicted ship type information using the lifestyle information for the first item and the first inference model, thereby deriving the second learning data, and the machine learning model Adenoma related information prediction method based on receiving lifestyle information for a plurality of second items from a second user who requests his/her predicted adenomatous information, and lifestyle information for the received second item and the first It may further include; a second service providing step of deriving the predicted ship type information using the two inference model, and providing it to a second user.

According to an embodiment of the present invention, by receiving the lifestyle information and providing the possibility of occurrence of adenoma derived from the lifestyle information, it is possible to exhibit the effect of presenting the screening test criteria for colorectal cancer.

According to an embodiment of the present invention, it is learned based on not only lifestyle information, colonoscopy result information and biopsy result information of various learning subjects, but also additional learning data generated by an inference model learned in a statistical manner using it. By providing information on the possibility of occurrence of adenoma corresponding to colorectal cancer-related information based on the machine learning model, the user can exert the effect of confirming the health status of the colon as an objective index through the provided probability index.

According to an embodiment of the present invention, when there is no data similar to the lifestyle information for predicting adenoma occurrence related information in the learning process of the first inference model to learn statistically, the detailed category information of the lifestyle information is set as a preset criterion. By filtering similar data by excluding according to , it is possible to exert the effect of more efficiently learning the first learning model.

According to an embodiment of the present invention, even when some data is not received, by supplementing the data based on the learning information collected in the past, the effect of improving the data reliability of the colorectal cancer-related information derived from the data can perform

1 schematically illustrates the operation of a server system according to a method for predicting colorectal cancer-related information based on lifestyle information according to an embodiment of the present invention.
2 schematically shows an internal configuration of a server system according to an embodiment of the present invention.
3 schematically shows the form of learning information processed by the possibility derivation unit according to an embodiment of the present invention.
4 schematically shows a criterion for classifying lifestyle information according to an embodiment of the present invention.
5 schematically illustrates a processing process of information derived by a likelihood derivation unit according to an embodiment of the present invention.
6 schematically shows a preset mapping table that is based on deriving the colorectal cancer possibility of a learning target by the possibility derivation unit according to an embodiment of the present invention.
7 schematically shows the form of data derived by the likelihood derivation unit and the prediction dataset derivation unit according to an embodiment of the present invention.
8 schematically shows the execution steps of the prediction result deriving unit according to an embodiment of the present invention.
9 schematically illustrates a criterion for excluding detailed category items of lifestyle information by the prediction result deriving unit according to an embodiment of the present invention.
10 schematically illustrates a processing process of information derived by a prediction result deriving unit according to an embodiment of the present invention.
11 schematically shows a ship type-related information prediction system based on a machine learning model according to an embodiment of the present invention.
12 schematically illustrates a process of generating normalized lifestyle information according to an embodiment of the present invention.
13 schematically illustrates a process of securing the first learning data and the second learning data according to an embodiment of the present invention.
14 schematically illustrates steps of a user using a first service according to an embodiment of the present invention.
15 schematically shows a compression process of the first learning data and the second learning data according to an embodiment of the present invention.
16 schematically illustrates a process of learning a second inference model using the first compressed learning data and the second compressed learning data according to an embodiment of the present invention.
17 schematically illustrates steps of providing a second service according to an embodiment of the present invention.
18 schematically shows processes for generating third compressed learning data according to an embodiment of the present invention.
19 schematically shows processes for generating fourth compressed learning data according to an embodiment of the present invention.
20 schematically illustrates a process of learning a third inference model according to an embodiment of the present invention.
21 schematically shows steps for providing high-risk vessel type information by the second service provider according to an embodiment of the present invention.
22 illustrates adenoma occurrence prediction information UI displayed on the user terminal by the operation of the second service providing unit according to an embodiment of the present invention.
23 illustrates a high-risk adenoma occurrence prediction information UI displayed on the user terminal by the operation of the second service provider according to an embodiment of the present invention.
24 shows data on the accuracy of predicting the occurrence of adenoma and predicting the occurrence of high-risk adenoma according to embodiments of the present invention.
25 schematically illustrates the configuration of a computing device according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily carry out the present invention. However, the present invention may be embodied in several different forms and is not limited to the embodiments described herein. And in order to clearly explain the present invention in the drawings, parts irrelevant to the description are omitted, and similar reference numerals are attached to similar parts throughout the specification.

Throughout the specification, when a part is "connected" with another part, this includes not only the case of being "directly connected" but also the case of being "electrically connected" with another element interposed therebetween. . Also, when a part "includes" a certain component, it means that other components may be further included, rather than excluding other components, unless otherwise stated.

Also, terms including an ordinal number such as first, second, etc. may be used to describe various elements, but the elements are not limited by the terms. The above terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. and/or includes a combination of a plurality of related listed items or any of a plurality of related listed items.

In this specification, a "part" includes a unit realized by hardware, a unit realized by software, and a unit realized using both. In addition, one unit may be implemented using two or more hardware, and two or more units may be implemented by one hardware. Meanwhile, '~ unit' is not limited to software or hardware, and '~ unit' may be configured to be in an addressable storage medium or may be configured to reproduce one or more processors. Thus, as an example, '~' denotes components such as software components, object-oriented software components, class components, and task components, and processes, functions, properties, and procedures. , subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays and variables. The functions provided in the components and '~ units' may be combined into a smaller number of components and '~ units' or further separated into additional components and '~ units'. In addition, components and '~ units' may be implemented to play one or more CPUs in a device or secure multimedia card.

The "user terminal" referred to below may be implemented as a computer or portable terminal that can access a server or other terminal through a network. Here, the computer includes, for example, a laptop equipped with a web browser (WEB Browser), a desktop, a laptop, and the like, and the portable terminal is, for example, a wireless communication device that guarantees portability and mobility. , PCS (Personal Communication System), GSM (Global System for Mobile communications), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code) Division Multiple Access)-2000, W-Code Division Multiple Access (W-CDMA), and Wibro (Wireless Broadband Internet) terminals may include all types of handheld-based wireless communication devices. In addition, "network" refers to a wired network such as a local area network (LAN), a wide area network (WAN) or a value added network (VAN), or a mobile radio communication network or satellite. It may be implemented in any kind of wireless network, such as a communication network.

1 schematically illustrates an operation of a server system according to a method for predicting colorectal cancer-related information based on lifestyle information according to an embodiment of the present invention.

In the server system 1000 shown in FIG. 1, by learning data about people's lifestyles obtained through health checkup results, and deriving and providing colorectal cancer-related information such as the possibility of colon polyps and colorectal cancer according to lifestyles. , the user can use the colorectal cancer-related information for a specific lifestyle as an index that can predict the onset of colorectal cancer according to the corresponding lifestyle. Preferably, the lifestyle information including a plurality of detailed category information includes one or more of age, gender, BMI, exercise habit, drinking habit, smoking habit, family medical history, and personal medical history. Lifestyle information receives health checkup information such as health checkup questions and answers and derives it from the received health checkup information based on keywords related to the items of the detailed category information, or intuitively inputs information for each item of the detailed category information can receive The server system 1000 of the present invention receives the learning information 1510 including such lifestyle information and colonoscopy result information and biopsy result information and learns the learning information 1510 to derive the polyp possibility and the colorectal cancer possibility. and deduce the possibility of polyp and colorectal cancer for specific lifestyle information. Thereafter, colorectal cancer-related result information 1530 can be derived based on the derived polyp possibility and colorectal cancer possibility, and the derived colorectal cancer-related result of a person with the inputted lifestyle information based on the inputted lifestyle information Information 1530 may be derived.

