CN119418850A

CN119418850A - HICH intelligent rehabilitation health system based on multi-round complementation of image and text recognition

Info

Publication number: CN119418850A
Application number: CN202411458521.1A
Authority: CN
Inventors: 蒋理; 石全红; 程崇杰; 郭宗铎; 徐睿; 夏宇隆
Original assignee: Chongqing Medical University
Current assignee: Chongqing Medical University
Priority date: 2024-10-18
Filing date: 2024-10-18
Publication date: 2025-02-11
Anticipated expiration: 2044-10-18
Also published as: CN119418850B

Abstract

The invention relates to the field of HICH auxiliary nursing systems, in particular to an HICH intelligent rehabilitation health system based on image and text recognition multiple-round complementation. The scheme comprises an acquisition module, a map construction module and a decision module, wherein the acquisition module is used for acquiring electrophysiological data, imaging data and clinical text data of a past treatment event of a patient, integrating the electrophysiological data, the imaging data and the clinical text data into training data and integrating the training data into first data, the map construction module is used for identifying the imaging data and the electrophysiological data into text data, performing word identification and semantic extraction on the text data, determining entity and entity relations through multi-map identification of extracted contents, establishing an HICH rehabilitation knowledge map, performing word identification and semantic extraction on the clinical text data, and checking and complementing the HICH rehabilitation knowledge map by using the extracted contents, and the decision module comprises a treatment weight unit and a Gaussian model unit. The problem that HICH rehabilitation diagnosis and treatment results given by AI identification calculation are low in reliability is solved.

Description

HICH intelligent rehabilitation big health system based on image and text recognition multiple-round complementation

Technical Field

The invention relates to the field of HICH auxiliary nursing systems, in particular to an HICH intelligent rehabilitation health system based on image and text recognition multiple-round complementation.

Background

Hypertensive cerebral hemorrhage (HYPERTENSIVE CEREBRAL HEMORRHAGE, HICH) is a common neurological disorder with extremely high mortality and disability rate. Studies have shown that about 80% of HICH patients (referred to herein as "patients") remain with various disabilities and complications, which makes it necessary for a large proportion of patients to undergo long-term rehabilitation therapy, which is not only labor intensive, but also quite dramatic in terms of economic resource consumption.

However, most patients and families cannot bear such huge consumption, and can only choose to self-care and recover at home, and as the patients cannot be guided and supported in a professional way at home, repeated or progressive deterioration often occurs, the life quality of the patients is seriously affected, and even the patients possibly die directly. Under the condition of limited medical resources at present, the professionals required by rehabilitation therapy are short, and the period of cultivating the professionals for rehabilitation therapy is long, the investment is large and the effect is slow, so that the economic cost of rehabilitation therapy is high. With the further development of AI technology, more and more information can be rapidly processed by AI technology.

At present, a part of diseases can be assisted in diagnosis and treatment through an AI technology, and after the AI technology is trained, the region of a part of suspected lesions is marked or identified in an image diagnosis and treatment stage, so that the diagnosis and treatment efficiency of doctors is improved. But the advanced learning (AI) technology is utilized to analyze the illness state of a patient more accurately and individually, finish auxiliary diagnosis and treatment decisions, output relevant suggestions of medical rehabilitation schemes, and finally provide a more convenient and accurate intelligent rehabilitation treatment scheme for HICH patients.

The main difficulty is that the HICH needs to acquire imaging data and electrophysiological data of a patient in the rehabilitation process, including CT, MRI and PET image data, and electroencephalogram (EEG), electromyogram, evoked potential and brain oxygen saturation, the data are too complex, the data structure is also complex, the data are trained by using an independent AI model, the training data amount is limited in the early stage due to too many input parameters (variables), and great personalized differences exist among patients, so that the reliability of the HICH rehabilitation result given by AI identification calculation is low.

Disclosure of Invention

The purpose of this scheme is to provide the high health system of HICH intelligence rehabilitation based on image and text recognition is complementary many times to solve the problem that the HICH rehabilitation diagnosis and treatment result reliability that AI discernment calculation gave is lower.

In order to achieve the above purpose, the scheme provides an HICH intelligent rehabilitation big health system based on image and text recognition multiple-round complementation, which comprises:

the acquisition module is used for acquiring electrophysiological data, imaging data and clinical text data of past treatment events of a patient, integrating the electrophysiological data, the imaging data and the clinical text data into training data and integrating the training data into first data;

The system comprises a map construction module, a HICH rehabilitation knowledge map, a clinical text data acquisition module, a HICH rehabilitation knowledge map acquisition module and a data analysis module, wherein the map construction module is used for recognizing the imaging data and the electrophysiological data into text data, performing text recognition and semantic extraction on the text data, determining entity and entity relation through multiple maps of extracted contents, and performing text recognition and semantic extraction on the clinical text data, and checking and complementing the HICH rehabilitation knowledge map by using the extracted contents;

The decision module comprises a treatment weight unit and a Gaussian model unit, wherein the treatment weight unit is a calculation model constructed based on TransG-DNN, the treatment weight unit combines an HICH rehabilitation guide and actual conditions in the consensus and rehabilitation treatment process, the Gaussian model unit is a Gaussian mixture model constructed based on an HICH rehabilitation knowledge map, and the decision module is used for putting first data into the Gaussian mixture model to calculate illness state information, and then putting the illness state information into the treatment weight unit to calculate a treatment scheme and treatment expectation.

The principle and the effect of the scheme are that the scheme performs entity extraction and mapping of entity relationship on electrophysiological data and imaging data of a patient, combines the extraction and mapping results, and performs multiple-round complementation. The treatment weight unit of the scheme is constructed by combining the latest domestic and foreign HICH rehabilitation guidelines and consensus with the actual conditions in the rehabilitation treatment process by the cooperation of special doctors such as neurosurgery, neurology, rehabilitation, radiology, nuclear medicine, electrophysiology center, psychological sanitation center and the like and AI researchers, so that the treatment weight unit can output reliable rehabilitation diagnosis and treatment results through the illness state. Meanwhile, the knowledge graph has the advantage of connecting structured and unstructured data, and the personalized HICH rehabilitation knowledge graph is constructed based on the actual condition of a patient, so that the problem that the structure of the imaging data and the electrophysiological data is complex in the scheme is solved, and a fine and powerful basic technical support is provided for the analysis of the complicated imaging data and electrophysiological data. According to the scheme, a Gaussian mixture model established based on the HICH rehabilitation knowledge graph is used as a basis, each triplet in the HICH rehabilitation knowledge graph is vectorized by utilizing a Gaussian mixture embedding model (TransG) to finish preliminary classification, the DNN training model is used on the basis, after the optimal training model is obtained, the obtained optimal training model is applied to training data, the classification effect and stability of the HICH rehabilitation decision model are realized, automatic HICH rehabilitation decision with high pre-judgment accuracy is finally finished, and the problems of numerous data and excessive input parameters (variables) are solved.

In conclusion, the problem that the HICH rehabilitation diagnosis and treatment result provided by AI identification calculation is low in reliability is solved.

