A patient position monitoring method based on a machine learning model

Abstract

The present invention discloses a patient body position monitoring method based on a machine learning model, which relates to the monitoring of a patient's body position. This method uses an intelligent mattress made of flexible intelligent sensors, designs experiments to collect the patient's sensor data, preprocesses the data, and then selects corresponding feature regions and attributes according to the data and clinical actual situations. A classification model is constructed using machine learning methods. Then, the real data of the patient is collected again. After the same data preprocessing, the previously constructed model is verified, so as to improve the model. Finally, a classification prediction model that can effectively predict whether the patient's body position has changed is constructed. The present invention has important clinical significance for the detection of changes in the patient's body position.

Description

Technical Field

[0001] The present invention belongs to the field of medical technology, and particularly relates to a method for monitoring a patient's body position based on a machine learning model. Background Art

[0002] In various types of applications, real-time monitoring of a patient's body position has important clinical significance, and various image guidance technologies have been developed.

[0003] Current image guidance systems are mainly divided into the following two categories: one is an X-ray based image guidance system, such as an electronic portal imaging device (EPID), a kilovoltage cone beam CT (CBCT), an orthogonal X-ray imaging system, a megavoltage cone beam CT (TOMO), a CT-on-rail system, etc. The other is a non-X-ray based image guidance system. There are many and complex types of such image guidance systems, mainly including a three-dimensional ultrasound guidance system (BAT system and clarity system), a nuclear magnetic resonance image guidance system, an electromagnetic navigation system, an optical surface tracking system, etc. Represented by the above two methods, the main defects of current clinically used image guidance systems are: generating additional radiation, being invasive, having large errors and low precision, or being unable to perform real-time online image guidance, and being expensive. These problems greatly limit the wide use of image guidance systems, and most current methods observe from the front, without a method for observing from the back. Therefore, an effective, non-invasive, real-time, and cost-effective method for monitoring a patient's body position is needed. Summary of the Invention

[0004] In order to improve the performance of the patient body position detection method, the present invention proposes a method for monitoring a patient's body position based on a machine learning model. According to the experimental data of an intelligent mattress made of flexible intelligent sensors collected, through a preprocessing step, and then using a machine learning method to establish a classification prediction model, it is possible to effectively and real-time predict whether a patient has a body position change.

[0005] The technical solution of the present invention is as follows:

[0006] A method for monitoring a patient's body position based on a machine learning model, comprising the following steps:

[0007] Step 1, data collection: Collect the sway track length and sway track area data of 5 collection areas of a precise intelligent mattress. The data collection frequency is to count the data every 5 seconds, and collect the data of two groups, namely the moving group and the non-moving group;

[0008] Step 2, data preprocessing: Data preprocessing includes data summarization, data numbering, null value processing, feature screening, feature engineering, and outlier screening;

[0009] Step 3, train and generate a classification prediction model: Select a machine learning model as the classification prediction model, and use the preprocessed dataset in Step 2 as the model input. The dataset is divided into a training set and a test set. According to the training situation, use the model parameter adjustment method to adjust the model parameters, and then verify the model effect on the newly collected validation set, and continuously optimize the model to achieve the classification prediction function;

[0010] Step 4, use the model to predict whether the patient's body position has changed.

[0011] Further, in the above Step 1, the five acquisition regions are the right scapula region, the left scapula region, the back region, the lumbar and abdominal region, and the sacrococcygeal region.

[0012] Further, in the above Step 1, the sway trajectory length and the sway trajectory area are related to the movement of the center of pressure (COP) of the region. The center of pressure (COP) of the region is the coordinate of the center of gravity point of the region; among them, after the pressure value data of each point collected by the pressure sensor is processed, the trajectory map of the center of pressure (COP) of the region is statistically analyzed. The sway trajectory length is the moving distance of the COP within the sampling frequency time, and the distance between two points is calculated using the Euclidean distance. The sway trajectory area is the covered area of the COP moving region.

[0013] Further, the data acquisition methods for the two groups in the above Step 1 are as follows: The data of the non-moving group are the body position data of the patient lying on the mattress and being required not to move. The data of the moving group are the body position data of the patient moving on the mattress under the guidance of the staff or the data collected in the simulation experiment; 30 groups of data of the non-moving group are collected at one time.

[0014] Further, in the above Step 2, select the data of the right scapula region, the left scapula region, and the back region from the five acquisition regions to construct a dataset for model construction and verification.

[0015] Further, in the above Step 2, feature engineering means adding 1 to all the data first and then taking the logarithm of the data.

