CN115227504B - An automatic raising and lowering hospital bed system based on EEG signal - Google Patents

Abstract

The invention discloses an automatic lifting sickbed system based on an electroencephalogram-eye electric signal, which belongs to the field of rehabilitation medical treatment and comprises an electroencephalogram sensor, a sickbed control system and a sickbed body, wherein the sickbed body has lifting, active calling and alarming functions; the brain wave sensor is connected with the sickbed control system and is used for collecting brain electric eye electric fusion signals of a patient and transmitting the brain electric eye electric fusion signals to the sickbed control system; the sickbed control system sends a control instruction according to an electroencephalogram-eye electric fusion signal of a patient, and the control instruction is used for controlling the sickbed body to lift, level, actively call and alarm, and the sickbed body responds according to the control instruction; lifting, active calling and intelligent alarm can be realized under the bedridden state of the patient by monitoring the electroencephalogram signals in real time.

Description

An automatic raising and lowering hospital bed system based on EEG signal

technical field

The invention belongs to the field of rehabilitation medicine, and in particular relates to an automatic raising and lowering sickbed system based on electroencephalogram-oculogram signals.

Background technique

For long-term bedridden patients, especially those with inconvenient limbs, it is necessary to adjust the state of the bed with the help of the nursing staff, which increases the labor intensity of the nursing staff, and the patient cannot independently adjust the state of the bed or call a doctor. Although the existing hospital beds can automatically realize functions such as lifting and alarming through simple buttons, they are only suitable for patients who can perform simple activities. For patients who are completely bedridden with physical inconvenience, they do not have any mobility. Ordinary intelligent hospital beds cannot fulfil requirements.

With the continuous development of science and technology, human beings have a variety of scientific research methods on the brain. Among them, the brain-computer interface method uses EEG, electro-oculogram signals to form a corresponding relationship with the movement of the eyes and eye appendages. Using these corresponding relationships, a The bed system that does not rely on limbs can help bedridden patients get up, actively call and intelligently alarm, which has become a research hotspot in rehabilitation medical engineering.

Contents of the invention

In order to solve the above technical problems, the present invention proposes an automatic lifting hospital bed system based on EEG signals, which can realize lifting, active calling and intelligent alarming when patients are lying in bed through real-time monitoring of EEG signals.

The present invention adopts following technical scheme:

An automatic raising and lowering hospital bed system based on EEG signals, including an EEG sensor, a bed control system and a bed body, the bed body has the functions of lifting, active calling and alarming;

The brainwave sensor is connected with the bed control system for collecting the patient's EEG fusion signal and transmitting it to the bed control system; the hospital bed control system sends control instructions according to the patient's EEG fusion signal, and the The control command is used to control the raising, leveling, active calling and alarm of the bed body, and the bed body responds according to the control command;

The hospital bed control system includes:

EEG fusion signal noise filtering module, which is used for noise filtering processing on EEG fusion signal;

EEG-oculo-fusion signal segmentation window module, which is used to divide the EEG-oculo-fusion signals after noise filtering into different groups with the same length;

The EEG-oculofusion signal pattern recognition module is used to identify the characteristic type of each group of EEG-oculofusion signals, and send instructions according to the characteristic type.

As a preference of the present invention, the brain wave sensor is wirelessly connected with the bed control system.

As a preference of the present invention, the hospital bed control system sends control instructions according to the patient's EEG fusion signal, including:

S1, performing noise filtering processing on the EEG fusion signal to obtain noise-filtered EEG data;

S2, dividing the noise-filtered EEG data into groups, obtaining EEG data groups, and calculating the waveform characteristics of each EEG data group, the waveform characteristics of the EEG data groups include average value and absolute average value, mean square and standard deviation;

S3, using the K-nearest neighbor algorithm to classify the waveform characteristics of each EEG data group relative to the waveform characteristics of the sample database data group, and obtain the characteristic type of each EEG data group; according to the affiliation of each EEG data group Characteristic types send control commands.

As a preference of the present invention, the characteristic types of each EEG data group include one blink, two blinks, eyeball rotation and EEG abnormalities, and the corresponding control instructions are respectively bed raising, laying down, active calling and Call the police.