The acquisition module further comprises an identification unit, wherein the identification unit is a calculation model based on U-net++ deep learning and a neural network, when the acquisition unit acquires multi-mode MRI and PET images in the imaging data, the identification unit is mainly used for automatically identifying the skull MRI and PET images in the rehabilitation process of a patient, and the identification unit is used for evaluating the integrity degree of brain parenchyma and white matter fiber bundles, the cerebral blood flow perfusion condition and the brain function retention condition of the patient and finally outputting a final result of automatically judging the MRI and PET images in a Chinese text form; the identification unit uses a method in CT image identification to carry out side, positioning and quantitative identification on MRI and PET image data, adds a Dense block and a convolution layer between an encoder and a decoder by using U-net++, adds a redesigned jump path on the basis of the original U-net++, fuses the output of a previous convolution layer of the same Dense block with the sampling output corresponding to a lower Dense block, enables the semantic level of the coded features to be closer to the semantic level of the feature mapping to be carried out in the decoder, adopts Dense jump connection to realize the jump path between the encoder and the decoder so as to ensure that all prior feature images are accumulated, and generates a feature map with complete resolution at a plurality of semantic levels through a Dense jump block on each jump path to improve the segmentation precision and the gradient flow, and in addition, the identification unit is used for increasing the depth supervision model to adjust the complexity of the model, adjust the balance between the calculation reasoning speed and the performance, carries out side, positioning and quantitative identification on the MRI and the image data, and taking the identification result obtained by one of the MRI and PET image data as a reference result, obtaining the identification obtained by other MRI and PET image data acquired by the same examination as a complementary result, sequentially verifying and comparing a plurality of identification results with the reference result, carrying out multi-round complementation on the reference result, and outputting the multi-round complementation result of the reference result as a final identification result.

When the acquisition unit acquires electrophysiological data, the identification unit carries out smoothing treatment on spike signals in a half-wave treatment mode, a main component analysis algorithm is adopted to carry out wavelet decomposition on the electrophysiological data to obtain signal components, then the main component analysis algorithm is applied to the signal components, and an independent component analysis algorithm is adopted to find a mixed signal matrix of artifact signals and electroencephalogram signals, so that various artifacts are separated from the mixed composite signals, the identification unit determines the threshold value of abnormal waves through threshold initial screening, if the amplitude, frequency or area characteristics under an effect curve break through the determined abnormal wave threshold value, the abnormal waves are determined, the identification unit adopts expert feature analysis, carries out initial screening according to the characteristics of various typical interference waveforms in the past, removes eye movement artifacts, blink artifacts and electrode artifacts, simultaneously uses a band-pass filter to carry out filtering treatment on the electroencephalogram signals in a useful frequency band, filters various interference signals outside the frequency band, carries out one-step treatment on the electrophysiological data by adopting a simple scaling method, stores and labels, then carries out automatic scaling treatment on the abnormal waves, and finally carries out a text-based on the data, and finally carries out a sparse-to-cycle analysis on the data, and finally carries out a data input-to a sparse-to a physiological data input-sparse-cycle data input-based on the obtained data, so as to obtain a final physiological data-cycle data-based on a sparse-phase-sparse-and-cycle data-input data-output data-input-sparse-cycle data-cycle analysis result, and a final-cycle data-input method.

Further, when the acquisition unit acquires the CT image, the identification unit divides and extracts corresponding characteristics of a cerebral parenchymal area hemorrhagic focus, edema brain tissue, normal brain tissue, each ventricle and an important brain pool in the brain window level of the skull CT through U-net++, automatic identification of the skull CT image of the HICH patient is completed, the automatic identification comprises side fixing, positioning and quantitative identification, and a final result of automatic interpretation of the CT image is finally output in a Chinese text form.

The identification unit can automatically judge the imaging data and the electrophysiological data, so that the workload of medical staff can be reduced, and a huge calculation basis can be rapidly provided for a spectrogram construction module and a decision module, so that the output speed and the output efficiency of the treatment scheme of the scheme are improved. Meanwhile, the recognition unit uses different discrimination modes for different examination data, and can output the final result of automatic interpretation more accurately, so that the reliability of the HICH rehabilitation diagnosis and treatment result given by the scheme is further improved. The recognition unit outputs the final result of automatic interpretation in the form of text, thereby facilitating the medical staff to check and change the final result, unifying the storage and reading modes of electrophysiological data, imaging data and clinical text data, and facilitating the reading and use of the atlas construction module and the decision module.

The device further comprises an analysis unit, wherein the analysis unit is used for comparing the final result output by the identification unit with the result obtained by the doctor and judging the accuracy, the specificity and the sensitivity of the identification unit, and the analysis unit is also used for comparing the number of the final results output by the identification unit with the doctor in the same unit time and simultaneously calculating and comparing the accuracy of the final results output by the identification unit and the doctor.

The judgment of doctors and the final result output by the identification unit are compared through the analysis unit, more accurate training data can be provided for the identification unit, the accuracy, the specificity and the sensitivity of the identification unit are improved, and the reliability of the HICH rehabilitation diagnosis and treatment result given by the scheme is further improved.

The medical care pillow is used for collecting cerebral oxygen saturation data of a patient, transmitting the cerebral oxygen saturation data as recovery collection data to the patient terminal, receiving and displaying the recovery collection data, transmitting the recovery collection data as electrophysiological data to the collection module, receiving and displaying the data transmitted by the decision module, and reading the collection data, receiving the data and inputting clinical text data to the collection terminal.

The rehabilitation pillow, the medical care terminal and the patient terminal are added, the rehabilitation pillow is used for providing reliable monitoring for patients, the patients can know the body condition of the patients through the rehabilitation pillow, the patients can also receive the reliable treatment scheme of the output of the decision making module through the patient terminal, and the patients can be provided with the treatment scheme of the current symptoms of the patients more in time when collecting the physiological data of the patients conveniently. Medical care terminal also can in time know patient's disease and recovery situation through the collection module, makes things convenient for medical staff to pass through medical care terminal input and more laminate the clinical text data of patient's current disease, further improves the HICH rehabilitation diagnosis and treatment result reliability that this scheme gave.

The recovery pillow comprises a shell and a plurality of near infrared light probes, wherein the near infrared light probes are embedded in the surface of the shell, a plurality of pressure sensors are further embedded in the surface of the shell, a temperature adjusting device is embedded in the shell, a power supply device and a processing device are fixedly connected in the shell, the processing device is in communication connection with the near infrared light probes, the pressure sensors and the temperature adjusting device, the power supply device is electrically connected with the near infrared light probes, the pressure sensors, the temperature adjusting device and the processing device, the near infrared light probes are used for collecting brain oxygen saturation data of a patient and transmitting the brain oxygen saturation data to the processing device, the pressure sensors are used for collecting pressure values received by all points on the recovery pillow and transmitting the pressure values to the processing device, the temperature adjusting device is used for obtaining the temperature of the shell and transmitting the temperature values to the processing device, the processing device is used for receiving brain oxygen saturation data, the pressure values and the shell temperature, analyzing the bearing on the recovery pillow according to the pressure values and the pressed point positions of the pressure values, and removing interference data in the brain oxygen saturation data through the bearing, so that brain oxygen saturation data are obtained, and finally the brain oxygen saturation data are transmitted to a patient terminal as physiological data.