The server system 1000 shown in FIG. 1 includes a processor for receiving the received learning information 1510 and deriving colorectal cancer-related information, and one or more memories for storing instructions executable by the processor. The processor includes a learning information receiving unit 1100 , a probability deriving unit 1200 , a prediction data set deriving unit 1300 , and a prediction result deriving unit 1400 .

Hereinafter, operations of the learning information receiving unit 1100 , the likelihood deriving unit 1200 , the prediction data set deriving unit 1300 , and the prediction result deriving unit 1400 will be described in detail.

2 schematically shows an internal configuration of a server system 1000 according to an embodiment of the present invention.

The server system 1000 of the present invention may perform a method of predicting colorectal cancer-related information based on lifestyle information. Briefly, by learning data on people's lifestyles obtained through health checkup results, information related to colorectal cancer such as the possibility of colon polyps and colorectal cancer according to lifestyles is derived and provided. The server system 1000 may include a learning information receiving unit 1100 , a likelihood deriving unit 1200 , a prediction data set deriving unit 1300 , and a prediction result deriving unit 1400 .

The learning information receiving unit 1100 receives lifestyle information, colonoscopy result information, and biopsy result information for a plurality of items of the learning target. Specifically, the lifestyle information of the learning target may be received from information including information on the lifestyle of the learning target, such as, for example, health checkup information. Preferably, the lifestyle information may include one or more of age, gender, BMI, exercise habit, drinking habit, smoking habit, family medical history, and personal medical history. The colonoscopy result information means, for example, information about a colonoscopy result of a learning target who has performed a diagnosis through a colonoscopy, such as a colonoscopy diagnosis certificate, and has a diagnosis result. The biopsy result information, for example, like a biopsy result report, means information about the biopsy result of a learning object that performs a biopsy and holds the biopsy result.

The possibility derivation unit 1200 clusters colonoscopy result information and biopsy result information of one or more learning subjects having specific lifestyle information, and based on the clustered information, polyps for a group having specific lifestyle information likelihood, and colorectal cancer. Specifically, first, colonoscopy result information and biopsy result information of one or more learning objects having the same specific lifestyle information among a plurality of learning objects are clustered. Thereafter, based on the clustered information, it is determined whether the learning target has polyps, and based on whether a plurality of learning objects having the same specific lifestyle information have polyps or not, the polyp possibility for a group with specific lifestyle information is derived. . In addition, based on the clustered information and a preset mapping table, the possibility of colorectal cancer for a group having specific lifestyle information is derived. Possibility derivation unit 1200, by repeating the above-described process for a plurality of specific lifestyle information, derives the possibility of polyp and colorectal cancer for each group having each specific lifestyle information of the entire learning object.

The prediction dataset derivation unit 1300 derives a prediction dataset 1520 for the polyp possibility and colorectal cancer possibility of a group having a plurality of specific lifestyle information. After the possibility of polyp and colorectal cancer for each group having specific lifestyle information is derived by the likelihood derivation unit 1200, the prediction dataset derivation unit 1300 includes the lifestyle information including a plurality of detailed category information, A prediction data set 1520 including the polyp possibility and the colorectal cancer possibility is derived. Preferably, the lifestyle information includes at least one of age, gender, BMI, exercise habit, drinking habit, smoking habit, family medical history, and personal medical history.

Thereafter, the prediction result deriving unit 1400, with respect to the input lifestyle information, based on the prediction data set 1520 derived by the prediction data set deriving unit 1300, the polyp of a person with the corresponding input lifestyle information It derives the results for the possibility and the possibility of colorectal cancer. Specifically, the prediction result deriving unit 1400 matches the received input lifestyle information and the lifestyle information included in the prediction data set 1520, and the detailed category information of each lifestyle information matches the degree of matching. Accordingly, the possibility of polyp and colorectal cancer of a person with input lifestyle information can be derived based on the detailed prediction dataset with a high degree of matching.

On the other hand, in the DB 1500 of the server system 1000, learning information 1510 including lifestyle information, colonoscopy result information and biopsy result information for a plurality of items of the learning target, age, gender, BMI, Lifestyle information including at least one of exercise habits, drinking habits, smoking habits, family history, and personal medical history, a prediction data set 1520 including the polyp possibility and colorectal cancer possibility, polyps of a person with input lifestyle information Colorectal cancer-related result information 1530 including the possibility and the possibility of colorectal cancer may be stored.

The server system 1000 shown in FIG. 2 may further include elements other than the elements shown, but for convenience, only the elements related to the method for predicting colorectal cancer-related information based on lifestyle information according to embodiments of the present invention. was displayed.

3 schematically shows the form of learning information 1510 processed by the possibility derivation unit 1200 according to an embodiment of the present invention.

The learning information 1510 received by the learning information receiving unit 1100 included in the server system 1000 of the present invention includes lifestyle information, colonoscopy result information, and biopsy result information for a plurality of items of the learning target. . Thereafter, the received learning information 1510 may be processed by the possibility derivation unit 1200 . Specifically, Fig. 3 (a) schematically shows lifestyle information, Fig. 3 (b) shows colonoscopy result information, and Fig. 3 (c) schematically shows biopsy result information. Life habit information is, for example, when health check-up information such as a health check-up questionnaire of the learning target as shown in FIG. Lifestyle information included in the examination questionnaire image can be derived. Alternatively, lifestyle information for a plurality of detailed category information that has already been pre-processed externally through a health checkup questionnaire may be directly received. On the other hand, the colonoscopy result information means, for example, information about the colonoscopy result of a learning target who has performed a diagnosis through a colonoscopy, such as a colonoscopy diagnosis certificate, and has the diagnosis result. The colonoscopy result information is clustered by the possibility derivation unit 1200, and as shown in FIG. can be derived In addition, the biopsy result information means information about the biopsy result of a learning target, for example, by performing a biopsy, such as a biopsy result report, and having the test result for the tissue in the large intestine. Biopsy result information In addition, as shown in (c) of FIG. 3, the polyp type (adenoma in FIG. 3) can be derived from the biopsy result information by the possibility derivation unit 1200. As such, the possibility derivation unit 1200, when the learning information 1510 is received, extracts keywords related to colon polyps from the learning information 1510, and includes the polyp size, polyp position, and polyp type of the learning target. The step of deriving polyp data is performed.

4 schematically shows a criterion for classifying lifestyle information according to an embodiment of the present invention.

The server system 1000 of the present invention may classify the received lifestyle information according to preset criteria of each detailed category information. The criteria by which the lifestyle information is classified may be classified as shown in FIG. 4 . Lifestyle information including at least one of age, gender, BMI, exercise habit, drinking habit, smoking habit, family medical history, and personal medical history is classified according to a preset criterion as shown in FIG. 4 according to each detailed category. An asterisk can be assigned and recognized. For example, in the case of gender among detailed categories, a gender item index of 00 may be assigned to women, and a gender item index of 01 may be assigned to men. In addition, as shown in FIG. 4, for detailed category items in which the degree of individual habit, such as drinking habits or smoking habits, may vary greatly, as shown in FIG. 4, for example, smoking period * average daily smoking amount and The index for each item can be derived by converting the degree of each subcategory item into a numerical value based on the same standard and assigning the smoking item index depending on whether the numerical value meets the preset standard. Lifestyle information can be recognized according to the derived item-by-item index, and lifestyle information can be converted into one ID (for example, 30-01-00-00-10- 10-01-01) can also be derived. By classifying the lifestyle information in this way, it is possible to exhibit the effect of objectively classifying the information of a plurality of learning objects.