As a preference of the present invention, said step S3 includes:

Calculate the distance between the waveform feature of the EEG data group to be tested and each EEG data group in the sample library data group, and obtain the distance set D={d _1,1 ,d _1,2 ,...,d _{1, m1} ，d _2,1 ，d _2,2 ，…，d _2,m2 ，…，d _4,1 ，d _4,2 ，…，d _4,m4 }, where m1 represents the data group in the sample library that belongs to The number of EEG data sets of the one-blink type, m2 represents the number of EEG data sets belonging to the double-blink type in the sample database data set, and m3 represents the EEG data sets of the eye movement type in the sample library data set The number of data groups, m4 represents the number of EEG data groups belonging to the EEG abnormal type in the sample library data group, d1 _{, m1} represents the waveform characteristics of the EEG data group to be tested and the m1th eye blink type The distance between the waveform features of the EEG data set, d _{2, m2} represents the distance between the waveform features of the EEG data set to be tested and the m2 second blink type of the EEG data set Distance, d _{3, m3} represent the distance between the waveform feature of the EEG data set to be tested and the waveform feature of the m3 eyeball movement type EEG data set, d _{4, m4} represent the EEG data set to be tested The distance between the waveform feature of the data set and the waveform feature of the m4th EEG abnormal type EEG data set;

Arrange the values in the distance set D in ascending order, take the first K distances, judge the number of four feature types in the sample library data group corresponding to the first K distances, and take the feature type with the largest number as the EEG to be tested The feature type to which the data group belongs.

As a preference of the present invention, the method for constructing the sample library data set includes:

S31, collect the EEG fusion signals of different patients in the case of one blink, two blinks, eyeball rotation and EEG abnormalities through the EEG sensor, perform noise filtering processing on the EEG fusion signals and divide them into groups, and extract each Waveform characteristics of the data group to obtain a database with a total data volume of p groups;

S32, randomly generate m sample database data groups, r training database data groups, and s test database data groups from the database whose total data volume is p groups; the sample database data groups, training database data groups Both the test database and the test library data set contain four types of data, namely one blink, two blinks, eye movement and EEG abnormalities, and are not repeated;

S33, repeating step S32, initializing several data sets, each data set is composed of randomly initialized sample database data set, training database data set, test database data set, K value parameter of the K nearest neighbor algorithm;

Calculate the distance set between the waveform features of each training database data group and the waveform features of each sample database data group, use the K nearest neighbor algorithm to classify the characteristics of a training database data group, and construct the objective function with the recognition rate and misjudgment rate, Select the best data set;

S34, taking the sample library data group in the optimal data set obtained in step S33 as the optimal sample library data group, calculating the waveform characteristics of the test library data group in the optimal data set and the waveforms of each sample library data group For the distance set between features, the K-nearest neighbor algorithm is used to classify the features of a test library data group, and the objective function is constructed based on the recognition rate and misjudgment rate to judge whether it meets the preset requirements;

If satisfied, the sample library data set in the optimal data set obtained in step S33 is the final optimal sample library data set;

If not, return to step S32 to re-initialize several data sets.

As a preference of the present invention, the step S33 further includes the steps of selecting, crossing and mutating the data groups in the initialized data sets by genetic algorithm, and replacing the original data sets with the new data sets obtained.

Compared with the prior art, the present invention has the advantages of:

1) The hospital bed system proposed by the present invention realizes automatic lifting, active calling and intelligent alarm by monitoring the EEG signal of the patient, and adopts the K nearest neighbor method to identify the characteristic type of the EEG signal, and the calculation speed is fast, which ensures the timely response of the hospital bed system ;

2) The present invention proposes a sample library construction method based on an embedded AI algorithm, which ensures the accuracy of the sample library and improves the accuracy of feature type recognition of EEG signals.

Description of drawings

Fig. 1 is the overall schematic diagram showing a kind of automatic raising and lowering hospital bed system based on EEG signal according to the embodiment of the present invention;

Fig. 2 is a schematic diagram of data processing of the hospital bed control system shown in the embodiment of the present invention;

Fig. 3 is a schematic flow chart of pattern recognition of EEG fusion signals shown in an embodiment of the present invention;

Fig. 4 is a schematic diagram of a sample library construction process based on an embedded AI algorithm shown in an embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below by means of embodiments in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solution of the present invention more clearly, but not to limit the protection scope of the present invention. Some block diagrams shown in the accompanying drawings are functional entities, which do not necessarily correspond to physically or logically independent entities, and these functional entities may be implemented in the form of software, or implemented in one or more hardware modules or integrated circuits functional entities, or implement these functional entities in different networks and/or processor devices and/or microcontroller devices.