The living environment of the patient has more factors for disturbing the acquisition of the cerebral oxygen saturation data, and the acquisition of the cerebral oxygen saturation data during sleeping of the patient does not influence the daily life of the patient, and can acquire more accurate cerebral oxygen saturation data. The pressure sensor can assist in determining the acquisition object of the rehabilitation pillow and exclude the data which are received by interference in the process of acquiring the cerebral oxygen saturation data.

Further, the plurality of pressure sensors are uniformly distributed on the shell, the processing device stores the received cerebral oxygen saturation data, the pressure value and the shell temperature into the cache, when the processing device is used for eliminating interference data in the cerebral oxygen saturation data, a three-dimensional model of a carrier is established through the pressure value and a pressed point for providing the pressure value as a temporary model, a reference model is preset in the processing module, the reference model is a three-dimensional model of the head and the neck of a patient which is input in advance, the processing module is used for comparing the temporary model with the reference model to obtain similarity, when the similarity is smaller than 90%, the cerebral oxygen saturation data which is acquired simultaneously with the pressure value is used as the interference data, and the pressure value and the shell temperature which is acquired simultaneously with the pressure value are simultaneously deleted, when the similarity is larger than or equal to 90%, the median in the cerebral oxygen saturation data which is acquired simultaneously with the pressure value is selected as the cerebral oxygen saturation acquisition data, when the similarity is between 90% and 97%, the cerebral oxygen saturation acquisition data are corrected through the temporary model, and the cerebral oxygen saturation acquisition data are transmitted to the patient terminal as the physiological data, and other cache data outside the temporary model are emptied.

The three-dimensional model of the carrier and the head and neck of the patient is established and compared, and the patient and the non-patient object can be accurately distinguished, so that the interference of the environment on the brain oxygen saturation data during acquisition is judged and eliminated, and the effect of accurate acquisition is achieved. Meanwhile, the state of the patient when the brain oxygen saturation data are acquired can be obtained through the temporary model, so that the interference of the patient on the acquisition process is judged, the brain oxygen saturation acquisition data are corrected according to the current state (self condition) of the patient, and the effect of accurate acquisition is further achieved.

Further, the temperature adjusting device is further used for adjusting the temperature of the shell, the processing device is preset with the highest cerebral oxygen saturation, the lowest cerebral oxygen saturation, the single waiting time and the maximum waiting time, and the processing device executes the following steps after sending the cerebral oxygen saturation acquisition data as electrophysiological data to the patient terminal and before emptying other data in the buffer memory:

s10, calculating a difference value between the brain oxygen saturation acquired data and the highest brain oxygen saturation;

s20a, if the difference value is continuously increased after three times of waiting, a first instruction is sent to a patient terminal;

S20b, if the continuous waiting for a plurality of times is carried out, and the accumulated waiting time for a plurality of times exceeds the maximum waiting time, a second instruction is sent to the patient terminal;

s20c, if the brain oxygen saturation acquired data is lower than the minimum brain oxygen saturation, a third instruction is sent to the patient terminal;

S20d, if the difference value is not less than 0 and the last difference value is less than 0, controlling the temperature regulating device to be started, controlling the temperature of the shell to be maintained at 34 ℃, and returning to S10 after counting down the single waiting time;

S20e, if the difference value is smaller than 0 and the last difference value is not smaller than 0, closing the temperature regulating device, and returning to S10;

S20f, if the difference value calculated in the last two times is not smaller than 0 and the difference value is larger than the last difference value, controlling the temperature of the shell to be maintained at 32 ℃, and returning to S10 after counting down the single waiting time;

S20g, if the difference value calculated in the last two times is not smaller than 0 and is not larger than the last difference value, returning to S10 after counting down the single waiting time;

The method comprises the steps that a first prompt, a second prompt, a third prompt and closing time are preset in a patient terminal, the first prompt, the second prompt or the third prompt are immediately displayed after the patient terminal receives the first instruction, the second prompt or the third prompt, the first prompt, the second prompt or the third prompt are circularly broadcast, an alarm signal is sent to an acquisition terminal if the time of the circular broadcast exceeds the closing time, the patient terminal does not receive the first instruction, the second instruction and the third instruction in the closing time if the circular broadcast is closed in the closing time, the delay time is selected after the circular broadcast is closed, the first instruction, the second instruction and the third instruction are not received in the delay time, and the alarm signal is sent to the acquisition terminal if the circular broadcast is closed in the closing time, and the alarm signal is selected after the circular broadcast is closed.

According to the scheme, on the basis of accurately collecting the cerebral oxygen saturation data, when abnormal data (when the cerebral oxygen saturation is too high) are detected, multiple judgments are made on the cerebral oxygen saturation data of a patient, judgment results are verified in a mode of reducing the contact temperature in the multiple judgments, misjudgement conditions can be reduced, the contact temperature of the brain and the neck of the patient can be reduced, the cerebral oxygen saturation of the patient can be reduced to a certain extent, sudden accidents of the patient are reduced, or more rescuing time is strived for the patient.

The system and the method have the advantages that the system and the method interact with the patient through the display and broadcasting mode of the patient terminal, the safety condition of the patient is further determined, if the patient does not respond, the patient is very likely to be in an isolated state without assistance and without ability response (such as a coma state, a speech disabled state, a movement disabled state and the like), an alarm signal is sent to the acquisition terminal, help can be provided for the patient in time, and more rescuing time is provided for the patient. If the patient responds and the delay time is selected, the patient is in a safe state, and the patient can wake up, so that the judgment of the patient is that the recovery pillow is wrong, and the recovery pillow is in a damaged state at the moment, and the patient terminal does not receive the first instruction, the second instruction and the third instruction, so that the patient can keep silent sleep. If the patient responds and selects to send the alarm signal, the alarm signal is sent to the acquisition terminal at the moment, so that help can be provided for the patient in time.

The temperature regulating device is further used for sending the acquired shell temperature and regulating record to the patient terminal after being started, the patient terminal receives the shell temperature and regulating record, the shell temperature and regulating record correspond to brain oxygen saturation acquisition data according to acquisition time and are converted into clinical text data and are sent to the acquisition module, the acquisition module receives an alarm signal, the alarm signal sending time is used as an accident time node, electrophysiological data, clinical text data and alarm signals before and after the accident time node are sent to the medical care terminal and the decision module, the medical care terminal sends emergency prompt sound after receiving the alarm signal and displays the received electrophysiological data and clinical text data, the decision module receives the alarm signal, marks the received electrophysiological data and clinical text data as the highest priority, calculates an emergency treatment scheme by using the electrophysiological data and the clinical text data, sends the emergency treatment scheme to the patient terminal, and plays the emergency treatment scheme after the patient terminal receives the emergency treatment scheme.

When the acquisition module receives the alarm signal, the patient is in the condition that needs help, and this scheme can provide the data of laminating patient's pathology the most for medical personnel this moment, lets medical personnel and decision-making unit provide the most professional help for the patient, further strives for time and the opportunity of rescuing for the patient. Meanwhile, the emergency treatment scheme is played, professional first-aid guidance can be provided for the patient and caregivers possibly existing around the patient, and the speed of the patient in danger separation is increased.

Drawings

Fig. 1 is a logic block diagram of embodiment 1 of the present solution.

Fig. 2 is a logic block diagram of embodiment 2 of the present solution.

Fig. 3 is a schematic structural diagram of the rehabilitation pillow according to embodiment 2.

