Method for predicting sudden pain in a subject

Abstract

To provide a method for predicting breakthrough pain in a cancer patient using an artificial intelligence model that processes time-series data collected from the cancer patient.SOLUTION: A method for predicting breakthrough pain in a subject comprises the steps of: receiving, by an analysis device, pain score data collected from a subject for a predetermined time; pre-processing, by the analysis device, the pain score data; and inputting, by the analysis device, the pre-processed pain score data into a pre-trained deep learning model and predicting whether or not breakthrough pain will be generated in the subject at a future time point based on a predicted value output by the deep learning model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a technique for predicting breakthrough pain in cancer patients using artificial intelligence. [Background technology]

[0002] Breakthrough pain refers to the sudden, transient worsening of pain experienced by cancer patients. Breakthrough pain reduces the quality of life of cancer patients. Currently, pain management in clinical settings involves the combination of long-acting analgesics and fast-acting analgesics. However, even with fast-acting analgesics, cancer patients are forced to suffer from breakthrough pain until the drug takes effect after administration. Summary of the Invention [Problem to be solved by the invention]

[0003] The present invention seeks to predict breakthrough pain in cancer patients using information collected from cancer patients over time. The present invention seeks to predict breakthrough pain in cancer patients using an artificial intelligence model that processes time series data collected from cancer patients. [Means for solving the problem]

[0004] The method for predicting sudden pain in a subject includes the steps of: receiving pain score data collected from the subject for a certain period of time by an analysis device; pre-processing the pain score data by the analysis device; inputting the pre-processed pain score data into a pre-trained deep learning model by the analysis device; and predicting whether the subject will experience sudden pain at a future time based on a predicted value output by the deep learning model.

[0005] The analysis device for predicting sudden pain in a subject includes an interface device that receives input of pain score data collected from the subject over a certain period of time, a storage device that stores a deep learning model that receives input of the patient's pain information and predicts whether or not sudden pain will occur, and a calculation device that preprocesses the input pain score data, inputs the preprocessed pain score data into a pre-trained deep learning model, and predicts whether or not the subject will experience sudden pain at a future time based on the predicted value output. [Effects of the Invention]

[0006] The present invention enables preemptive measures by predicting the occurrence of sudden pain based on clinical information of cancer patients collected clinically at medical institutions. Furthermore, in addition to pain control, the present invention can also be used for research on actual treatments for cancer patients and the development of biomarkers related to prognosis. [Brief explanation of the drawings]

[0007] [Figure 1] 1 is an exemplary diagram of a system for predicting sudden pain using information on a cancer patient. [Figure 2] FIG. 2 is an exemplary diagram of the learning process (200) of the deep learning model. [Figure 3] FIG. 10 is an exemplary diagram of a process for pre-processing pain data. [Figure 4] FIG. 1 is an exemplary diagram of a deep learning model that receives input of a patient's preprocessed pain information and predicts future breakthrough pain. [Figure 5] FIG. 1 is an illustrative diagram of a verification process for a deep learning model for predicting sudden pain. [Figure 6] AUC curve for the LSTM-based prediction model. [Figure 7] 1 is an exemplary diagram of an analysis device for predicting sudden pain; DETAILED DESCRIPTION OF THE INVENTION

[0008] The present invention can be modified in various ways and can have various embodiments, and a specific embodiment will be described in detail below with reference to the drawings. However, it is not intended to limit the present invention to the specific embodiment, and it should be understood that the present invention includes all modifications, equivalents, and alternatives that fall within the spirit and scope of the technology described below.

[0009] Terms such as "first," "second," "A," and "B" may be used to describe various components, but the components are not limited by these terms and are used solely to distinguish one component from another. For example, a first component may be termed a "second component," and similarly, a second component may be termed a "first component," without departing from the scope of the present invention. The term "and / or" includes a combination of multiple related listed items or any of multiple related listed items.