5 schematically shows a processing process of information derived by the likelihood derivation unit 1200 according to an embodiment of the present invention, and FIG. It schematically shows a preset mapping table as a basis for this.

The possibility derivation unit 1200 of the present invention extracts keywords related to colon polyps from the colonoscopy result information and the biopsy result information of the learning subject, and selects one of the polyp size, polyp location, and polyp type of the learning subject. Deriving polyp data including the above (S100); It is determined whether the polyp is possessed by the learning target according to one or more of the polyp size, polyp location, and polyp type, and the specific lifestyle information based on whether two or more learning objects having the same specific lifestyle information have the polyp. Deriving the polyp possibility for (S110); and deriving the colorectal cancer possibility for the specific lifestyle information according to information on the polyp size and polyp type of two or more learning objects having the same specific lifestyle information, and a preset mapping table; (S120) carry out

Specifically, in step S100, the possibility derivation unit 1200 extracts keywords related to colon polyps from the colonoscopy result information and the biopsy result information of the learning target, and selects one or more of the polyp size, polyp location, and polyp type of the learning target. It is possible to derive polyp data including As described above in the description of FIG. 3 , the possibility derivation unit 1200 derives polyp data including at least one of a polyp position, a polyp size, and a polyp type. Figure 5 (a) shows the learning information 1510 of one or more learning objects having specific lifestyle information from which polyp data was derived by the possibility derivation unit 1200 . According to the criteria of FIG. 4, lifestyle information including a plurality of detailed category information is displayed as an index by item, and the polyp location, polyp size, and polyp type according to the colonoscopy result information and biopsy result information of each learning target are shown. has been In the case of learning information #1 and learning information #4, it indicates that keywords related to colon polyps were not extracted from the colonoscopy result and biopsy result information of the subject, which means that no polyps were found in the large intestine of the subject. means it wasn't

Thereafter, in step S110, the possibility derivation unit 1200 determines whether the polyp possesses the learning target according to one or more of the polyp position, polyp size, and polyp type. According to the bar shown in Figure 5 (b), in the case of learning object #1 and learning object #4, colon polyps in the colonoscopy result and biopsy result information in the learning information 1510 of the learning object as described above. It is shown that polyp data was not derived based on that keywords related to the were not extracted. In this way, it is possible to determine whether the polyp possessing the learning target according to whether the polyp data including at least one of the polyp location, polyp size, and polyp type is derived. If one or more of the polyp location, polyp size, and polyp type is derived, it is determined that the polyp is possessed. to determine

After it is determined whether polyps possessing a plurality of learning objects having the same specific lifestyle information, the possibility derivation unit 1200 is based on the polyps possession of two or more learning objects having the specific lifestyle information. Polyp possibility can be derived for habit information. In one embodiment of the present invention, as shown in FIG. 5(b), lifestyle information indicating age-sex-BMI-exercise habit-drinking habit-smoking habit-family medical history-personal medical history is 30-01- As 00-00-10-10-01-01, five learning subjects with the same specific life information are shown. Among them, it is shown that two of the learning objects #1 and #4 do not have polyps, and the three learning objects #2, #3, and #5 have polyps. Based on this, the polyp possibility can be derived according to the following derivation equation 1.

[derivative formula 1]

Number of learning subjects with specific lifestyle information with polyps / Total number of learning subjects with specific lifestyle information *100 (%)

When an example shown in (b) of FIG. 5 is applied to Equation 1, a polyp probability of Equation 1 = 3/5 * 100 = 60% is derived. Therefore, the polyp probability of the learning target having the lifestyle information 30-01-00-00-10-10-01-01 is derived as 60% as shown in (c) of FIG. 5 .

In this way, the possibility derivation unit 1200 may derive the polyp possibility for specific lifestyle information.

Thereafter, the step of deriving the colorectal cancer possibility for the specific lifestyle information performed by the likelihood derivation unit 1200 (S120) is information on the polyp size and polyp type of two or more learning objects having specific lifestyle information. and deriving the colorectal cancer possibility for each learning target according to a preset mapping table (S120-1); and deriving the colorectal cancer possibility for the specific lifestyle information based on the colorectal cancer possibility for each learning target; (S120-2).

Specifically, in step S120-1, the likelihood derivation unit 1200 derives the colorectal cancer possibility for each learning information 1510 having specific lifestyle information through a preset mapping table as shown in FIG. 6 . As shown in FIGS. 5 and 6 , the mapping table serves as a criterion for deriving the possibility of colorectal cancer according to the polyp size and polyp type. According to FIG. 5 (b), the polyp size of learning information #2 is 4, and the polyp type is shown as villous adenoma. Based on this, in the criteria of the mapping table of FIG. 6, the polyp type is chorionic adenoma, and since it corresponds to a polyp size of 3 to 6 cm or less, the possibility of colorectal cancer is derived as 25% as shown in FIG. can Similarly, in the case of the learning information #3 and the learning information 1510 #5, it is shown that the colorectal cancer possibility was derived for each learning information 1510 according to the mapping table of FIG. 6 . In this way, the likelihood derivation unit 1200 derives the colorectal cancer possibility for each learning object according to the information on the polyp size and the polyp type of two or more learning objects having specific lifestyle information and a preset mapping table.

Then, in step S120-2, the colorectal cancer possibility is derived for the specific lifestyle information based on the colorectal cancer possibility for each learning target. The possibility derivation unit 1200 may derive the colorectal cancer possibility with respect to the specific lifestyle information according to Equation 2 below based on the colorectal cancer possibility for each of two or more learning objects having specific lifestyle information.

[derivative formula 2]

In Equation 2, N means the total number of learning objects having specific lifestyle information.

When the example shown in (b) of FIG. 5 is applied to Equation 2, the possibility of colorectal cancer for specific life information of Equation 2 = ((25 + 80 + 2)/5) = 21.4% is derived. Therefore, the colorectal cancer possibility for the lifestyle information 30-01-00-00-10-10-01-01 is derived as 21.4% as shown in FIG. 5(c).

In this way, the likelihood derivation unit 1200 may derive the colorectal cancer possibility for specific lifestyle information.

7 schematically shows the form of data derived by the likelihood derivation unit 1200 and the prediction data set derivation unit 1300 according to an embodiment of the present invention.

As described above, the possibility derivation unit 1200 derives polyp data including at least one of a polyp location, a polyp size, and a polyp type of a learning target based on lifestyle information, colonoscopy result information, and biopsy result information, The possibility of polyp and colorectal cancer are derived for specific lifestyle information. 7 (a) shows an example form of the learning information 1510 including polyp data including at least one of the polyp position, polyp size, and polyp type by the possibility derivation unit 1200 . If there is a value for the polyp data, the learning target with the corresponding learning information 1510 has a polyp, and if there is no value for the polyp data, the learning target with the corresponding learning information 1510 does not have a polyp. can be identified as