The following is a further detailed description of the embodiments:

The reference numerals in the drawings of the specification comprise 1, a light hole, 2, a pressure sensor, 3, a shell, 4, a power supply device, 5, a processing device, 6, a near infrared light probe, 7 and a temperature regulating device.

Detailed Description

The conception and the technical effects produced by the present invention will be clearly and completely described in conjunction with the embodiments below to fully understand the objects, features and effects of the present invention. It is apparent that the described embodiments are only some embodiments of the present invention, but not all embodiments, and that other embodiments obtained by those skilled in the art without inventive effort are within the scope of the present invention based on the embodiments of the present invention:

Example 1:

multiple rounds of complementary HICH intelligent rehabilitation and wellness systems (as shown in fig. 1) based on image and text recognition, comprising:

the acquisition module is used for acquiring electrophysiological data, imaging data and clinical text data of past treatment events of a patient to be integrated into training data, and also used for acquiring electrophysiological data, imaging data and clinical text data of current rehabilitation treatment of the patient to be integrated into first data;

The map construction module is used for recognizing the imaging data and the electrophysiological data into text data, performing text recognition and semantic extraction on the text data and the clinical text data, determining entity and entity relation through the multiple image certificates of the extracted content, and establishing an HICH rehabilitation knowledge map;

The acquisition module also comprises a recognition unit, the recognition unit is based on a calculation model of U-net++ deep learning and a neural network, when the acquisition unit acquires multi-mode MRI and PET images in image data, the recognition unit is mainly used for automatically recognizing the head MRI and PET images in the rehabilitation process of a patient, the recognition unit is used for evaluating the brain parenchyma and white matter fiber bundle integrity degree, the cerebral blood flow perfusion condition and the brain function retention condition of the patient and outputting final results for automatically judging the MRI and PET images in a form of Chinese text finally, the recognition unit uses a method in CT image recognition to fix, position and quantitatively recognize MRI and PET image data, the recognition unit uses U-net++ to add a Dense block and a convolution layer between an encoder and a decoder, on the basis of original U-net++, a redesigned jump path is added, the semantic level of the former convolution layer is fused with the sampling output corresponding to a lower layer, the semantic level of the coded features is enabled to be more approximate to the features to be needed in a decoder, simultaneously, the recognition unit is used for realizing the fact that the gradient map is connected with the gradient map between the encoder and the jump block, the gradient map is used for adjusting the overall gradient map and the gradient map is adjusted by the comparison between the two-level and the current gradient map, the recognition unit is used for adjusting the overall gradient map and the gradient map is adjusted by the calculation model, and acquiring identification obtained by other MRI and PET image data acquired by the same examination as a complementary result, sequentially verifying and comparing a plurality of identification results with a reference result, performing multiple rounds of complementation on the reference result, and outputting the multiple rounds of complementation result of the reference result as a final identification result.

When the acquisition unit acquires electrophysiological data, the recognition unit carries out smoothing treatment on spike signals in a half-wave treatment mode, carries out wavelet decomposition on the electrophysiological data by adopting a principal component analysis algorithm to obtain signal components, then applies a principal component analysis algorithm to the signal components, and searches a mixed signal matrix of artifact signals and electroencephalograms by adopting an independent component analysis algorithm, thereby separating various artifacts from mixed composite signals, the recognition unit determines the threshold value of abnormal waves through threshold initial screening, if the amplitude, frequency or area characteristics under an effect curve break through the determined abnormal wave threshold value, the abnormal waves are judged, the recognition unit adopts expert feature analysis, carries out primary screening according to the characteristics of various typical interference waveforms in the past, removes eye movement artifacts, artifacts and electrode artifacts, simultaneously completes filtering treatment on the electroencephalograms in a useful frequency band by using a band-pass filter, filters various interference signals outside the frequency band, carries out normalization treatment on the electrophysiological data by adopting a simple method, stores the segmented storage treatment and the labeling treatment, then carries out automatic sparse-to the pre-treated automatic sparse-electronic-encoder, carries out the final memory cycle data according to the text-based on the obtained data, and finally carries out the data-scaling analysis on the data of the final neural-cycle data, thus finally carries out the data-cycle-based on the data, and finally-cycle-based data-scaling analysis result of the data, thus finally carries out the final-cycle-based data.

When the acquisition unit acquires CT images, the identification unit divides cerebral parenchymal area hemorrhagic stoves, edema brain tissues, normal brain tissues, each ventricle and an important brain pool in the brain CT window level of the skull through U-net++, extracts corresponding characteristics of the brain parenchymal area hemorrhagic stoves, edema brain tissues, normal brain tissues, each ventricle and the important brain pool, completes automatic identification of the skull CT images of the HICH patient, comprises side fixing, positioning and quantitative identification, and finally outputs a final result of automatic interpretation of the CT images in a Chinese text mode.

When the identification unit identifies and interprets the electrophysiological data and the CT image, one of the acquired data (the electrophysiological data and the CT image) is used as reference data, other data acquired at the same time is used as complement data, the complement data is used for sequentially comparing and verifying the interpretation results of the reference data, and the interpretation results after multiple rounds of complementation of the reference data are used as final interpretation results to be output.

The device also comprises an analysis unit, wherein the analysis unit is used for comparing the final result output by the identification unit with the result obtained by the doctor and judging the accuracy, the specificity and the sensitivity of the identification unit, and the analysis unit is also used for comparing the number of the final results output by the identification unit with the doctor in the same unit time and simultaneously calculating and comparing the accuracy of the final results output by the identification unit and the doctor.

The analysis unit takes the first interpretation result output by the identification unit as a reference result, takes the interpretation result output after the first interpretation result as a comparison result, uses the comparison result to carry out multi-round complementation on the reference result, and then takes clinical text data (data recorded by medical staff) as the comparison result to carry out multi-round complementation on the reference result again, wherein the interpretation result after the multi-round complementation of the reference result is used for constructing a map.

In the case of a specific implementation of the method,

The acquisition unit acquires electrophysiological data, imaging data and clinical text data of past treatment events of a patient to be integrated into training data, and acquires electrophysiological data, imaging data and clinical text data of current rehabilitation treatment of the patient to be integrated into first data.

After the acquisition unit acquires image data (namely image data) such as CT, MRI and PET of the HICH patient, the image data is stored in a DICOM format in the system of the scheme, the identification and labeling of the image data are completed through the identification unit, the image data are subjected to denoising treatment, related images are labeled, a deep neural network model row is constructed for image segmentation, image characteristic information is extracted, automatic identification of the multi-mode image data of the HICH patient is realized, and finally, the final result of automatic interpretation is stored in training data and first data in a Chinese text form.

After the acquisition unit acquires electrophysiological data and cerebral oxygen metabolism data (including electroencephalogram (EEG), electromyogram, evoked potential, cerebral oxygen saturation and the like) of an HICH patient in the rehabilitation process, the identification unit extracts power spectral density of the electrophysiological data (including EEG, electromyogram, evoked potential and the like), finishes classification of a cerebral oxygen saturation data set, converts one-dimensional signals in the data into two-dimensional time-frequency domain signals, removes partial noise and artifacts, then performs feature extraction and classification on the electrophysiological data and the cerebral oxygen saturation data, finishes feature identification, and finally stores a final result of automatic interpretation in a Chinese text form into training data and first data.