[0010] In the terms used in this specification, the singular expression should be understood to include the plural expression unless the context clearly indicates otherwise, and the term "comprise" or similar should be understood to mean that a stated feature, number, step, operation, component, part, or combination thereof is present, but not to exclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0011] Before proceeding to a detailed description of the drawings, it should be made clear that the division of components in this specification is merely a division according to the main function of each component. That is, two or more components described below may be incorporated into one component, or one component may be divided into two or more components according to more specific functions. Furthermore, each component described below may additionally perform some or all of the functions of other components in addition to its own main function, and some of the functions of each component may be exclusively performed by other components.

[0012] Furthermore, in performing a method or method of operation, each step constituting the method may be performed in a different order than specified unless the context clearly dictates a specific order, i.e., each step may be performed in the same order as specified, substantially simultaneously, or in the reverse order.

[0013] The present invention is a technique for predicting breakthrough pain in cancer patients. Furthermore, the technique of the present invention can be used to predict breakthrough pain in patients with a specific type of disease, regardless of the type of disease.

[0014] The present invention predicts breakthrough pain using time-series data that can be collected from cancer patients. The time-series data can include pain scores assessed over time and other clinical information, such as age, gender, disease type, and other EMR (Electronic Medical Record) data.

[0015] Hereinafter, an analysis device will be described as using a learning model to predict sudden pain based on time-series data of a cancer patient. The analysis device can be implemented as various devices capable of processing data and controlling the operation of the learning model. For example, the analysis device can be implemented as a PC, a server on a network, a smart device, a chipset with a dedicated program, etc.

[0016] The learning model can be any one of various types of models, such as a decision tree, a random forest, KNN (K-nearest neighbor), Naive Bayes, SVM (support vector machine), and ANN (artificial neural network). In particular, a deep learning model capable of processing time-series data can be used as the learning model.

[0017] 1 shows an example of a system 100 for predicting breakthrough pain using information on a cancer patient. In FIG. 1, an example of the analysis device is shown as a computer terminal 130 and a server 140.

[0018] The user terminal 110 collects a pain score (NRS) for a subject (patient). The NRS score falls within the range of [0, 10]. The NRS score is classified into mild pain, moderate pain, and severe pain according to the score. The pain score may be information calculated by medical staff evaluating the subject's condition over time. The user terminal 110 may store the collected pain score (NRS) in an EMR (Electronic Medical Record) 120 or a separate database (DB).

[0019] The learning device 50 analyzes patient information and constructs a deep learning model for predicting sudden pain. The learning device 50 refers to a computer device capable of preprocessing patient data and deep learning model training. The learning device 50 can construct a deep learning model using predetermined training data. The learning process and model structure will be described later. The computer terminal 130 and server 140 acquire the constructed deep learning model and use it for subject analysis.

[0020] In FIG. 1 , user A can analyze a patient's pain score and clinical information using computer terminal 130. Computer terminal 130 can receive input of the patient's pain score and clinical information from user terminal 110 or EMR 120 via a wired or wireless network. In some cases, computer terminal 130 can be a device physically connected to user terminal 110. Computer terminal 130 can pre-process the patient's pain score and / or clinical information. Computer terminal 130 can predict breakthrough pain based on values ​​output by a deep learning model that receives input of the patient's pain score and clinical information. User A can check the analysis results on computer terminal 130.

[0021] The server 140 may receive the patient's pain score and clinical information from the user terminal 110 or the EMR 120 via a wireless network. The server 140 may pre-process the patient's pain score and / or clinical information. The server 140 may predict sudden pain based on the value output by a deep learning model that receives the patient's pain score and clinical information. The server 140 may transmit the analysis results to the user A's terminal.

[0022] The computer terminal 130 and / or server 140 may also store the results of the analysis in the EMR 120 .

[0023] Below, we will explain the deep learning model that predicts sudden pain in cancer patients and the learning process of the deep learning model.

[0024] The inventors used a dataset of 3,431 patients out of 34,301 who were admitted to the Department of Hematology and Oncology at Samsung Medical Center, their affiliated research institution, between July 2016 and February 2020, excluding 2,697 who underwent surgery and 28,173 who had an NRS (non-zero numerical rating scale) score of less than 20 points.

[0025] 2 shows an example of a learning process 200 of the deep learning model. The deep learning model is a model that predicts (classifies) breakthrough pain using patient information. Therefore, the deep learning model may be called a predictive model.