On the other hand, the possibility derivation unit 1200, when there is no tissue examination result information in the learning information 1510 of the learning target, the learning information received in the learning information receiving unit 1100 of the learning information 1510 collected in the past Based on the collected learning information 1510 in the past that the lifestyle information and the lifestyle information of the learning target are the same, the colonoscopy result information is similar within a preset standard, and the biopsy result information exists, A colorectal cancer possibility is derived for the learning information 1510 received from the learning information receiving unit 1100 . Specifically, as shown in (a) of FIG. 7 , in the learning information 1510 #1, values for the location and size of the polyp exist, but values for the type of the polyp do not exist. This means that the biopsy result information was not received. As such, in one embodiment of the present invention, there may be a case where there is no tissue examination result information in the learning information 1510 of the learning object received by the server system 1000 . In such a case, the possibility derivation unit 1200 is the learning information 1510 having the same lifestyle information as the learning information 1510 of the learning target without the received biopsy result information among the learning information 1510 collected in the past. to filter Thereafter, by comparing the polyp location and polyp size of one or more learning information 1510 having the same lifestyle information with the polyp location and polyp size included in the received learning information 1510, similar learning information 1510 within a preset standard. ) to filter out. According to (a) and (b) of Figure 7, a plurality of learning objects having the same lifestyle information 40-00-00-00-10-10-01-01 as the learning information #1 without a value for the polyp type Learning information 1510 is shown, and when the polyp position and polyp size of each learning information 1510 are compared, the polyp position and polyp size of learning information #1 and learning information #5 are the same as R and 4. is shown. Accordingly, the possibility derivation unit 1200 derives the polyp type of the learning information #1 as the same polyp type Type4 as the learning information #5. Preferably, among the learning information 1510 of one or more learning objects having the same lifestyle information collected in the past, if the polyp location and the polyp size derived from the colonoscopy result information are the same learning information 1510, When the polyp type of the corresponding learning information 1510 is derived as the polyp type of the learning information 1510 received from the learning information receiving unit 1100, and the learning information 1510 having the same polyp location and polyp size does not exist, The polyp location is the same, and the polyp size can be derived as the polyp type of the learning information 1510 received from the learning information receiving unit 1100 that has received the polyp type of the learning information 1510 that is within a preset range.

In this way, the possibility derivation unit 1200 derives all the information about the biopsy result information based on the learning information 1510 collected in the past when there is no biopsy result information, Then, through the method described in FIGS. 5 and 6, the possibility of colon cancer is derived for the learning information 1510 received in the step of receiving the learning information 1510, and the possibility of colorectal cancer is derived for the corresponding specific lifestyle information. .

In this way, the possibility derivation unit 1200 supplements the learning information 1510 based on some learning information 1510 even when all of the learning information 1510 has not been received, thereby improving data reliability. can perform

On the other hand, the prediction dataset deriving unit 1300, the polyp possibility of a group having a plurality of the specific lifestyle information based on the polyp possibility and colorectal cancer possibility for the specific lifestyle information derived by the possibility deriving unit 1200 and a prediction data set 1520 for the possibility of colorectal cancer. The prediction dataset 1520 includes lifestyle information including a plurality of detailed categories, the polyp possibility, and the colorectal cancer possibility, and the lifestyle information is, age, gender, BMI, exercise habits, drinking habits, smoking Including habits, family history, and personal medical history. As shown in (c) of FIG. 7 , the polyp potential and colorectal cancer possibility for specific lifestyle information derived by the likelihood derivation unit 1200 are lifestyle information 40-00-00-00-10-10-01- The polyp probability A1 for 01, and the colorectal cancer possibility B1 are derived, and the prediction dataset derivation unit 1300 converts the entire learning information 1510 into a plurality of the specific lifestyle habits as shown in FIG. A prediction dataset 1520 including lifestyle information for each group with information, polyp potential for lifestyle information, and colorectal cancer potential is derived. Based on the prediction data set 1520 derived in this way, the prediction result deriving unit 1400 to be described later may predict the polyp possibility and the colorectal cancer possibility with respect to the input lifestyle information.

8 schematically shows the execution steps of the prediction result deriving unit 1400 according to an embodiment of the present invention, and FIG. 9 is lifestyle information by the prediction result deriving unit 1400 according to an embodiment of the present invention. It schematically shows the criteria for excluding detailed category items of

The prediction result deriving unit 1400 of the present invention, with respect to the input lifestyle information, based on the prediction data set 1520 derived from the prediction data set deriving unit 1300, a person who has the corresponding input lifestyle information of polyp and colorectal cancer can be derived. Specifically, the prediction result deriving unit 1400 derives a degree of similarity according to the matching degree of the input lifestyle information and the lifestyle information included in the prediction data set 1520, and the degree of similarity meets a preset criterion Deriving similar group prediction data by determining whether or not (S1000 to S1300); and deriving the polyp possibility and colorectal cancer possibility of a person with the input lifestyle information based on the similar group prediction data (S1400).

In step S1000, a degree of similarity is derived according to the degree of matching between the input lifestyle information and the lifestyle information included in the prediction dataset 1520 . By matching the input lifestyle information input with the lifestyle information of the prediction dataset 1520 derived in the step of deriving the prediction dataset 1520, it is determined whether matching according to the detailed category of the lifestyle information is matched, and depending on whether the matching similarity can be derived. Age, gender, BMI, exercise habit, drinking habit, smoking habit, family history, and personal medical history included in the lifestyle information are determined whether each category is matched, and the degree of similarity is derived according to the matching. The equation for deriving the similarity is shown in Derived Equation 3 below.

[derivative formula 3]

(Number of detailed category items of lifestyle information matched / Total number of detailed category items)*100

An example in which the prediction result deriving unit 1400 derives the degree of similarity is as follows.

[Example 1]

Input lifestyle information: 30-01-00-00-10-10-01-01

Detailed Prediction Dataset Lifestyle Information: 50-00-01-00-12-10-01-01

Matching :X-X-X-O-X-O-O-O

If the above example is applied to Equation 3, a similarity of (4/8)*100 = 40% can be derived. In this way, the prediction result deriving unit 1400 derives the similarity between the input lifestyle information and the prediction data set 1520 .

Thereafter, in step S1100, it is determined whether the degree of similarity derived in step S1000 meets a preset criterion. For example, assuming that the degree of similarity derived from the preset criterion is 80% or more, in an embodiment of the present invention, since the number of detailed category items of lifestyle information is 8, the lifestyle information of 7 or more items is In the same case, a similarity of 80% or more can be derived. Accordingly, the prediction result deriving unit 1400 determines the detailed prediction dataset from which the similarity of 80% or more is derived among the prediction datasets 1520 . Thereafter, if there is a detailed prediction data set having a similarity that meets a preset criterion, step S1200 is performed.

Step S1200 is a step performed by the prediction result derivation unit 1400 when there is a detailed prediction data set in which the degree of similarity meets a preset criterion. One or more detailed prediction datasets that match , are derived as similar group prediction data.

On the other hand, step S1300 is a step performed by the prediction result derivation unit 1400 when there is no detailed prediction data set that meets the preset criteria for the degree of similarity. In step S1300, a plurality of detailed category information of the input lifestyle information part of it is excluded according to the preset criteria. As shown in FIG. 9, the degree of similarity derived according to the matching degree of items of eight detailed category information according to age, gender, BMI, exercise habit, drinking habit, smoking habit, family medical history, and personal medical history is based on a preset standard. If there is no matching detailed prediction data set, 6 detailed category information according to BMI, exercise habits, drinking habits, smoking habits, family and personal medical history, excluding age and gender, among multiple detailed category information of lifestyle information based on As such, the prediction result deriving unit 1400 may exclude a part of the plurality of detailed category information according to a preset criterion as shown in FIG. 9 . Preferably, the input lifestyle information includes at least one of age, sex, BMI, exercise habit, drinking habit, smoking habit, family medical history, and personal medical history, and the step of deriving it as the similar group prediction data includes age, , gender, BMI, exercise habits, drinking habits, smoking habits, family history, and personal medical history to determine whether the first degree of similarity meets a preset standard, and if the first similarity does not meet the preset standard, It is determined whether the second similarity for BMI, exercise habit, drinking habit, smoking habit, family medical history, and personal medical history meets a preset criterion, and if the second similarity does not meet the preset criterion, BMI, smoking It is determined whether the third similarity for habit, drinking habit, and personal medical history meets a preset criterion, and the similar group prediction data is derived.