The acquisition unit acquires clinical text data of the HICH patient in the treatment and rehabilitation process (including Electronic Medical Record (EMR) during hospitalization in the acute stage after the illness and recorded rehabilitation related text data during rehabilitation treatment), and classifies the clinical text data through text information rules preset in the system.

The atlas construction module carries out pretreatment on training data and first data through natural language processing technologies such as word segmentation, syntactic analysis, information extraction, word sense disambiguation, automatic abstract and the like, solves the problems of text classification and keyword extraction of the training data and the first data, semantic disambiguation and the like, and obtains a pretreated Chinese medical information character string related to HICH rehabilitation. Based on various pre-established HICH rehabilitation related dictionaries, the acquired Chinese medical information character strings are divided into character strings with independent semantics, and analysis of different dimensionalities is carried out. And matching the character strings subjected to the analysis of different dimensions in the respective analysis dimensions by a mechanical word segmentation method.

And determining the HICH rehabilitation knowledge graph as a directed label graph consisting of HICH rehabilitation related entities and relations thereof, and connecting various entities through relations by constructing an entity-relation-entity triple structure related to HICH rehabilitation in a form of a main guest, thereby forming a netlike knowledge graph. Here, we denote an entity (i.e. "HICH rehabilitation related entity") by a node, and by a relationship (i.e. a medical event relationship between TBI diagnosis and treatment related entities) by an edge, and define as follows.

I. The HICH rehabilitation related entity is set as H, and is defined as various medical entities which can be uniquely identified in the HICH rehabilitation related text information record and are related to HICH rehabilitation.

Setting the HICH rehabilitation medical event relationship as R, which is defined as medical event relationship generated or existing between different HICH rehabilitation related entities.

Setting the HICH rehabilitation knowledge graph as K, which is defined as a directed label graph. And R is defined as an edge set of the knowledge graph, and represents medical event connection generated or existing among all HICH rehabilitation related entities.

The related entities in the HICH rehabilitation process are defined and classified, and the entities to be identified are classified into the types of ' patients, basic information, symptoms, signs, auxiliary examination results, examination schemes, prognosis conditions, rehabilitation schemes ', and the like '. Then, a two-way long-short-term memory network model conditional random field (Bidirectional long short term memory-Conditional random fields, biLSTM-CRF) is adopted, a corpus is constructed, a model is generated, and the model is applied to complete entity extraction (or named entity identification).

The HICH rehabilitation related medical event relationship in the HICH patient EMR is classified and the entity relationship to be identified is classified into the types of 'belongingto relationship (belong_to), instance relationship (instance_of), attribute relationship (attribute_of), diagnosis relationship (diagnose), examination relationship (check), treatment relationship (treatment)', and the like. Meanwhile, since the attribute extraction can be regarded as relationship extraction essentially, the "HICH rehabilitation related medical event relationship" extraction in the present project will include entity relationship extraction and attribute extraction. And combining the kernel functions corresponding to the feature-based relation extraction method and the tree kernel function-based method to form a feature-tree kernel function combined method, so as to finish HICH rehabilitation related medical event relation extraction.

And then, performing the steps of entity disambiguation, reference resolution, knowledge combination and the like to complete TBI related knowledge fusion management.

And finally, performing ontology construction and extraction, knowledge reasoning and quality assessment to complete TBI diagnosis and treatment related knowledge processing management.

The decision-making module is firstly used for jointly cooperating with AI researchers by special doctors such as neurosurgery, neurology, rehabilitation, radiology, nuclear medicine, electrophysiology center, psychological sanitation center and the like, and initially constructing a treatment weight unit for emergency diagnosis and emergency treatment decision of an HICH patient according to the latest domestic and foreign HICH rehabilitation guidelines and consensus and combining actual conditions and AI algorithm requirements in the rehabilitation treatment process. Meanwhile, a Gaussian model unit based on TransG-DNN is constructed, so that automatic decision of HICH rehabilitation is realized, wherein the automatic decision has high prejudgement accuracy. Based on the constructed HICH rehabilitation knowledge graph, vectorizing each triplet in the HICH rehabilitation knowledge graph by utilizing a Gaussian mixture embedding model (TransG) in a Gaussian model unit to finish preliminary classification, training a model (using training data) by DNN on the basis, and applying the obtained optimal training model to first data after obtaining the optimal training model to realize the classification effect and stability of an HICH rehabilitation decision model, and finally finishing the HICH rehabilitation automatic decision with high pre-judgment accuracy.

Treatment expectations for HICH patients include predictive assessment (e.g., ICH improvement and grading scale, new ICH scale, ICH function return scoring scale, function outcome scale), and actual rehabilitation assessment (e.g., mRS, GOS, consciousness and mental state assessment, intelligent assessment, motor function assessment, balance ability assessment, walking function assessment, daily life action assessment). The treatment scheme comprises formulating rehabilitation targets (near term and long term rehabilitation targets), pharmaceutical treatment, physical treatment, daily life activity training, sports treatment, psychological treatment, nerve regulation treatment, rehabilitation training based on brain-computer interfaces, traditional Chinese medicine treatment and the like.

The analysis unit judges the accuracy, the specificity and the sensitivity of the identification unit:

the method is used for completing accuracy analysis of automatic identification of the multi-mode imaging image based on U-Net++. The multi-mode imaging automatic interpretation result output by the scheme is compared with the result obtained by a doctor, and the accuracy, the specificity and the sensitivity of the imaging result output by the scheme are judged.

The method is used for completing accuracy analysis of automatic identification of electrophysiological and brain oxygen metabolism data based on neural network and deep learning. The electrophysiological and cerebral oxygen metabolism data automatic interpretation result output by the scheme is compared with the result obtained by a doctor, and the accuracy, the specificity and the sensitivity of the electrophysiological and cerebral oxygen metabolism result output by the scheme are judged.

And the accuracy analysis is used for completing rehabilitation decision based on knowledge graph reasoning. And comparing the therapeutic scheme output by the scheme with a result obtained by a doctor, and judging the accuracy, the specificity and the sensitivity of the rehabilitation decision output by the scheme.

The method is used for completing the efficiency analysis of the automatic identification of the multi-mode imaging image based on U-Net++. In the same unit time, the quantity of the multi-mode imaging images processed by the scheme is compared with that of doctors, and the accuracy of the two images is calculated and compared.

The method is used for completing the efficiency analysis of the automatic identification of electrophysiological and brain oxygen metabolism data based on the neural network and deep learning. In the same unit time, the quantity of the electrophysiological and cerebral oxygen metabolism data processed by the scheme is compared with that of a doctor, and the accuracy of the electrophysiological and cerebral oxygen metabolism data and that of the doctor are calculated and compared simultaneously.

For performing an efficiency analysis of a knowledge-graph inference based treatment regimen. In the same unit time, the number of the processing steps for completion is compared with that of doctors, and the accuracy of the two steps is calculated and compared.