[0026] The database (DB) can store the patient's pain score and clinical information included in the dataset. In this case, the pain score corresponds to time-series data collected over a certain period of time. The medical team collected the patient's pain score using the NRS scale at regular intervals. The medical team collected pain scores at 5:00, 1:00, and 9:00 p.m. every day. In addition, the medical team also collected information on any sudden pain experienced by the patient at other times. The clinical information may include age, gender, and other items. The inventor defined an NRS score of 4, which is the standard for administering narcotic analgesics according to guidelines, as breakthrough pain in cancer patients. In this case, an NRS score of 4 or higher is considered breakthrough pain.

[0027] The learner may pre-process the patient's raw data in some manner (step 210).

[0028] Figure 3 shows an example of the process of preprocessing pain data.

[0029] The learning device segments the pain score, which is time-series information, into fixed time intervals. The learning device assigns a pain score to each interval according to multiple time intervals (bins). If multiple pain scores exist in one interval, the learning device can set the score for that interval as the highest score. One bin can be defined as a fixed time length (e.g., τ hours). Figure 3(A) shows an example of segmenting pain scores into fixed time intervals (τ = 3 hours).

[0030] Breakthrough pain can be assessed differently within a single patient or between different patients. For example, breakthrough pain can have a specific pattern for each patient depending on the patient's medical condition. Breakthrough pain can also be assessed differently for each patient depending on the patient's management and condition. The learning device can process the entire data in units of a certain time period. The inventor processed pain score data in 24-hour units (0:00 to 24:00). In addition, the learning device can zero-pad portions that are not present in the original data to set the start and end of the pain score data to 0:00. Figure 3(B) shows an example of a process for processing pain score data collected continuously for n days in 24-hour units.

[0031] FIG. 3 illustrates that pain score data is divided into a certain time interval (τ=3 hours) and then divided into days, but the order of the process may be reversed.

[0032] The learning device can convert pain score data for n days divided into τ time bins from a 1×((24 / τ)) vector format to a (24 / τ)×n matrix format. Figure 3(C) shows an example of pain score data processed in 24-hour units converted into a matrix format. One row in Figure 3(C) is pain score data for a one-day time interval, and the vertical columns are examples of pain score data collected in n-day units.

[0033] Thus, the input data may also utilize further clinical information such as the patient's gender and age in addition to the pain score.

[0034] The learning device can then build a prediction model using the pain score data preprocessed in the matrix form. Furthermore, the learning device can also build a prediction model by further using clinical information along with the pain score data, which is time-series data.

[0035] The inventors constructed six different types of deep learning models. The deep learning models include a recurrent neural network (RNN), a long short-term memory (LSTM), a gated recurrent unit (GRU), a bidirectional long short-term memory (Bi-LSTM), a hybrid of the convolutional neural network, long short-term memory (CNN-LSTM), and a transformer. The inventors then evaluated the performance of the models constructed by varying the length of the time interval of the data. The inventors constructed models by dividing the time interval length into units τ∈{1, 2, 3, 4, 6, 8, 12}.

[0036] The learning device divides the collected data into training data and validation data (step 220). The inventors used 80% of the data from 2,745 people in the dataset of 3,431 people as training data, and 20% of the data from 686 people as test data.

[0037] The learning device performs learning of the prediction model using the learning data (step 220). The learning device extracts one input data from the learning data and performs the learning process. The learning device compares the value (sudden pain prediction value) output by the prediction model that received the input data with the correct value and updates the parameters of the prediction model. The learning device repeats the learning process using data that belongs to the learning data.

[0038] Once the learning process is complete, the learning device performs verification of the trained prediction model (step 230). The learning device selects one of the verification data and performs the verification process. The learning device inputs the selected data into the prediction model, outputs a sudden pain prediction value, and performs verification by comparing the predicted result with the correct value.