The flowcharts shown in the figures are illustrative only and do not necessarily include all steps. For example, some steps can be decomposed, and some steps can be combined or partly combined, so the actual execution sequence may be changed according to the actual situation.

Fig. 1 is an overall schematic diagram showing an automatic lifting hospital bed system based on EEG signals according to an embodiment of the present invention. The automatic lifting hospital bed system includes an electroencephalogram sensor, a hospital bed control system and a hospital bed body. The hospital bed body has lifting, The functions of leveling, active calling and alarming can be realized through motion modules or alarming modules that realize different functions, and the four functions are controlled by sending control commands from the bed control system.

The electroencephalogram sensor is used for the collection and transmission of EEG signals and electrooculogram signals, and is connected with the bed control system to collect the patient's EEG fusion signals and transmit them to the bed control system; during the signal acquisition process, the EEG The signal and the electro-ocular signal are fused together for output. In this embodiment, the brainwave sensor is wirelessly connected to the hospital bed control system, for example, by using wireless communication methods such as bluetooth and wifi.

The hospital bed control system is used to perform noise filtering, feature extraction, and pattern recognition on the patient's EEG fusion signal, and convert it into a bed control instruction, that is, send a control instruction according to the patient's EEG fusion signal. The control instructions of the hospital bed are used to control the raising and lowering of the bed body, active calls and alarms, and the bed body responds according to the control instructions. The hospital bed control system includes:

EEG fusion signal noise filtering module, which is used for noise filtering processing on EEG fusion signal;

The EEG-oculo-fusion signal segmentation window module is used to divide the EEG-oculo-fusion signals after noise filtering into different groups with the same length.

The EEG-oculo-fusion signal pattern recognition module is used to extract the waveform features of the EEG-oculo-fusion signals and identify the feature type of each group of EEG-oculo-fusion signals, and send instructions according to the feature type. In this embodiment, as shown in Figure 2, the characteristic types of each EEG data group include one blink, two blinks, eyeball rotation and EEG abnormality, and the corresponding control instructions are bed raising, lowering, etc. Ping, active call and alarm. The EEG abnormality refers to the EEG signal collected when the patient is in acute cerebral infarction, cerebral insufficiency, epilepsy, etc.; waveform features include average value, absolute average value, sample mean square deviation, sample standard deviation, etc. .

As shown in Figure 3, the K-nearest neighbor algorithm is used to classify and judge each EEG data group relative to the sample library, and obtain the feature type of each EEG data group; according to the feature type of each EEG data group, the hospital bed The control system issues a bed control command.

In a specific implementation of the present invention, for a certain EEG data set, first calculate its waveform features; then use a distance function, such as Euclidean distance, to calculate the waveform features of the EEG data set and the sample database data set The distance set between the waveform features of each EEG data group is recorded as the distance set D={d _1,1 ,d _1,2 ,...,d _1,m1 ,d _2,1 ,d _2,2 ,... , d _{2, m2} ,..., d _4,1 , d _4,2 ,..., d _{4, m4} }, wherein, m1 represents the number of EEG data groups belonging to one blink type in the sample library data group, m2 Represents the number of EEG data groups belonging to the double blink type in the sample library data group, m3 represents the number of EEG data groups belonging to the eye movement type in the sample library data group, m4 represents the number of EEG data groups in the sample library data group The number of EEG data sets belonging to the abnormal type of EEG, d _{1, m1} represents the distance between the waveform feature of the EEG data set to be tested and the waveform feature of the m1th one-blink type EEG data set , d _{2, m2} represent the distance between the waveform feature of the EEG data set to be tested and the waveform feature of the m2 second blink type EEG data set, d _{3, m3} represent the EEG data set to be tested The distance between the waveform feature of the data set and the waveform feature of the m3 eyeball movement type EEG data set, d _{4, m4} represents the waveform feature of the EEG data set to be tested and the m4th EEG abnormal type The distance between the waveform features of the EEG data set; finally, the values in the distance set D are sorted in ascending order to obtain a new set of distances, for example, the new set of distances obtained is D'={d _2,1 , d _1,1 ，d _2,4 ，…，d _2,m2 ，d _3,1 ，d _4,1 ，…，d _4,m4 }, take the first K distances {d _2,1 ，d _1,1 , d _2,4 ,..., d _2,m2 }, it is judged that the number of secondary blink types in the sample library data set corresponding to the first K distances is the largest, so this EEG data set belongs to the secondary blink type.