Example 2

As shown in fig. 2, the multi-round complementary HICH intelligent rehabilitation big health system based on image and text recognition further comprises a rehabilitation pillow, a medical care terminal and a patient terminal, wherein the rehabilitation pillow is in communication connection with the patient terminal, the medical care terminal is in communication connection with the acquisition module and the decision module, the patient terminal is in communication connection with the acquisition module and the decision module, the rehabilitation pillow is used for acquiring brain oxygen saturation data of a patient and transmitting the brain oxygen saturation data as rehabilitation acquisition data to the patient terminal, the patient terminal is used for receiving and displaying the rehabilitation acquisition data and transmitting the rehabilitation acquisition data as electrophysiological data to the acquisition module, the patient terminal is also used for receiving and displaying the data transmitted by the decision module, and the medical care terminal is used for reading the acquisition data, receiving the data and inputting clinical text data to the acquisition terminal.

As shown in fig. 3, the rehabilitation pillow comprises a shell 3 and a plurality of near infrared light probes 6, wherein a plurality of light holes 1 which are arranged in a matrix are arranged on the surface of the shell 3, which is contacted with a patient, the near infrared light probes 6 are fixedly adhered to the bottom of the light holes 1, the near infrared light probes 6 are embedded in the surface of the shell 3, a plurality of pressure sensors 2 are further embedded in the surface of the shell 3, a temperature adjusting device 7 is embedded in the shell 3, a power supply device 4 and a processing device 5 are further fixedly adhered in the shell 3, and flexible and fluffy materials such as cotton, polyester fibers, foam and the like are further filled between the shell 3 and the power supply device 4 to improve the comfort level of use. The processing device 5 is in communication connection with the near infrared light probe 6, the pressure sensor 2 and the temperature regulating device 7, the power supply device 4 is electrically connected with the near infrared light probe 6, the pressure sensor 2, the temperature regulating device 7 and the processing device 5, the near infrared light probe 6 is used for collecting brain oxygen saturation data of a patient and transmitting the brain oxygen saturation data to the processing device 5, the pressure sensor 2 is used for collecting pressure values received by all points on a rehabilitation pillow and transmitting the pressure values to the processing device 5, the temperature regulating device 7 is used for obtaining the temperature of the shell 3 and transmitting the temperature to the processing device 5, the processing device 5 is used for receiving the brain oxygen saturation data, the pressure values and the temperature of the shell 3, analyzing a bearing object on the rehabilitation pillow according to the pressure values and the pressure points providing the pressure values, removing interference data in the brain oxygen saturation data through the bearing object, thus obtaining brain oxygen saturation collection data, and finally transmitting the brain oxygen saturation collection data to a patient terminal as electrophysiological data.

As shown in fig. 3, the plurality of pressure sensors 2 are uniformly distributed on the housing 3, the processing device 5 stores the received brain oxygen saturation data, the pressure value and the temperature of the housing 3 into the cache, when the processing device 5 is used for eliminating the interference data in the brain oxygen saturation data, a three-dimensional model of a carrier is established through the pressure value and the pressed point for providing the pressure value as a temporary model, a reference model is preset in the processing module, the reference model is a three-dimensional model of the head and the neck of a patient which is input in advance, the processing module is used for comparing the temporary model with the reference model to obtain similarity, when the similarity is smaller than 90%, the brain oxygen saturation data which is acquired simultaneously with the pressure value is used as the interference data, and the pressure value and the temperature of the housing 3 which is acquired simultaneously with the pressure value are simultaneously deleted, when the similarity is larger than or equal to 90%, the median of the brain oxygen saturation data which is acquired simultaneously with the pressure value is selected as brain oxygen saturation acquisition data, when the similarity is between 90% and 97%, the brain oxygen saturation acquisition data are corrected through the temporary model, and the brain oxygen saturation acquisition data are transmitted to the patient terminal as physiological data, and other data outside the temporary model are emptied.

When the similarity between the temporary model and the reference model is between 90% and 97%, the shape of the patient is slightly changed, and at the moment, the temporary model is used for complementing the correction reference model to enable the reference model to be more fit with the current real-time condition of the patient, so that the real data which is most suitable for the real-time condition of the patient can be used for complementing the historical data in such a way, the historical data can be updated, and the identification accuracy of the rehabilitation pillow to the patient (the patient) is maintained or even improved under the condition of multiple updating (namely, the real-time data is used for complementing the historical data for multiple times), the occurrence of misjudgment conditions is reduced, and the use experience of the patient is improved.

The temperature adjusting device 7 is further configured to adjust the temperature of the housing 3, and the processing device 5 is preset with a maximum cerebral oxygen saturation, a minimum cerebral oxygen saturation, a single waiting time and a maximum waiting time, and after the cerebral oxygen saturation acquired data is sent to the patient terminal as electrophysiological data, the processing device 5 performs the following steps before emptying other data in the buffer memory:

S20d, if the difference value is not less than 0 and the last difference value is less than 0, controlling the temperature regulating device 7 to be started, controlling the temperature of the shell 3 to be maintained at 34 ℃, and returning to S10 after counting down the single waiting time;

s20e, if the difference value is smaller than 0 and the last difference value is not smaller than 0, closing the temperature regulating device 7, and returning to S10;

S20f, if the difference value calculated in the last two times is not smaller than 0 and the difference value is larger than the last difference value, controlling the temperature of the shell 3 to be maintained at 32 ℃, and returning to S10 after counting down the single waiting time;

When the health pillow collects abnormal data, the abnormal condition acquisition does not belong to the actual condition of the patient, at the moment, the verification and complement of the judgment result (made by the health pillow) by the patient or the family member of the patient are needed, and when the verification and complement operation is not timely provided by the patient or the family member of the patient, the judgment result (made by the health pillow) can be verified by other physiological data of the patient, namely, the judgment result (made by the health pillow) is complemented in multiple rounds by using other physiological data of the patient, so that the judgment accuracy of the health pillow is increased.

The temperature adjusting device 7 is further used for sending the acquired shell 3 temperature and adjustment record to the patient terminal after the temperature adjusting device is started, the patient terminal receives the shell 3 temperature and adjustment record, the shell 3 temperature and adjustment record corresponds to brain oxygen saturation acquisition data according to acquisition time and is converted into clinical text data to be sent to the acquisition module, the acquisition module receives an alarm signal, the alarm signal sending time is used as an accident time node, electrophysiological data, clinical text data and alarm signals before and after one hour of the accident time node are sent to the medical care terminal and the decision module, the medical care terminal sends emergency prompt sound after receiving the alarm signal and displays the received electrophysiological data and clinical text data, the decision module receives the alarm signal, marks the received electrophysiological data and clinical text data as the highest priority, calculates an emergency treatment scheme by using the electrophysiological data and the clinical text data, sends the emergency treatment scheme to the patient terminal, and plays the emergency treatment scheme after the patient terminal receives the emergency treatment scheme.

In the specific implementation, the patient A is taken as a user of the rehabilitation pillow, the accompanying A is taken as a nursing staff of the patient A, the mobile phone of the patient is taken as a patient terminal, and the computer of a doctor is taken as a medical care terminal.

The three-dimensional model of the head and neck of the patient A, which has been entered in the processing module before using the rehabilitation pillow, is used as a reference model A.

