Anesthesia and consciousness depth monitoring system
By combining a MEMS targeted sensor with an intraoperative ultrasound-calibrated signal acquisition system, along with a three-level noise reduction and dual-dimensional evaluation system, the signal specificity and black-box problems of traditional anesthesia monitoring systems have been solved. This enables precise monitoring and personalized intervention for patients with brain diseases, ensuring accurate judgment and safety of the anesthesia status.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- YUYAO PEOPLES HOSPITAL
- Filing Date
- 2026-03-01
- Publication Date
- 2026-04-21
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Traditional anesthesia consciousness depth monitoring systems lack targeted monitoring of key nuclei, have poor signal specificity, and are difficult to accurately capture abnormal physiological signals in patients with brain diseases. Furthermore, they lack intraoperative brain structure displacement calibration mechanisms, leading to monitoring biases, strong black box nature of assessment models, and a single early warning mechanism that cannot adapt to personalized needs.
A signal acquisition system using MEMS targeted sensors and intraoperative ultrasound calibration, combined with a three-level noise reduction scheme, is constructed using a CNN, LSTM, and SHAP fusion architecture to build a two-dimensional assessment system. Combined with brain disease-specific parameter adaptation and a graded early warning mechanism, an anesthesia and drug administration synergy module is built to form a closed loop of monitoring-assessment-intervention.
It achieves precise capture and stable processing of amygdala-specific neural signals, solves the problem of insufficient specificity in traditional monitoring, constructs an interpretable assessment framework, enables personalized dosing recommendations, reduces intraoperative risks, and improves the accuracy and safety of monitoring.
Smart Images
Figure CN121890954A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical device technology, specifically to an anesthesia and depth of consciousness monitoring system. Background Technology
[0002] In clinical surgery, monitoring the depth of consciousness under anesthesia is a crucial step in ensuring surgical safety, directly impacting the stability of the patient's vital signs during surgery and postoperative recovery. With the development of biomedical engineering technology, anesthesia monitoring equipment has evolved from early physiological sign monitoring to more precise and personalized methods. Currently, most mainstream monitoring systems assess the depth of anesthesia based on electroencephalogram (EEG) signal analysis and are widely used in various surgical procedures. This is especially important for patients with brain diseases such as epilepsy and brain tumors, where precise control of the depth of anesthesia during surgery is paramount, requiring a balance between anesthetic efficacy and brain function protection. This places higher demands on the specificity and adaptability of monitoring systems.
[0003] Traditional anesthesia consciousness depth monitoring systems suffer from numerous technical shortcomings. These systems often employ generalized whole-brain EEG signal acquisition, failing to target key nuclei such as the amygdala related to consciousness regulation and epileptic seizures. This results in poor signal specificity, making it difficult to accurately capture abnormal physiological signals in patients with brain diseases. Furthermore, existing systems lack intraoperative brain structure displacement calibration mechanisms, making them susceptible to monitoring biases due to changes in body position and surgical procedures. The assessment models are often black-box algorithms, lacking interpretable assessment logic and failing to establish a synergistic relationship between monitoring results and anesthetic administration, making it difficult to adapt to the individualized needs of patients with brain diseases. In addition, traditional systems have simplistic early warning mechanisms with insufficient lead time, failing to allow adequate intervention time for medical staff and posing certain clinical safety risks. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an anesthesia and consciousness depth monitoring system. This system integrates a MEMS targeted sensor with an intraoperative ultrasound-calibrated signal acquisition system, coupled with a three-level noise reduction signal preprocessing scheme, to accurately capture specific neural signals in the amygdala region. The timing is then calibrated by a whole-brain signal synchronization module, and a two-dimensional evaluation system is constructed through a fusion architecture of CNN, LSTM, and SHAP. Combined with brain disease-specific parameter adaptation and a graded early warning mechanism, an anesthesia administration coordination module is finally built through a gradient boosting tree algorithm, forming a complete monitoring-evaluation-intervention closed loop.