[0039] Figure 4 shows an example of a deep learning model that receives preprocessed pain information from a patient and predicts a future breakthrough pain. The analysis device receives past pain score data from the patient to be predicted. The analysis device preprocesses the pain score data in a certain way. The preprocessing process is the same as described above. The analysis device can convert the pain score data for n days, divided into bins of time length τ, into a 24 / τ x n matrix format. Figure 4 shows an example where time τ = 3 hours. The analysis device inputs the matrix-format pain data into the constructed deep learning model. The deep learning model can be composed of multiple convolutional layers (e.g., Layer 1 to Layer 3) and a dense layer. The dense layer can use an activation function to ultimately output information predicting the likelihood of a breakthrough pain occurring for each time interval (bin). Based on the information output by the deep learning model, the analysis device can predict whether a breakthrough pain will occur within 24 hours from the last time point when the patient's information was collected. For example, the analysis device can predict that sudden pain is likely to occur in the second period (03:00 to 06:00), the fourth period (09:00 to 12:00), and the fifth period (12:00 to 15:00) in Figure 4.

[0040] Figure 5 shows an example of the verification process for a deep learning model that predicts sudden pain. The analysis device uses 72-hour pain score data to predict sudden pain for the next 24 hours. In this case, the analysis device can predict the likelihood of sudden pain occurring based on the pain score predicted by the deep learning model for each time period. For example, if the pain score predicted for a specific time period is 4 or greater, the analysis device can determine that sudden pain will occur in that period (binary value 1), and if the pain score predicted for a specific time period is less than 4, the analysis device can determine that sudden pain will not occur in that period (binary value 0).

[0041] The analysis device can verify the performance of the deep learning model by comparing the predicted value of sudden pain for a time interval with the true value based on the value output by the model. Matthews correlation coefficient (MCC) can be used as a verification index. MCC is a value in the range [-1, 1], with values ​​closer to +1 indicating higher accuracy. Table 1 below shows the performance of each model. In Table 1, the input data is divided into data from the past 24 hours, 72 hours, and 120 hours. The input data used differs from each other in the time interval τ∈{1, 2, 3, 4, 6, 8, 12}.

[0042] [Table 1]

[0043] As shown in Table 1, the LSTM-based model showed high overall performance. The LSTM model performed best with an MCC of 0.4927 in the example using a 12-hour time interval for 120 hours of data.

[0044] Figure 6 shows the AUC curves for the LSTM-based prediction model. Figure 6(A) shows the results using pain score data from the past 24 hours, Figure 6(B) shows the results using pain score data from the past 72 hours, and Figure 6(C) shows the results using pain score data from the past 120 hours. The LSTM-based prediction model showed significant performance for data from various time intervals.

[0045] 7 shows an example of an analysis device 300 for predicting sudden pain. The analysis device 300 corresponds to the above-described analysis devices (130 and 140 in FIG. 1). The analysis device 300 can be physically embodied in various forms. For example, the analysis device 300 can be in the form of a computer device such as a personal computer, a network server, a chipset dedicated to data processing, etc.

[0046] The analysis device 300 may include a storage device 310 , a memory 320 , a processing device 330 , an interface device 340 , a communication device 350 , and an output device 360 ​​.

[0047] The storage device 310 can store the patient's past pain score data. The pain score data can be data collected for a predetermined period of time, such as the past 24 hours, the past 72 hours, or the past 120 hours.

[0048] The storage device 310 can store clinical information of the patient (such as gender and age).

[0049] The storage device 310 may store a deep learning model for predicting sudden pain. In this case, the storage device 310 may store multiple models constructed using time data of different lengths.

[0050] The storage device 310 can store the analysis results.

[0051] The memory 320 may store data and information generated in the process of the analysis device 300 predicting sudden pain using the patient's pain score data.

[0052] The interface device 340 is a device that receives predetermined commands and data input from the outside.

[0053] The interface device 340 can receive input of the subject's pain score data from a physically connected input device or an external storage device, and can also receive input of the subject's clinical information from a physically connected input device or an external storage device.

[0054] The interface device 340 can also transmit the predicted results of the breakthrough pain to an external object. The predicted results can consist of a predicted value of breakthrough pain for 24 hours from the end of data collection for the patient.

[0055] The communication device 350 refers to a component that receives and transmits predetermined information through a wired or wireless network.