The patient A does not use the rehabilitation pillow for sleeping in daytime, the mobile phone of the patient A is placed on the rehabilitation pillow, the pressure sensor 2 for detecting the pressure on the rehabilitation pillow transmits the detected pressure value to the processing device 5, and meanwhile, the near infrared light probe 6 and the temperature regulating device 7 are started to collect brain oxygen saturation data and the temperature of the shell 3. The processing device 5 records the point position of each pressure sensor 2, after receiving the pressure values, the processing device 5 finds out the point position corresponding to the pressure sensor 2 transmitting the pressure values as the compression point position, and establishes a three-dimensional model as a temporary model (a three-dimensional model corresponding to the mobile phone) through all the pressure values and the compression point positions by corresponding the compression point positions to the pressure values one by one. The processing device 5 compares the reference model A with the temporary model to obtain the similarity 1, wherein the similarity 1 is 2 percent and is smaller than 90 percent, the rehabilitation pillow deletes cerebral oxygen saturation data which is acquired simultaneously with the pressure value from the cache as interference data, and deletes the pressure value and the temperature of the shell 3 which is acquired simultaneously with the pressure value.

After a period of time, the accompanying A is at noon nap by using the rehabilitation pillow, the processing device 5 of the rehabilitation pillow compares the reference model A with the temporary model established at this time to obtain the similarity 2, the similarity 2 is 80% and is smaller than 90%, the rehabilitation pillow deletes the cerebral oxygen saturation data which is simultaneously acquired with the pressure value from the cache as interference data, and simultaneously deletes the pressure value and the temperature of the shell 3 which is simultaneously acquired with the pressure value.

On the evening of the day, the patient A sleeps by using the rehabilitation pillow, the processing device 5 of the rehabilitation pillow compares the reference model A with the temporary model established at this time to obtain the similarity 3, the similarity 3 is 98 percent and is more than 90 percent, the rehabilitation pillow selects the median in the cerebral oxygen saturation data acquired at the same time with the pressure value as cerebral oxygen saturation acquisition data (8.09), the cerebral oxygen saturation acquisition data is transmitted to the patient terminal as electrophysiological data, meanwhile, the temperature (36.2 ℃) of the shell 3 acquired at the same time with the pressure value is transmitted to the patient terminal, and the electrophysiological data and the temperature of the shell 3 are transmitted to the acquisition module by the patient terminal.

The maximum cerebral oxygen saturation preset in the processing device 5 is 10.09, the minimum cerebral oxygen saturation is 7.88, the single waiting time is 5 minutes, and the maximum waiting time is 30 minutes.

The rehabilitation pillow calculates that the difference value between the brain oxygen saturation acquired data and the highest brain oxygen saturation is-2 and less than 0, and the rehabilitation pillow continuously acquires the brain oxygen saturation data of the patient.

After a period of time, the rehabilitation pillow calculates that the difference between the brain oxygen saturation acquisition data and the highest brain oxygen saturation is 0.5, the temperature regulating device 7 is controlled to be started (the temperature regulating device 7 is started again to ensure the opening state), the temperature of the shell 3 is controlled to be maintained at 34 ℃, after a countdown single waiting time (5 minutes), the difference between the brain oxygen saturation acquisition data and the highest brain oxygen saturation is calculated again to be 0.2, the difference between the last two times is not smaller than 0, the difference (0.2) is not larger than the last difference (0.5), after the countdown single waiting time (5 minutes), the difference between the brain oxygen saturation acquisition data and the highest brain oxygen saturation is calculated again to be-0.2, the current difference is smaller than 0 (-0.2) and the last difference is not smaller than 0 (0.2), the temperature regulating device 7 is controlled to be closed, and the rehabilitation pillow continues to acquire the brain oxygen saturation data of a patient.

After a period of time, the rehabilitation pillow calculates the difference between the brain oxygen saturation acquisition data and the highest brain oxygen saturation to be 0.4,5 minutes, then the difference is measured again to be 0.9, the control device controls the temperature of the shell 3 to be maintained at 32 ℃, the difference is measured again to be 1.3 after 5 minutes, the processing device 5 judges three continuous waiting times and the difference is continuously increased, the processing device 5 sends a first instruction to a patient terminal (mobile phone), the mobile phone receives the instruction and displays a first prompt, the first prompt is circularly broadcast, the accompanying A listens to the first prompt broadcast by the mobile phone and then immediately detects the state of the patient A, the accompanying A is found to be in an abnormal state, the accompanying A is closed and then selects to send an alarm signal, then the mobile phone sends the alarm signal to the acquisition module, the acquisition module receives the alarm signal, then sends the alarm signal sending time to the medical terminal and the decision module as an accident time node, and sends emergency prompt sound after the medical terminal (doctor computer) receives the alarm signal, and displays the received medical physiological data and the clinical text data, and the emergency text data are displayed, and the clinical text data are displayed and the various items are input to the acquisition module after the doctor computer receives the voice data. And after receiving the alarm signal, the decision module marks the received electrophysiology data and clinical text data as the highest priority, calculates an emergency treatment scheme by using the electrophysiology data and the clinical text data, sends the emergency treatment scheme to a mobile phone of a patient, plays the emergency treatment scheme after receiving the emergency treatment scheme, and carries out first aid for the patient according to the emergency treatment scheme broadcasted by the mobile phone by accompanying A.

The foregoing is merely exemplary embodiments of the present application, and specific structures and features that are well known in the art are not described in detail herein. It should be noted that modifications and improvements can be made by those skilled in the art without departing from the structure of the present application, and these should also be considered as the scope of the present application, which does not affect the effect of the implementation of the present application and the utility of the patent. The scope of the application is to be determined by the appended claims, and the description of the embodiments in this specification should be construed to include all embodiments falling within the scope of the claims.

Claims

1. Multiple-round complementary HICH intelligent rehabilitation big health system based on image and text recognition, which is characterized by comprising:

2. The multi-round complementary HICH intelligent rehabilitation big health system based on image and text recognition according to claim 1, wherein the acquisition module further comprises a recognition unit, the recognition unit is based on a calculation model of U-net++ deep learning and a neural network, when the acquisition unit acquires multi-mode MRI and PET images in the imaging data, the recognition unit mainly carries out automatic recognition on the skull MRI and PET images in the patient rehabilitation process, and the recognition unit is used for evaluating the brain parenchyma and white matter fiber bundle integrity degree, the brain blood flow perfusion condition and the brain function retention condition of the patient and finally outputting the final result of automatic interpretation of the MRI and PET images in the form of Chinese text; the identification unit uses a method in CT image identification to carry out side, positioning and quantitative identification on MRI and PET image data, adds a Dense block and a convolution layer between an encoder and a decoder by using U-net++, adds a redesigned jump path on the basis of the original U-net++, fuses the output of a previous convolution layer of the same Dense block with the sampling output corresponding to a lower Dense block, enables the semantic level of the coded features to be closer to the semantic level of the feature mapping to be carried out in the decoder, adopts Dense jump connection to realize the jump path between the encoder and the decoder so as to ensure that all prior feature graphs are accumulated, reaches the current node through Dense convolution blocks on each jump path, generates a feature map with complete resolution at a plurality of semantic levels, improves the segmentation precision and the gradient flow, and is used for increasing the depth supervision model to adjust the complexity of the model through pruning, and the identification unit takes the identification result obtained by one of the MRI and PET image data as a reference result after carrying out side, positioning and quantitative identification on the MRI and PET image data, obtains the identification obtained by other MRI and PET image data collected by the same examination as a complementary result, sequentially verifies and compares a plurality of identification results with the reference result, carries out multi-round complementation on the reference result, and outputs the multi-round complementation result of the reference result as a final identification result.