In a specific implementation of the present invention, the present invention uses an embedded AI algorithm to construct a sample library, the purpose of which is to quickly find the optimal sample library data set, and use the optimal sample library data set for EEG signals Feature classification to further improve the signal recognition rate.

As shown in Figure 4, the sample library construction process based on the embedded AI algorithm is as follows:

1. Construction constraints:

1.1) In order to increase the diversity of the sample library and avoid falling into local extremums, the data set of the sample library is formed by randomly generating a blink data set, a double blink data set, an eyeball movement data set, and an abnormal EEG data set.

s.t.rand (m1 blink data sets)

rand(m2 secondary blink data sets)

rand(m3 eyeball movement data sets)

rand (m4 abnormal EEG data sets)

1.2) Using the K-nearest neighbor algorithm and the distance function, the waveform features of the EEG data set are calculated, and the feature classification is judged. The K value parameter in the K nearest neighbor algorithm takes a smaller value or a larger value, which will lead to large deviations in the feature recognition rate and feature misjudgment rate. In order to obtain a suitable K value, the method of randomly generating the K value is adopted. Avoid getting stuck in local extrema.

s.t.rand (K)

1.3) Similarly, in order to avoid falling into local extremum, the training database data set is randomly generated.

s.t.rand (r training database data sets)

1.4) The K value should be less than the number of sample library data groups.

s.t.K<m

1.5) The number of sample database data groups and training database data groups should be less than the number of EEG data groups.

s.t.m<p

r<p

m+r<p

m1+m2+m3+m4=m

Second, build a multi-objective function:

The feature recognition rate and feature misjudgment rate are used to characterize the feature classification accuracy and misjudgment rate of the EEG data set. For the feature classification results of the EEG data set, the higher the feature recognition rate, the better, and the smaller the feature misjudgment rate, the better. Therefore, the multi-objective function is based on the maximum feature recognition rate and the minimum feature misjudgment rate.

Max{feature recognition rate}

Min{feature misjudgment rate}

3. Optimizing the sample library data set

The sample library construction model is a linear programming problem. The solution process of the sample library construction model is actually the solution process of the linear programming problem. In order to quickly and accurately solve the sample library construction model, the embedded AI algorithm is used for calculation.

Taking the basic genetic algorithm as an example, after population selection, crossover algorithm, and mutation algorithm, update the evolutionary cycle process of the current optimal chromosome. The basic steps are:

3.1) Randomly generate sample library data sets (first blink data, second eye blink data, eyeball movement data, EEG abnormal data), K value, training library data sets, and the remaining brainwave data sets are test library data sets (at p In each EEG data group, the remaining data groups except the sample database data group and the training database data group) form a data set, and carry out feasibility check;

3.2) Repeat 3.1) step to randomly generate POP_SIZE data sets, each data set includes sample database data set, K value, training database data set, test database data set;

3.3) In a data set, calculate the distance set between the waveform feature of a training library data set and the waveform feature of each sample library data set, adopt the K nearest neighbor algorithm to carry out feature classification to a training library data set, and so on, Carry out feature classification for each training library data group in the data set, and calculate the feature recognition rate and feature misjudgment rate of the data set;

3.4) Arrange the POP_SIZE data sets in ascending or descending order according to the feature recognition rate and feature misjudgment rate of the data sets, and exchange the corresponding data set content (sample library data group, K value), and save the current best data set ;

3.5) Use roulette selection, crossover algorithm, and mutation algorithm to update the data set to generate new POP_SIZE data sets;

3.6) Repeat steps 3.3) to 3.4) to generate the current best data set, compare it with the best data set of the previous generation, and save a better data set;