[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: an anesthesia and depth of consciousness monitoring system, the system comprising:
[0006] Signal acquisition module: includes an array of sensors fabricated using MEMS technology, a miniature intraoperative ultrasound probe, and a differential input acquisition circuit. The sensor surface is coated with a polylactic acid-based conductive polymer film, and the acquisition channels are independently adjustable.
[0007] Signal preprocessing module: It adopts a three-level noise reduction scheme of adaptive power frequency filtering, wavelet threshold noise reduction and electromyographic noise separation to extract specific frequency band signals of amygdala and epileptic abnormal discharge characteristics, generate standardized feature vectors, and has a built-in temperature compensation module.
[0008] Whole brain signal synchronization module: includes multi-channel EEG acquisition electrodes, with a sampling rate consistent with the signal acquisition module, extracts relevant EEG assessment parameters, and calibrates signal timing through a timestamp synchronization mechanism;
[0009] Dual-dimensional fusion assessment module: It adopts a fusion architecture of CNN, LSTM and SHAP, and assigns weights of nucleus signals and whole brain signals based on the patient's brain images and disease type, and outputs quantitative values of consciousness depth, brain functional status assessment results and signal contribution heatmaps.
[0010] Early warning module: When a specific epileptic abnormal discharge characteristic or a quantified value of the depth of consciousness exceeds a preset range, a graded early warning is triggered;
[0011] Parameter adaptation module: Pre-stores parameter templates specific to brain diseases, which can be automatically matched based on the patient's preoperative brain imaging information;
[0012] Anesthesia Dosing Coordination Recommendation Module: Includes a clinical case database, outputs dosing recommendations through gradient boosting tree algorithm, and supports manual correction and feedback optimization of the model.
[0013] Furthermore, the MEMS sensor array is a 3×3 layout with a flexible design, a contact resistance of ≤5kΩ, and a frequency of ≥10MHz for the miniature intraoperative ultrasound probe. When amygdala displacement of ≥2mm is detected, the sensor acquisition range and sensitivity are finely adjusted. Displacement calibration is achieved through a registration algorithm based on image feature point matching between preoperative brain images and intraoperative ultrasound images.
[0014] Furthermore, the differential input acquisition circuit of the signal acquisition module is used to suppress common-mode interference, the acquisition channel adjustment range is adapted to individual differences in the amygdala position of different patients, and the adjusted acquisition parameters are adapted to the feature extraction process of the signal preprocessing module.
[0015] Furthermore, the signal preprocessing module separates electromyographic noise through independent component analysis, and the determination of epileptic abnormal discharge characteristics is based on the following formula: ,in, To detect the amplitude of the signal, To detect the frequency of the signal, The threshold for amplitude determination. The frequency determination threshold is used to extract parameters including the amplitude, frequency, and duration of spikes and sharp waves. The temperature compensation module is adapted to ambient temperature changes of ±5℃.
[0016] Furthermore, the multi-channel EEG acquisition electrodes of the whole-brain signal synchronization module have 8 channels, which are deployed in the prefrontal, parietal, and temporal lobe regions respectively. The timestamp synchronization mechanism ensures that the acquisition time deviation between the nucleus signals and the whole-brain signals is ≤1ms. The synchronized signals are directly used as input data for the dual-dimensional fusion evaluation module.
[0017] Furthermore, the weight allocation of nucleus signals and whole-brain signals in the dual-dimensional fusion assessment module is based on the following formula: ,in, To integrate the evaluation weights, For dynamic adjustment coefficients, For the basic weights of the nucleus signal, As the basic weight of whole-brain signals, for ordinary patients Values range from 30% to 40%, for patients with brain diseases. Values range from 40% to 50%. Adjust in real time according to the intraoperative displacement calibration parameters.
[0018] Furthermore, the CNN network of the dual-dimensional fusion evaluation module contains 3 convolutional layers and 2 pooling layers, and the LSTM network contains 2 hidden layers and 1 fully connected layer. The feature contribution ratio of the signal contribution heatmap generated by the SHAP algorithm is calculated using the following formula: ,in, The contribution percentage of the i-th feature. Let be the SHAP value of the i-th feature. The sum of SHAP values for all features involved in the evaluation is used to indicate the contribution percentage of the 4-12Hz theta wave, the 13-30Hz beta wave, and the epileptic abnormal discharge features.