After excluding some of the plurality of detailed category information matching the input lifestyle information and the prediction data set 1520 as described above, the prediction result deriving unit 1400 again performs the steps S1000 to S1200, so that the similarity is One or more detailed prediction datasets that meet the set criteria can be derived as similar group prediction data.

In step S1400, the prediction result derivation unit 1400, which derives the similar group prediction data through the execution of the steps S1000 to S1300, based on the similar group prediction data, the possibility of polyp and colorectal cancer of a person with input lifestyle information draw possibilities

In an embodiment of the present invention, when the input lifestyle information is input as shown in FIG. A detailed prediction data set like b) can be derived as similar group prediction data. It is shown that the lifestyle information of the detailed prediction data set shown in FIG. 10(b) includes the same 20-00-00-00-10-10-01-01 as the input lifestyle information. After deriving such similar group prediction data, the prediction result deriving unit 1400 is based on the similar group prediction data for specific lifestyle information corresponding to 20-00-00-00-10-10-01-01. The possibility of polyp and colorectal cancer of a person who has input lifestyle information corresponding to 20-00-00-00-10-10-01-01 is derived. Accordingly, (c) of FIG. 10 shows that A1 and colorectal cancer possibility B2, which are polyp probability values of similar group prediction data for specific lifestyle information corresponding to 20-00-00-00-10-10-01-01, are 20 It is shown that the polyp probability value of a person who has input lifestyle information corresponding to -00-00-00-10-10-01-01 is derived as A1 and the colorectal cancer possibility value is derived as B2.

Although FIG. 10 (b) describes a case in which all items of detailed category information of lifestyle information match the input lifestyle information as described above, in an embodiment of the present invention, the input lifestyle information and prediction There may be a case where all items of detailed category information of the lifestyle information of the dataset 1520 do not match. 10(c) shows similar group prediction data derived according to a preset criterion when there is no detailed prediction data set in which all items of detailed category information match. As shown in (c) of FIG. 10, the similarity group prediction data that meets the preset criteria according to the degree of matching with 20-00-00-00-10-10-01-01 and 5 detailed prediction datasets this was derived In the case of detailed prediction dataset #1, the items of the remaining category information except for gender are the same, and in the case of detailed prediction dataset #2, the items of the remaining category information except for age are the same, and the remaining detailed prediction dataset #3, In the case of #4 and #5, the same items of category information except for drinking habit, smoking habit, and BMI, respectively, are shown. In this case, the prediction result deriving unit 1400 is 20-00-00-00 based on the similar group prediction data for the specific lifestyle information corresponding to 20-00-00-00-10-10-01-01. -10-10-01-01 The possibility of polyp and colorectal cancer of a person with input lifestyle information is derived. Accordingly, FIG. 10(e) shows the polyp potential and colon of five detailed prediction data sets, which are similar group prediction data for specific lifestyle information corresponding to 20-00-00-00-10-10-01-01. An input life in which the average values of cancer likelihood values, A1+A2+A3+A4+A5/5 and B1+B2+B3+B4+B5/5, correspond to 20-00-00-00-10-10-01-01 It is shown that the polyp probability value of a person with habit information is A1+A2+A3+A4+A5/5 and the colorectal cancer probability value is derived as B1+B2+B3+B4+B5/5.

Preferably, the possibility of polyp and colorectal cancer of a person with input lifestyle information can be derived based on the following Equations 4 and 5.

[derivative formula 4]

[derivative formula 5]

Preferably, the step of deriving similar group prediction data performed by the prediction result deriving unit 1400 includes all items of a plurality of detailed category information of the input lifestyle information while the degree of similarity meets a preset criterion. When there is a matching detailed prediction data set, only the corresponding detailed prediction data set is derived as the similar group prediction data, and while the degree of similarity meets a preset criterion, the items of a plurality of detailed category information of the input lifestyle information are When there is no matching detailed prediction data set, one or more derived detailed prediction data sets are derived as the similar group prediction data.

In this way, the prediction result deriving unit 1400 can derive the polyp possibility and colorectal cancer possibility of a person having input lifestyle information based on the prediction data set 1520 derived by learning the learning information 1510 . have.

11 schematically shows a ship type-related information prediction system based on a machine learning model according to an embodiment of the present invention.

As shown in FIG. 11, the ship type-related information prediction system based on the machine learning model, performed by the server system, is a second learning data derivation unit 2100, a compressed data generation unit 2200, and a second inference model learning unit. 2300 , a third inference model learning unit 2400 , a service providing unit 2500 , and a DB.

The second learning data deriving unit 2100 includes a learning information receiving unit, a functional deriving unit, a prediction data set deriving unit, and a prediction result deriving unit, which have been described with reference to FIGS. 1 to 10 . Parts overlapping with those described with reference to FIGS. 1 to 10 with respect to these configurations will be omitted. However, in one embodiment of the system shown in FIG. 11, the learning information receiving unit, the functional deriving unit, the prediction data set deriving unit, and the prediction result deriving unit are operated based on the information on the presence or absence of the line type, and the prediction data set deriving unit is operated. A prediction result derivation unit that performs a prediction operation based on the prediction data set derived by , is included in the first inference model.

Specifically, the second learning data derivation unit 2100 is a first including adenoma information including lifestyle information for a plurality of first items for each individual collected from a medical institution, and information on the presence or absence of an adenoma according to a colonoscopy result. Using the first inference model statistically learned by the learning data, the predicted adenomatous information on the lifestyle information for the plurality of first items for each individual is derived, and the life for the plurality of first items for each individual is derived. A second learning data derivation step of deriving second learning data including habit information and Zen species information is performed.

That is, the second learning data derivation step uses the second inference model statistically learned with a prediction dataset derived based on the collected lifestyle information and Zen type information to predict the Zen type information for any lifestyle information, and , through which virtual learning data is extracted.

The lifestyle information on the plurality of first items for each individual used to derive the second learning data corresponds to the lifestyle information of users input through the service shown in FIG. 14 to be described later, or generates random numbers or existing The lifestyle information may correspond to lifestyle information artificially generated according to a variable based on random number generation.

In an embodiment of the present invention, not only the first learning data, but also the artificial neural network-based second inference model and the third inference model to be described later by the second learning data obtained by the statistically learned first inference model. The training data is expanded, and accordingly, the inference accuracy of the second inference model and the third inference model can be improved.

Compressed data generation unit 2200 generates first compressed learning data including lifestyle information and Zen type information for a second item in which some of the plurality of first items are removed from the first learning data, and 2 A compressed data generation step of generating second compressed learning data including lifestyle information on a second item in which some of the plurality of first items is removed from the second learning data, and Zen type information is performed.

Lifestyle information according to the first item in the first learning data and the second learning data may have a plurality of living information categories. In the present invention, in order to improve the learning accuracy and reduce the computational load, after compressing the learning data into the second item in which some items are deleted from the first item, the artificial neural network-based second inference model and the second inference model are compressed into the compressed learning data. 3 Train the inference model.

The second inference model learning step includes an artificial neural network based on the first compressed data and the second compressed data, and inferring the possibility of information on the presence or absence of an adenoma with respect to the inputted lifestyle information for the second item. Train an inference model.