3.7) Steps 3.5) to 3.6) are repeated until the termination condition is satisfied, and the current best data set is output as the optimal solution, and the sample library data group in the current best data set is the optimal sample library;

3.8) In the current best data set, calculate the distance set between the waveform features of a test library data set and the waveform features of each sample library data set, and use the K nearest neighbor algorithm to carry out feature classification for a test library data set, with By analogy, feature classification is performed on each test library data group in the data set, and the feature recognition rate and feature misjudgment rate of the test library data group are calculated. If the goal is achieved, the feature recognition rate of the test library data set is above 85%, and the feature misjudgment rate is below 8‰, and the optimal sample library data set in the current best data set is feasible. If the goal is not achieved, repeat steps 3.1) to 3.8) to retrain the sample library data set until the optimal sample library that meets the target is output.

In this embodiment, the optimal sample library data set is obtained, and in the feature classification of the EEG data set, the feature recognition rate is further increased to over 85%, and the feature misjudgment rate is reduced to below 8‰.

When the automatic lifting hospital bed system of the present invention is in use, firstly, the brain wave sensor collects a certain length of patient's EEG fusion signal for noise filtering, extracts the waveform features, and compares the waveform features with the optimized sample library data group, The K-nearest neighbor algorithm is used to classify the waveform features to be recognized, and the control command is sent to the bed body according to the classification result, and the bed body responds according to the control command. The hospital bed control scheme is as follows: one blink corresponds to the raising of the hospital bed, two blinks corresponds to the leveling of the hospital bed, eye rotation corresponds to the active call, and EEG abnormality corresponds to the intelligent alarm.

The hospital bed system proposed by the present invention realizes automatic lifting, active calling and intelligent alarm by monitoring the EEG signal of the patient, adopts the K nearest neighbor method to identify the type of the EEG signal, and has a fast operation speed, which ensures the timely response of the hospital bed system; after optimization The sample library ensures the accuracy of the type recognition of EEG signals.

What are listed above are only specific embodiments of the present invention. Obviously, the present invention is not limited to the above embodiments, and many variations are possible. All deformations that can be directly derived or associated by those skilled in the art from the content disclosed in the present invention should be considered as the protection scope of the present invention.

Claims (5)

1. An automatic lifting sickbed system based on electroencephalogram signals is characterized by comprising an electroencephalogram sensor, a sickbed control system and a sickbed body, wherein the sickbed body has lifting, active calling and alarming functions;

the brain wave sensor is connected with the sickbed control system and is used for collecting brain electric eye electric fusion signals of a patient and transmitting the brain electric eye electric fusion signals to the sickbed control system; the sickbed control system sends a control instruction according to an electroencephalogram-eye electric fusion signal of a patient, and the control instruction is used for controlling the sickbed body to lift, level, actively call and alarm, and the sickbed body responds according to the control instruction;

the sickbed control system comprises:

the electroencephalogram electrooculogram fusion signal noise filtering module is used for carrying out noise filtering processing on the electroencephalogram electrooculogram fusion signal;

the electroencephalogram electrofusion signal segmentation window module is used for segmenting the electroencephalogram electrofusion signals subjected to noise filtering processing into different groups with consistent lengths;

the electroencephalogram electrofusion signal mode identification module is used for identifying the characteristic type of each group of electroencephalogram electrofusion signals, and sending an instruction according to the characteristic type, and comprises the following components:

s1, performing noise filtering processing on an electroencephalogram electrooculogram fusion signal to obtain noise-filtered electroencephalogram electrooculogram data;

s2, dividing the electroencephalogram data subjected to noise filtering into groups to obtain electroencephalogram data sets, and calculating waveform characteristics of each electroencephalogram data set, wherein the waveform characteristics of the electroencephalogram data sets comprise average values, absolute average values, mean square deviations and standard deviations;

s3, classifying waveform characteristics of each electroencephalogram ocular data set relative to waveform characteristics of the sample library data set by adopting a K nearest neighbor algorithm to obtain the belonging characteristic type of each electroencephalogram ocular data set; transmitting a control instruction according to the characteristic type of each electroencephalogram-eye electric data set;

the construction method of the sample library data set comprises the following steps:

s31, collecting electroencephalogram electrooculogram fusion signals of different patients under the conditions of primary blinking, secondary blinking, eyeball rotation and electroencephalogram abnormality through an electroencephalogram sensor, carrying out noise filtering processing on the electroencephalogram electrooculogram fusion signals, dividing the electroencephalogram electrooculogram fusion signals into groups, extracting waveform characteristics of each data group, and obtaining a database with total data quantity of p groups;

s32, randomly generating m sample library data sets, r training library data sets and S test library data sets from the database with the total data quantity of p groups; the sample library data set, the training library data set and the test library data set all comprise four types of data, namely primary blink, secondary blink, eyeball rotation and electroencephalogram abnormality, and the data are not repeated;

s33, repeating the step S32, initializing a plurality of data sets, wherein each data set is composed of a randomly initialized sample library data set, a training library data set, a test library data set and K value parameters of a K neighbor algorithm;

calculating a distance set between the waveform characteristics of each training library data set and the waveform characteristics of each sample library data set, classifying the characteristics of one training library data set by adopting a K nearest neighbor algorithm, constructing an objective function according to the recognition rate and the misjudgment rate, and selecting an optimal data set;

s34, taking the sample library data set in the optimal data set obtained in the step S33 as an optimal sample library data set, calculating a distance set between waveform characteristics of a test library data set in the optimal data set and waveform characteristics of each sample library data set, classifying the characteristics of one test library data set by adopting a K neighbor algorithm, constructing an objective function according to the recognition rate and the misjudgment rate, and judging whether the preset requirement is met;

if yes, the sample library data set in the optimal data set obtained in the step S33 is a final optimal sample library data set;

if not, returning to the step S32, and reinitializing a plurality of data sets.

2. The automatic lifting sickbed system based on the electroencephalogram signals as claimed in claim 1, wherein the electroencephalogram sensor is in wireless connection with a sickbed control system.

3. The automatic lifting sickbed system based on the electroencephalogram signals according to claim 1, wherein the characteristic types of each electroencephalogram data set comprise primary blinking, secondary blinking, eyeball rotation and electroencephalogram abnormality, and the corresponding control instructions are sickbed lifting, flattening, active calling and alarming respectively.

4. The automatic lifting sickbed system based on the electroencephalogram signals according to claim 1, wherein the step S3 comprises:

calculating the distance between the waveform characteristics of the electroencephalogram data set to be tested and each electroencephalogram data set in the sample base data set to obtain a distance set D= { D _1,1 ，d _1,2 ，…，d _1,m1 ，d _2,1 ，d _2,2 ，…，d _2,m2 ，…，d _4,1 ，d _4,2 ，…，d _4,m4 }, itWherein m1 represents the number of the electroencephalogram data groups belonging to the first blink type in the sample library data group, m2 represents the number of the electroencephalogram data groups belonging to the second blink type in the sample library data group, m3 represents the number of the electroencephalogram data groups belonging to the eyeball rotation type in the sample library data group, m4 represents the number of the electroencephalogram data groups belonging to the electroencephalogram abnormality type in the sample library data group, d _1,m1 Representing the distance between the waveform characteristics of the electroencephalogram data set to be tested and the waveform characteristics of the electroencephalogram data set of the m1 st blink type, d _2,m2 Representing the distance between the waveform characteristics of the electroencephalogram data set to be tested and the waveform characteristics of the electroencephalogram data set of the m < 2 > second blink type, d _3,m3 Representing the distance between the waveform characteristics of the electroencephalogram data set to be tested and the waveform characteristics of the electroencephalogram data set of the m3 rd eyeball rotation type, d _4,m4 Representing the distance between the waveform characteristics of the electroencephalogram data set to be tested and the waveform characteristics of the electroencephalogram data set of the m4 th electroencephalogram anomaly type;

and (3) carrying out ascending arrangement on the values in the distance set D, taking the first K distances, judging the number of four feature types in the sample library data set corresponding to the first K distances, and taking the feature type with the largest number as the feature type of the electroencephalogram data set to be tested.

5. The automatic lifting sickbed system based on the electroencephalogram signals according to claim 1, wherein in the step S33, a genetic algorithm is adopted to select, cross and mutate the data sets in the initialized data sets, and the obtained new data set is compared with the original data set to save a better data set.

Images