[0016] Further, in the above Step 3, the dataset adopts the random division method, where 80% of the data is divided into the training set, and the remaining 20% of the data is divided into the test set.

[0017] Further, in the above Step 3, the selected machine learning model is the LightGBM model.

[0018] Further, in the above Step 3, the model parameter adjustment method is the grid search method.

[0019] Further, it is characterized in that the trained model can realize the classification prediction function for the input data and effectively predict whether the patient's body position changes at the current moment.

[0020] The beneficial effects of the present invention are as follows: The intelligent mattress made of flexible intelligent sensors constructs an accurate machine learning classification model by using the real data collected in the actual application process, and can realize non-invasive and real-time monitoring of the patient's body position, which has important clinical significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is the method flow chart of the present invention based on the machine learning model;

[0022] Figure 2 is the AUC curve graph of the training set and the validation set. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] To further illustrate the embodiments, the present invention provides drawings. These drawings are part of the disclosure of the present invention, and their main function is to illustrate the embodiments and can be used in conjunction with the relevant descriptions to explain the operating principle of the embodiments. With reference to these contents, those of ordinary skill in the art should be able to understand other possible implementation manners and the advantages of the present invention. The present invention will be further described below in conjunction with the drawings and specific implementation manners.

[0024] As Figure 1 shown, a method for monitoring a patient's body position based on a machine learning model of the present invention includes the following steps:

[0025] Step 1, data collection: Collect the wobbling track length and wobbling track area data of 5 collection areas on the intelligent mattress. The data collection frequency is to count the data every 5 seconds, and collect the data of two groups, namely the moving group and the non-moving group;

[0026] Step 2, data preprocessing: Data preprocessing includes data summarization, data numbering, null value processing, feature screening, feature engineering, and outlier screening;

[0027] Step 3, train and generate a classification prediction model: Select a machine learning model as the classification prediction model, use the data set preprocessed in Step 2 as the model input. The data set is divided into a training set and a test set. According to the training situation, use the model parameter adjustment method to adjust the model parameters, and then verify the model effect on the newly collected validation set, and continuously optimize the model to realize the classification prediction function;

[0028] Step 4, use the model to predict whether the patient's body position changes.

[0029] In the said Step 1, the five data acquisition regions are the right scapular region (Region A), the left scapular region (Region B), the back region (Region C), the lumbar and abdominal region (Region D), and the sacrococcygeal region (Region E).

[0030] I. Experimental Environment and Equipment

[0031] In this experiment, an intelligent mattress based on flexible pressure sensor technology is used. A mattress formed by splicing two flexible sensors with a size of 30*30 cm is placed on the bed and connected to a computer through a data transmission line. Software for real-time observing pressure changes and recording experimental data is configured on the computer. The software uses TCP / IP communication rules to transmit data with the signal acquisition module. After the data of the pressure sensors are processed by the software, the statistically processed data are saved in the computer. The relevant parameters of this mattress are shown in the following table.

[0032]

[0033] II. Experimental Process

[0034] During the experimental process, when the patient lies correctly on the intelligent mattress following the doctor's instructions, start observing the change of the values transmitted back by the pressure sensors and record the experimental data. In the experiment, the sway trajectory area and sway trajectory length data of five regions are collected. The collection frequency is to count once every 5 seconds, that is, to count the COP sway trajectory length and sway trajectory area within 5 seconds. Two sets of data need to be collected during the collection process, namely the moving group and the non-moving group. The data of the non-moving group are the body position data when the patient lies on the mattress and is required not to move. The data of the moving group are the body position data when the staff guides the patient to move on the mattress or the data collected in the simulation experiment. An average of about 30 non-moving group data are collected for one patient at a time. A total of 10108 pieces of data are collected during the experimental process, including 5103 pieces in the moving group and 5005 pieces in the non-moving group. This part of the data is used for model training and construction. In addition, another 10422 pieces of data are collected to construct an independent validation set.

[0035] III. Data Processing

[0036] Data processing includes: data aggregation, data numbering, null value processing, feature screening, feature engineering, etc. First, aggregate and merge the data of different patients, and number them according to the patient's name, sampling date, sampling group, and sampling order. Check for null value data and delete the data with null values. The occurrence of null value data in some areas may be related to the patient's body position changes and unstable data signal transmission of the sensor. According to clinical experience and literature research, select the data in the right scapula area (area A), left scapula area (area B), and back area (area C) to construct a data set. Feature engineering is to add 1 to the data values and then take the logarithm. After completing the data feature engineering, draw a box plot and view the parameter data such as the maximum value, minimum value, mean value, median value, and standard deviation of the data in different areas. Combining the experimental situation, if there is a part of the data that is extremely large or extremely small, consider this part of the data as outlier data and delete this part of the data to obtain the final data set for constructing the model. Then collect data according to the above process to construct an independent validation set.