3. The multi-round complementary HICH intelligent rehabilitation health system based on image and text recognition of claim 2, wherein when the acquisition unit acquires electrophysiological data, the recognition unit performs smoothing treatment on spike signals in a half-wave treatment mode, performs wavelet decomposition on the electrophysiological data by adopting a principal component analysis algorithm to obtain signal components, then applies the principal component analysis algorithm on the signal components, and searches a mixed signal matrix of artifact signals and electroencephalogram signals by adopting an independent component analysis algorithm, thereby separating and treating various artifacts from the mixed composite signals; the recognition unit adopts expert feature analysis to perform preliminary screening according to the features of various conventional interference waveforms to remove eye movement artifacts, blink artifacts and electrode artifacts, and simultaneously uses a band-pass filter to complete the filtering processing of the electroencephalogram signals to obtain electroencephalogram signals of useful frequency bands, filters various interference signals outside the frequency bands, adopts a simple scaling method to perform normalization processing, segmentation storage processing and labeling processing on electrophysiological data, then inputs the preprocessed electrophysiological sample data into a sparse automatic encoder, extracts the features of the electrophysiological data based on the sparse automatic encoder, inputs the result output by the sparse automatic encoder into a long-short-period memory cyclic neural network, analyzes the extracted data features, inputs the analysis result based on the long-short-period memory cyclic neural network into a softmax classifier, thereby completing classification of the electrophysiological data, and finally outputting the final result of automatic interpretation of the electrophysiological data in the form of Chinese text.

4. The multi-round complementary HICH intelligent rehabilitation health system based on image and text recognition according to claim 3, wherein when the acquisition unit acquires CT images, the recognition unit segments cerebral parenchymal area hemorrhagic stoves, edema brain tissues, normal brain tissues, each ventricle and important brain pools in the brain window level of the skull CT through U-net++, extracts corresponding characteristics thereof, completes automatic recognition of the skull CT images of the HICH patient, comprises side fixing, positioning and quantitative recognition, and finally outputs the final result of automatic interpretation of the CT images in the form of Chinese text.

5. The multi-round complementary HICH intelligent rehabilitation health system based on image and text recognition of claim 4, further comprising an analysis unit for comparing the final result output by the recognition unit with the result obtained by the doctor and judging the accuracy, the specificity and the sensitivity of the recognition unit, wherein the analysis unit is further used for comparing the number of the final results output by the recognition unit with the doctor in the same unit time and calculating and comparing the accuracy of the two results.

6. The multi-round complementary HICH intelligent rehabilitation health system based on image and text recognition of claim 5, further comprising a rehabilitation pillow, a medical care terminal and a patient terminal, wherein the rehabilitation pillow is in communication connection with the patient terminal, the medical care terminal is in communication connection with the acquisition module and the decision module, the patient terminal is in communication connection with the acquisition module and the decision module, the rehabilitation pillow is used for acquiring brain oxygen saturation data of a patient and transmitting the brain oxygen saturation data as rehabilitation acquisition data to the patient terminal, the patient terminal is used for receiving and displaying the rehabilitation acquisition data and transmitting the rehabilitation acquisition data as electrophysiological data to the acquisition module, the patient terminal is also used for receiving and displaying the data transmitted by the decision module, and the medical care terminal is used for reading the acquisition data, receiving the data and inputting clinical text data to the acquisition terminal.

7. The multi-wheel complementary HICH intelligent rehabilitation big health system based on image and text recognition, according to claim 6, wherein the rehabilitation pillow comprises a shell and a plurality of near infrared light probes, the near infrared light probes are embedded in the surface of the shell, the pressure sensors are embedded in the surface of the shell, the temperature adjusting device is embedded in the shell, the power supply device and the processing device are fixedly connected in the shell, the processing device is in communication connection with the near infrared light probes, the pressure sensors and the temperature adjusting device, the power supply device is electrically connected with the near infrared light probes, the pressure sensors and the processing device, the near infrared light probes are used for collecting brain oxygen saturation data of a patient and transmitting the brain oxygen saturation data to the processing device, the pressure sensors are used for collecting pressure values received by all points on the rehabilitation pillow and transmitting the pressure values to the processing device, the temperature adjusting device is used for obtaining the shell temperature and transmitting the processing device, the processing device is used for receiving the brain oxygen saturation data, the pressure values and the shell temperature, analyzing the carrier on the pillow according to the pressure values and the pressure points providing the pressure values, and then the carrier is used for removing oxygen saturation data in the brain oxygen saturation data of the brain interference data, and finally transmitting oxygen saturation data of the patient to a physiological terminal is obtained.

8. The multi-round complementary HICH intelligent rehabilitation big health system based on image and text recognition is characterized in that a plurality of pressure sensors are uniformly distributed on a shell, a processing device stores received brain oxygen saturation data, pressure values and shell temperature into a cache, when the processing device is used for eliminating interference data in the brain oxygen saturation data, a three-dimensional model of a carrier is established through the pressure values and pressure points providing the pressure values as a temporary model, a reference model is preset in the processing module, the reference model is a three-dimensional model of a head and a neck of a patient which are input in advance, the processing module is used for comparing the temporary model with the reference model to obtain similarity, when the similarity is smaller than 90%, brain oxygen saturation data which are acquired simultaneously with the pressure values are deleted from the cache as interference data, and shell temperature which is acquired simultaneously with the pressure values are deleted, when the similarity is larger than or equal to 90%, a median of the brain oxygen saturation data which are acquired simultaneously with the pressure values is selected as brain oxygen saturation acquisition data, when the similarity is between 90% and 97%, the brain oxygen saturation data are corrected temporarily, and then the brain oxygen saturation data are transmitted to other terminal physiological data as the temporary data of the patient.

9. The multiple complementary rounds of image and text recognition based HICH intelligent rehabilitation and health system according to claim 8, wherein the temperature adjusting device is further configured to adjust the temperature of the shell, the processing device is preset with a maximum cerebral oxygen saturation, a minimum cerebral oxygen saturation, a single waiting time and a maximum waiting time, and the processing device performs the following steps after sending the cerebral oxygen saturation acquisition data as electrophysiological data to the patient terminal and before emptying other data in the buffer memory:

10. The multi-round complementary HICH intelligent rehabilitation health system based on image and text recognition of claim 9, wherein the temperature adjusting device is further used for sending the acquired shell temperature and adjustment record to the patient terminal after being started, the patient terminal receives the shell temperature and adjustment record, corresponds the shell temperature and adjustment record to the brain oxygen saturation acquisition data according to acquisition time and converts the acquisition time and the brain oxygen saturation acquisition data into clinical text data and sends the clinical text data to the acquisition module, the acquisition module receives an alarm signal, takes the alarm signal sending time as an accident time node, sends electrophysiological data, clinical text data and alarm signals before and after the accident time node for one hour to the medical terminal and the decision module, sends out emergency prompt sound after the medical terminal receives the alarm signal, displays the received electrophysiological data and the clinical text data, the decision module receives the alarm signal, marks the received electrophysiological data and the clinical text data as the highest priority, calculates an emergency treatment scheme by using the electrophysiological data and the clinical text data, sends the emergency treatment scheme to the patient terminal, and plays the emergency treatment scheme after the patient terminal receives the emergency treatment scheme.