Such a second inference model may output adenomatous information, for example, predictive information on the possibility of occurrence of adenoma with respect to the inputted lifestyle information according to the second item.

Hereinafter, the operation of the detailed components of the second learning data derivation unit 2100 will be described. Basically, the detailed components of the second learning data derivation unit 2100 perform an operation corresponding to the operation of the element with the same name of FIG. 2 described with reference to FIGS. 1 to 10 , but in the following Only the essential parts of the present invention will be described.

The learning information receiving unit receives the first learning data including lifestyle information for a plurality of first items for each individual collected from a medical institution, and adenoma information including information on the presence or absence of adenoma according to the result of colonoscopy. ;do. In the learning information receiving step, a normalization step described with reference to FIG. 12 may be performed.

The possibility derivation unit clusters the first learning data according to the specific lifestyle information according to the first item, and based on the clustered information, the adenomatous type for the group having the specific lifestyle information according to the first item. Take steps to derive possibilities.

The prediction dataset derivation unit derives a prediction dataset including the likelihood of adenoma of a group having the specific lifestyle information according to a plurality of the first items.

The prediction result deriving unit derives a result on the likelihood of adenomatosis of a person having the inputted lifestyle information based on the prediction data set with respect to the inputted lifestyle information for the plurality of first items for each individual.

The second learning data may be extracted by the operations of the learning information receiving unit, the likelihood deriving unit, the prediction data set deriving unit, and the prediction result deriving unit as described above.

On the other hand, since the same prediction data set does not exist for some inputted lifestyle information, it may be difficult to derive statistical prediction results for it. In order to supplement this, in a preferred embodiment of the present invention, the prediction result deriving unit determines the similarity according to the degree of matching between the inputted lifestyle information for the plurality of first items for each individual and the lifestyle information included in the prediction dataset. deriving a plurality of similarity group prediction data by determining whether the degree of similarity meets a preset criterion; And by using a representative value of numerical information on the information on the presence or absence of adenoma in the plurality of similar group prediction data, the information on the presence or absence of adenoma of a person having lifestyle information for a plurality of first items for each individual inputted above deriving probability information about the information; may be performed.

Detailed operations of the ship type-related information prediction system or method based on the above-described machine learning model will be described with reference to FIGS. 12 to 16 .

In another embodiment of the present invention, the server system shown in FIG. 11 may operate as a high-risk vessel type-related information prediction system based on a machine learning model, performed by the server system.

In this case, the second learning data derivation unit 2100 includes adenoma information including lifestyle information on a plurality of first items for each individual collected from medical institutions, and information on the presence or absence of adenoma according to the result of colonoscopy. 1 By using the first inference model statistically learned by the learning data, the predicted adenomatous information on the lifestyle information for the plurality of first items for each individual is derived, and the plurality of first items for each individual is derived. A second learning data derivation step of deriving second learning data including lifestyle information and Zen species information; is performed.

In addition, the compressed data generating unit 2200 generates first compressed learning data including lifestyle information and Zen type information for a second item in which some of the plurality of first items are removed from the first learning data, performing a compressed data generation step of generating second compressed learning data including lifestyle information for a second item in which some of the plurality of first items are removed from the second learning data, and Zen type information;

In addition, the second inference model learning unit 2300 includes an artificial neural network by the first compressed data and the second compressed data, and the possibility of information on the presence or absence of an adenoma with respect to the lifestyle information for the input second item. A second inference model that makes inferences is trained.

In addition, the third inference model learning unit 2400 expands to include lifestyle information for a plurality of second items for each individual collected from medical institutions, and information on the presence or absence of adenomas and information on the presence or absence of high-risk adenomas according to the results of colonoscopy. third compressed learning data including ship type information; and lifestyle information for a plurality of second items for each individual, lifestyle information for a plurality of second items for each individual, and information on the presence or absence of adenomas generated based on the second inference model. The third reasoning that trains the third inference model for inferring the possibility of information on the presence or absence of high-risk adenoma with respect to the lifestyle information for the input second item, including the artificial neural network, by the fourth compressed learning data including the information The model learning step is performed.

The detailed operation of the high-risk adenoma-related information prediction system or method based on the above-described machine learning model will be described with reference to FIGS. 12 to 16 .

12 schematically illustrates a process of generating normalized lifestyle information according to an embodiment of the present invention.

The process shown in FIG. 12 may be performed by the learning information receiver.

The information on the presence or absence of an adenoma in the first learning data is obtained by extracting a keyword related to a polyp or adenoma from the colonoscopy result information and the biopsy result information collected from the medical institution to extract the polyp location, polyp size, and polyp type for the individual. Deriving the first polyp data including one or more of, and determining the information on the presence or absence of the adenoma from the first polyp data can be derived in the form of normalized information.

The colonoscopy result information and the biopsy result information collected from the medical institution may correspond to EMR information of the medical institution. Such information corresponds to information that has not been pre-processed or normalized.

The learning information receiving unit converts the atypical lifestyle information collected by the medical institution into the lifestyle information according to the first item. For example, if there is EMR data in a different format at different medical institutions, it is converted into lifestyle information according to the first item, so that the learning data of the first inference model, the second inference model, and the third inference model, which will be described later can be used

On the other hand, if the information collected from the medical institution is the colonoscopy result information and the biopsy result information in the form of text, the learning information receiving unit extracts a keyword related to the polyp or adenoma, and the polyp location, polyp size and polyp for the individual. The first polyp data including one or more of the types is derived.

Information on the presence or absence of adenoma can be determined from such polyp data, and the information on the presence or absence of adenoma extracted in this way can serve as a kind of ground truth. In an embodiment of the present invention, the information on the presence or absence of adenoma may have any one of two values in the form of 1 or 0.

13 schematically illustrates a process of securing the first learning data and the second learning data according to an embodiment of the present invention.

In (A) of FIG. 13 , the second learning data extracting unit 2100 includes lifestyle information for a plurality of first items for each individual collected from a medical institution, and information on the presence or absence of adenoma according to the result of colonoscopy. The first inference model is statistically learned by the first learning data including information.

In an embodiment of the present invention, the prediction result deriving unit operates based on the prediction data set, and by continuously collecting the prediction data sets derived by the above-described prediction data set deriving unit, as a result, the first inference model is learned. can see.

As such, the amount of information used for learning the first inference model may be limited because it is information collected at an actual medical institution. Therefore, the first inference model can compensate for the lack of learning data to some extent by allowing the prediction result derivation unit to perform the complementary algorithm as in FIGS. 9 and 10 described above.

Meanwhile, the second inference model of the present invention corresponds to an artificial neural network-based deep learning model, and preferably, the second inference model and the third inference model may include an artificial neural network model based on logistic regression analysis.

In the case of such a second inference model, since it may be difficult to learn only with the first learning data collected at an actual medical institution, in the present invention, it is possible to expand the learning data by using the first statistically learned inference model. 2You can create training data.

As shown in (B) of FIG. 13 , in the embodiments of the present invention, predicted adenomatous information about lifestyle information for a plurality of first items for each individual is derived using a first inference model, and the arbitrary The second learning data including lifestyle information and Zen type information for the plurality of first items for each individual is derived.

That is, an arbitrary lifestyle information input value is input to the first inference model, and the output predicted selection information becomes a kind of label and can be secured as the second learning data. The predicted adenomatous information in FIG. 13B may be expressed, for example, as a numerical value between 0 and 1. Such a numerical value corresponds to a kind of probability value, which may be used for learning a second inference model to be described later.

14 schematically illustrates steps of a user using a first service according to an embodiment of the present invention.