[0037] IV. Machine Learning Model Training

[0038] Use the data set obtained through the aforementioned processing, and use the train_test_split method in the sklearn package to divide the data set, where the training set data accounts for 80% and the test set data accounts for 20%.

[0039] Use the data in the training set, take the wavering trajectory area and wavering trajectory length in areas A, B, and C as input features, and use 1 for body position change and 0 for no body position change as the output for binary classification. Train through the LightGBM algorithm, and comprehensively evaluate the model using the prediction accuracy, AUC value, ROC curve, etc. (as Figure 2 shown), use the gridsearch grid search method to optimize the model parameters. The result of training on the test set is that the accuracy ACC value is 0.89 and the AUC value is 0.96. The comprehensive prediction accuracy obtained on the validation set is 0.72 and the AUC value is 0.86. After the model is constructed, when an input of real-time observation data is given, the predicted category can be given to determine whether the patient has had a body position change, thus realizing the detection function.

[0040] Relying on the real experiments carried out in the hospital, a machine learning model constructed with a large amount of effective experimental data collected. This method can effectively predict the body position changes of patients and has important clinical significance.

[0041] The above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A patient position monitoring method based on a machine learning model, characterized in that, It includes the following steps: Step 1, data collection: Collect the data of the sway track length and sway track area of 5 collection areas on the intelligent mattress. The data collection frequency is to count the data every 5 seconds, and collect the data of two groups, namely the moving group and the non-moving group; Step 2, data preprocessing: Data preprocessing includes data summarization, data numbering, null value processing, feature screening, feature engineering and outlier screening; Step 3, train and generate a classification prediction model: Select a machine learning model as the classification prediction model, use the dataset preprocessed in Step 2 as the model input. The dataset is divided into a training set and a test set. According to the training situation, use the model parameter adjustment method to adjust the model parameters, and then verify the model effect on the newly collected validation set, and continuously optimize the model to achieve the classification prediction function; Step 4, use the model to predict whether the patient's body position has changed; In the above Step 1, the 5 data collection areas are the right scapula area, the left scapula area, the back area, the lumbar and abdominal area, and the sacrococcygeal area respectively; In the above Step 1, the sway track length and sway track area are related to the movement of the center of pressure (COP) of the area. The center of pressure COP of the area is the coordinate of the center of gravity point of the area; among them, after the pressure value data of each point collected by the pressure sensor is processed, the trajectory map of the center of pressure COP of the area is counted. The sway track length is the moving distance of the COP within the sampling frequency time, and the distance between two points is calculated using the Euclidean distance. The sway track area is the covered area of the COP moving area.

2. The patient position monitoring method based on a machine learning model according to claim 1, characterized in that, The data collection methods of the two groups in the above Step 1 are as follows: The data of the non-moving group are the body position data of the patient when lying on the mattress and being required not to move, and the data of the moving group are the body position data when the staff guides the patient to move on the mattress or the data collected in the simulation experiment; 30 groups of the non-moving group data are collected at one time.

3. The patient position monitoring method based on a machine learning model according to claim 1, characterized in that, In the above Step 2, screen the data of the right scapula area, the left scapula area, and the back area from the 5 collection areas to construct a dataset for model construction and verification.

4. The patient position monitoring method based on a machine learning model according to claim 1, characterized in that, In the above Step 2, feature engineering means that for all data, first add 1 to the data, and then take the log of the data.

5. The patient position monitoring method based on a machine learning model according to claim 1, characterized in that, In the above Step 3, the dataset adopts the random division method, where 80% of the data is divided into the training set, and the remaining 20% of the data is divided into the test set.

6. The patient position monitoring method based on a machine learning model according to claim 1, characterized in that, In the above Step 3, the selected machine learning model is the LightGBM model.

7. The patient position monitoring method based on a machine learning model according to claim 1, characterized in that, In the above Step 3, the model parameter adjustment method is the grid search method.

8. The patient position monitoring method based on a machine learning model according to any one of claims 1-7, characterized in that, The trained model can achieve the classification prediction function for the input data, and effectively predict whether the patient's body position has changed at the current moment.

Images