[0056] The communication device 350 can receive pain score data for the subject from an external object. The communication device 350 can receive clinical information for the subject.

[0057] Alternatively, the communication device 350 may transmit the predicted traffic outage results to an external object such as a user terminal.

[0058] On the other hand, the interface device 340 means that it includes an interface that transfers data and information received from the communication device 350 to the inside of the analysis device 300 .

[0059] The output device 360 ​​is a device that outputs predetermined information, and can output interfaces, analysis results, etc., required for the data processing process.

[0060] The computing device 330 may pre-process the subject's initial pain score data in a predetermined manner. As described above, the pre-processing may include dividing the data into time segments and generating pain score data in a matrix format by processing the data in 24-hour units.

[0061] The computing device 330 inputs the (preprocessed) pain score data into a pre-trained deep learning model, and can predict whether or not the subject has experienced sudden pain based on the probability value output by the deep learning model.

[0062] In this case, a deep learning model that matches the input data according to the time length of past data collected from the patient and / or the length of the time interval unit into which pain data is divided can be used.

[0063] The computing device 330 can determine whether a sudden pain will occur in a specific section of the future time section based on the output value of the deep learning model.

[0064] The computing device 330 may be a device such as a processor, AP, or a chip with a built-in program that processes data and performs predetermined operations.

[0065] Furthermore, the method for predicting sudden pain in cancer patients as described above may be implemented as a program (or application) including an executable algorithm that can be executed by a computer. The program may be provided by being stored in a non-transitory or non-transitory computer-readable medium.

[0066] A non-transitory readable medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short period of time such as a register, cache, memory, etc. Specifically, the various applications or programs described above can be stored and provided in a non-transitory readable medium such as a CD, DVD, hard disk, Blu-ray disc, USB, memory card, ROM (read-only memory), PROM (programmable read-only memory), EPROM (erasable PROM), EPROM or EEPROM (electrically EPROM), or flash memory.

[0067] The temporary readable medium refers to various types of RAM such as static RAM, SRAM, dynamic RAM, DRAM, synchronous DRAM, SDRAM, double data rate SDRAM, DDRSDRAM, enhanced SDRAM, ESDRAM, synclink DRAM, SLDRAM, and direct Rambus RAM (DRRAM).

[0068] The present embodiment and the drawings attached to this specification only clearly show a part of the technical ideas contained in the present invention, and it is obvious that any modifications and embodiments that can be easily inferred by a person skilled in the art within the scope of the technical ideas contained in the above-mentioned specification and drawings of the present invention fall within the scope of the rights of the present invention.

Claims

1. receiving pain score data collected from the subject over a period of time including multiple time intervals by the analysis device; the analysis device preprocessing the pain score data by assigning pain scores to time interval units from the pain score data; The analysis device inputs the pre-processed pain scores for each time interval into a pre-trained deep learning model; A method for predicting sudden pain in a subject, comprising: a step in which the analysis device predicts whether or not the subject will experience sudden pain at a future time based on a predicted value output by the deep learning model; The pain score data includes time series data of pain scores, The deep learning model is trained to output whether or not there will be a sudden pain at a future time point using training data including pain scores obtained from cancer patients for each time interval. method.

2. The method for predicting sudden pain in a subject according to claim 1 , wherein the deep learning model outputs a probability value of sudden pain occurrence for each time interval.

3. An interface device for receiving input of pain score data collected from a subject over a period of time including multiple time intervals; A storage device for storing a deep learning model that predicts whether sudden pain will occur; An analysis device for predicting sudden pain in a subject, the analysis device including: a calculation device that preprocesses the input pain score data by allocating pain scores to time interval units from the pain score data, inputs the preprocessed pain scores for each time interval unit into a pre-trained deep learning model, and predicts whether or not the subject will have sudden pain at a future time point based on the output predicted value; The pain score data includes time series data of pain scores, The deep learning model is trained to output whether or not there will be a sudden pain at a future time point using training data including pain scores obtained from cancer patients for each time interval. Analyzer.

4. The analysis device for predicting sudden pain in a subject according to claim 3 , wherein the deep learning model outputs a probability value of sudden pain occurrence for each time interval.

Images