The server system receives lifestyle information for a plurality of first items from a first user who requests his/her predicted line type information, and uses the received lifestyle information for the first item and the first inference model. By deriving the predicted ship type information, the first service provided to the first user is performed, so that the second learning data can be derived.

That is, the user of the first user terminal can check the prediction information about his/her own adenoma using the statistically learned first inference model, and the lifestyle information input by the user will become a part of the second learning data again. can

15 schematically shows a compression process of the first learning data and the second learning data according to an embodiment of the present invention.

The compressed data generating unit 2200 generates first compressed learning data including lifestyle information and Zen type information for a second item in which some of the plurality of first items are removed from the first learning data, Generates second compressed learning data including lifestyle information and Zen type information on a second item in which some of the plurality of first items are removed from the second learning data.

Preferably, the first item and the second item include age, gender, BMI, exercise habit, drinking habit, smoking habit, family medical history, and personal medical history, and the second item includes some of the first items excluded. More preferably, in the case of the second item, it may correspond to a form in which the exercise habit is excluded from the first item.

16 schematically illustrates a process of learning a second inference model using the first compressed learning data and the second compressed learning data according to an embodiment of the present invention.

The second inference model learning unit 2300 includes an artificial neural network based on the first compressed data and the second compressed data to infer the possibility of information on the presence or absence of an adenoma with respect to the lifestyle information for the input second item. Train the second inference model.

17 schematically illustrates steps of providing a second service according to an embodiment of the present invention.

The second service providing unit 2520 receives the lifestyle information for the plurality of second items from the second user who requests his/her predicted line type information, and receives the lifestyle information for the second item and A second service providing step of deriving predicted ship type information using the second inference model and providing it to a second user; is performed.

For example, when the user inputs his/her BMI, drinking habit, smoking habit, family history, and personal medical history through his/her terminal, the second service provider 2520 uses the learned second inference model to Information on the presence or absence of adenoma predicted based on a machine learning model can be provided.

18 schematically shows processes for generating third compressed learning data according to an embodiment of the present invention.

Hereinafter, a detailed technical configuration of a method and system for predicting high-risk adenoma-related information based on a machine learning model will be described. High-risk adenoma corresponds to an adenoma that can develop into colorectal cancer among adenomas, and in embodiments of the present invention, possibility or prediction information of high-risk adenoma is derived based on the inputted lifestyle information.

In order to predict such a high-risk adenoma, the present invention uses a third inference model based on an artificial neural network, and the third inference model is learned in a different way from the second inference model. In the system for predicting such high-risk adenomas, the first learning data is

As shown in FIG. 18, keywords related to polyps or adenomas are extracted from the colonoscopy result information and the biopsy result information collected from the medical institution, and one or more of the polyp location, polyp size, and polyp type for the individual is included. To derive the first polyp data, and to determine the information on the presence or absence of adenoma of the learning target from the first polyp data is derived in the form of normalized information.

On the other hand, the information on the presence or absence of high-risk adenoma of the third compressed learning data is derived in the form of normalized information by determining the information on the presence or absence of the high-risk adenoma from the first polyp data.

The third learning data, such as Oia, includes information on living habits according to item 1, information on the presence or absence of adenoma, and information on the presence or absence of high-risk adenoma, which is information corresponding to ground truth collected from medical institutions.

Similarly, the third compressed learning data that is compressed by the compressed data generating unit 2200 and finally used for learning the third inference model is lifestyle information for a plurality of second items for each individual collected from a medical institution, and information on the presence or absence of adenoma according to the results of colonoscopy and information on the presence or absence of high-risk adenoma.

19 schematically shows processes for generating fourth compressed learning data according to an embodiment of the present invention.

As described above, the third inference model is learned by the third compressed learning data and the fourth compressed learning data generated using the first and second inference models.

19 , first, lifestyle information according to the second item is input to the second inference model. At this time, the inputted lifestyle information corresponds to information collected by the user of the first service providing unit 2510 or the second service providing unit 2520 as described above, or information generated by random number generation.

The lifestyle information according to the second item is input to the second inference model, and information on the predicted adenoma is derived.

Thereafter, the fourth compressed learning data is generated based on the lifestyle information on the plurality of second items for each individual, lifestyle information on the plurality of second items for each individual, and the second inference model. It is created to include information on the presence or absence of adenoma. My

20 schematically illustrates a process of learning a third inference model according to an embodiment of the present invention.

The third inference model learning unit 2400 trains the third inference model by using the third compressed learning data and the fourth compressed learning data as described above. The third compressed learning data is ground truth, and information on the presence or absence of adenoma is designated as a value of 0 or 1, and information on the presence or absence of high-risk adenoma is designated as a value of 0 or 1.

Meanwhile, in the fourth compressed learning data, there is no information on the presence or absence of high-risk adenoma, but information on the presence or absence of adenoma has a value between 0 and 1. In this way, the learning data of the third inference model can be expanded by expanding the third compressed data, and the relationship between lifestyle information and probability information on the presence or absence of adenoma and probability information on the presence or absence of high-risk adenoma can be learned. can do it

21 schematically illustrates steps of providing high-risk vessel type information by the second service providing unit 2520 according to an embodiment of the present invention.

As described above, the server system receives lifestyle information for a plurality of first items from the first user requesting his/her predicted line type information, and receives lifestyle information for the first item and the first By deriving the predicted ship type information using the inference model, and the first service provided to the first user is performed, the second learning data is derived.

On the other hand, the ship type-related information prediction method based on the machine learning model receives lifestyle information for a plurality of second items from a second user who requests his/her predicted ship type information, and receives information about the received second item. A second service providing step of deriving the predicted ship type information using the lifestyle information and the second inference model and the third inference model, and providing it to the second user; may be performed.

As the first inference model, the second inference model, and the third inference model are constructed in this way, the user can receive the predicted high-risk adenoma information based on the machine learning model by inputting his/her lifestyle information.

22 illustrates adenoma occurrence prediction information UI displayed on the user terminal by the operation of the second service providing unit 2520 according to an embodiment of the present invention. 23 illustrates a high-risk adenoma occurrence prediction information UI displayed on a user terminal by the operation of the second service providing unit 2520 according to an embodiment of the present invention.

24 shows data on the accuracy of predicting the occurrence of adenoma and predicting the occurrence of high-risk adenoma according to embodiments of the present invention.

The model performance was checked by checking the performance statistics through 5-fold cross validation. The results are as shown in FIG. 24 .

- Total data: 78,792 cases

- Data collected during the period: August 01, 2008 - December 27, 2019

- Ratio of training data and test data (8:2)

Tables 1 and 2 below show prediction results of model performance in the second inference model and the third inference model, respectively, according to embodiments of the present invention.

[Table 1]

[Table 2]

25 schematically illustrates an internal configuration of a computing device according to an embodiment of the present invention.

The server system 1000 for performing the method for predicting colorectal cancer-related information according to FIG. 1 may include one or more modules of the computing device illustrated in FIG. 13 .

25, the computing device 11000 includes at least one processor 11100, a memory 11200, a peripheral interface 11300, an input/output subsystem ( I/O subsystem) 11400 , a power circuit 11500 and a communication circuit 11600 may be included at least. In this case, the computing device 11000 may include the system 1000 for deriving the asymmetric characteristic information of the face or may be connected to the server system 1000 by the input/output subsystem 11400 .

The memory 11200 may include, for example, a high-speed random access memory, a magnetic disk, an SRAM, a DRAM, a ROM, a flash memory, or a non-volatile memory. have. The memory 11200 may include a software module, an instruction set, or other various data required for the operation of the computing device 11000 .

In this case, access to the memory 11200 from other components such as the processor 11100 or the peripheral device interface 11300 may be controlled by the processor 11100 . The processor 11100 may be configured as a single or a plurality of processors, and may include a GPU and a TPU type processor in order to improve the processing speed.

Peripheral interface 11300 may couple input and/or output peripherals of computing device 11000 to processor 11100 and memory 11200 . The processor 11100 may execute a software module or an instruction set stored in the memory 11200 to perform various functions for the computing device 11000 and process data.

The input/output subsystem 11400 may couple various input/output peripherals to the peripheral interface 11300 . For example, the input/output subsystem 11400 may include a controller for coupling a peripheral device such as a monitor, keyboard, mouse, printer, or a touch screen or sensor as needed to the peripheral interface 11300 . According to another aspect, input/output peripherals may be coupled to peripheral interface 11300 without going through input/output subsystem 11400 .

The power circuit 11500 may supply power to all or some of the components of the terminal. For example, the power circuit 11500 may include a power management system, one or more power sources such as batteries or alternating current (AC), a charging system, a power failure detection circuit, a power converter or inverter, a power status indicator, or a power source. It may include any other components for creation, management, and distribution.

The communication circuit 11600 may enable communication with another computing device using at least one external port.

Alternatively, as described above, if necessary, the communication circuit 11600 may transmit and receive an RF signal, also known as an electromagnetic signal, including an RF circuit, thereby enabling communication with other computing devices.

This embodiment of FIG. 25 is only an example of the computing device 11000, and the computing device 11000 may omit some components shown in FIG. 25 or further include additional components not shown in FIG. 25, or 2 It may have a configuration or arrangement that combines two or more components. For example, a computing device for a communication terminal in a mobile environment may further include a touch screen or a sensor in addition to the components shown in FIG. 25 , and may include various communication methods (WiFi, 3G, LTE) in the communication circuit 1160 . , Bluetooth, NFC, Zigbee, etc.) may include a circuit for RF communication. Components that may be included in the computing device 11000 may be implemented in hardware, software, or a combination of both hardware and software including an integrated circuit specialized for one or more signal processing or applications.

Methods according to an embodiment of the present invention may be implemented in the form of program instructions that can be executed through various computing devices and recorded in a computer-readable medium. In particular, the program according to the present embodiment may be configured as a PC-based program or an application dedicated to a mobile terminal. The application to which the present invention is applied may be installed in the user terminal through a file provided by the file distribution system. For example, the file distribution system may include a file transmission unit (not shown) that transmits the file according to a request of the user terminal.

The device described above may be implemented as a hardware component, a software component, and/or a combination of the hardware component and the software component. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA). , a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions, may be implemented using one or more general purpose or special purpose computers. 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 execution of the software. For convenience of understanding, although one processing device is sometimes described as being used, one of ordinary skill in the art will recognize that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that may include For example, the processing device may include a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as parallel processors.

Software may comprise a computer program, code, instructions, or a combination of one or more thereof, which configures a processing device to operate as desired or is independently or collectively processed You can command the device. The software and/or data may be any kind of machine, component, physical device, virtual equipment, computer storage medium or device, to be interpreted by or to provide instructions or data to the processing device. , or may be permanently or temporarily embody in a transmitted signal wave. The software may be distributed over networked computing devices, and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through 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 medium may be specially designed and configured for the embodiment, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the methods and systems of the present invention have been described with reference to specific embodiments, some or all of their components or operations may be implemented using a computer system having a general purpose hardware architecture.

The above description of the present invention is for illustration, and those of ordinary skill in the art to which the present invention pertains can understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present invention. will be. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a distributed manner, and likewise components described as distributed may also be implemented in a combined form.

The scope of the present invention is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be interpreted as being included in the scope of the present invention. do.

Claims (6)

As a method of predicting ship type-related information based on a machine learning model, performed by a server system,
A first inference model statistically learned by first learning data including lifestyle information for a plurality of first items for each individual collected from medical institutions and adenoma information including information on the presence or absence of adenoma according to colonoscopy results The second learning data including lifestyle information and Zen type information for the plurality of first items for each individual is derived by deriving the predicted adenomatous information on the lifestyle information for the plurality of first items for each individual using a second learning data derivation step of deriving;
Generates first compressed learning data including lifestyle information and Zen type information for a second item in which some of the plurality of first items are removed from the first learning data, and in the second learning data, the plurality of first items A compressed data generating step of generating second compressed learning data including lifestyle information and Zen type information for a second item from which some of the first items are removed; and
A second reasoning for learning a second inference model that includes an artificial neural network and infers the possibility of information on the presence or absence of an adenoma with respect to the inputted lifestyle information for the second item by using the first compressed data and the second compressed data. A method for predicting adenoma related information based on a machine learning model, including a model learning step.
The method according to claim 1,
The second learning data derivation step is,
A learning information receiving step of receiving first learning data including lifestyle information for a plurality of first items for each individual collected from a medical institution, and adenoma information including information on the presence or absence of an adenoma according to a colonoscopy result;
The first learning data is clustered with adenomatous information according to the specific lifestyle information according to the first item, and based on the clustered information, the probability of adenomatosis for a group having the specific lifestyle information according to the first item is derived. possibility extraction stage;
a predictive data set deriving step of deriving a predictive data set including the adenomatous probability of a group having the specific lifestyle information according to a plurality of the first items; and
A prediction result derivation step of deriving a result on the possibility of adenoma of a person having the inputted lifestyle information based on the prediction data set with respect to the inputted lifestyle information for a plurality of first items for each individual; A method of predicting adenoma related information based on a machine learning model.
3. The method according to claim 2,
The step of deriving the prediction result is,
The degree of similarity is derived according to the matching degree between the lifestyle information for the plurality of first items for each individual input and the lifestyle information included in the prediction dataset, and it is determined whether the degree of similarity meets a preset standard. deriving a plurality of similar group prediction data; and
Information on the presence or absence of adenoma of a person who has lifestyle information for a plurality of first items for each individual inputted above by using a representative value of numerical information for information on the presence or absence of adenoma in the plurality of similar group prediction data A method of predicting adenoma related information based on a machine learning model, including; deriving probability information for.
The method according to claim 1,
Information on the presence or absence of adenoma of the first learning data is,
By extracting a keyword related to a polyp or adenoma from the colonoscopy result information and the biopsy result information collected from the medical institution, the first polyp data including at least one of the polyp location, polyp size, and polyp type for the individual is derived, , Adenoma-related information prediction method based on a machine learning model, which is derived in the form of normalized information by determining the information on the presence or absence of the adenoma from the first polyp data.
The method according to claim 1,
The first item and the second item include age, gender, BMI, exercise habit, drinking habit, smoking habit, family medical history and personal medical history,
The second item is a method of predicting ship type-related information based on a machine learning model, in which some of the first item is excluded.
The method according to claim 1,
In the second learning data derivation step,
The server system receives the lifestyle information for a plurality of first items from the first user who requests his/her predicted line type information, and uses the received lifestyle information for the first item and the first inference model. By deriving the predicted ship type information and performing the first service provided to the first user, the second learning data is derived,
The method of predicting ship type-related information based on the machine learning model is,
Receives lifestyle information for a plurality of second items from a second user who requests his/her predicted adenomatous information, and predicts adenomatous disease using the received lifestyle information for the second item and the second inference model A method of predicting ship type related information based on a machine learning model, further comprising; a second service providing step of deriving information and providing it to a second user.

Images