[0019] Furthermore, the exclusive parameter template of the parameter adaptation module is generated based on clinical data from more than 500 patients with corresponding diseases, covering epilepsy, brain tumors, and traumatic brain injury, as well as different severity levels. It supports manual fine-tuning based on patient age, weight, and surgical type. The basic parameters of the template are directly related to the weight allocation formula of the dual-dimensional fusion assessment module.
[0020] Furthermore, the warning module provides three types of tiered warnings: Level 1 warnings are indicated by a green indicator light, Level 2 warnings are indicated by a yellow indicator light and a 1Hz sound, and Level 3 warnings are indicated by a red indicator light, a 2Hz high-frequency sound, and a pop-up window on the screen. The warning lead time is ≥5 seconds, the warning sensitivity can be adjusted as needed, and the warning trigger threshold is consistent with the epileptic abnormal discharge judgment threshold of the signal preprocessing module.
[0021] Furthermore, the clinical case database of the anesthetic drug administration collaborative suggestion module contains 1000+ clinical cases. The database correlates the monitoring results with the anesthetic drug dosage, and the output drug administration suggestions clearly define the adjustment direction and range. The clinical validation accuracy is ≥90%. The gradient boosting tree algorithm takes the dual-dimensional fusion evaluation results, patient age, weight, surgical type, and brain disease type as input features, and generates personalized drug administration suggestions by iteratively training and optimizing model parameters. It is also directly related to the output results of the dual-dimensional fusion evaluation module.
[0022] Compared with existing technologies, this anesthesia and depth of consciousness monitoring system has the following advantages:
[0023] I. This invention constructs a signal acquisition system combining a MEMS targeted sensor with intraoperative ultrasound calibration, coupled with a three-level noise reduction signal preprocessing scheme, to achieve precise capture and stable processing of specific neural signals in the amygdala region. It overcomes the limitations of traditional anesthesia monitoring that relies on generalized signals across the entire brain. Combined with a dynamic weighted dual-dimensional fusion assessment mode, it can accurately match the physiological signal characteristics of patients with brain diseases such as epilepsy and brain tumors, solving the problem of insufficient specificity in traditional monitoring. This allows medical staff to better assess the anesthesia status in relation to the patient's actual condition, reducing intraoperative risks caused by monitoring bias.
[0024] Second, this invention utilizes an interpretable evaluation architecture that integrates CNN, LSTM, and SHAP, combined with an anesthesia administration coordination module built using the gradient boosting tree algorithm, forming a closed-loop process from monitoring and evaluation to intervention recommendations. This solves the black-box problem of traditional AI monitoring models, making the evaluation logic intuitive and verifiable through feature contribution heatmaps, while simultaneously achieving direct linkage between monitoring results and personalized medication recommendations. The tiered early warning mechanism and parameter adaptation function further align with the needs of clinical operations, reducing the judgment burden on medical staff and freeing anesthesia management for patients with brain diseases from the constraints of uniform standards, allowing for flexible adjustments based on the individual patient's condition.
[0025] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description
[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0027] Figure 1 This is a schematic diagram of the module composition and data flow of the anesthesia and depth of consciousness monitoring system of the present invention;
[0028] Figure 2 This is a schematic diagram of the signal acquisition and preprocessing process of the present invention;
[0029] Figure 3 This is a schematic diagram of the dual-dimensional fusion evaluation and interpretability output process of the present invention. Detailed Implementation
[0030] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structure, features and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.
[0031] Example 1
[0032] This embodiment provides an application of an anesthesia and consciousness depth monitoring system in intraoperative anesthesia monitoring of epilepsy patients. Through targeted capture and displacement calibration of the signal acquisition module, noise suppression and feature extraction of the signal preprocessing module, temporal calibration of the whole-brain signal synchronization module, interpretability analysis of the dual-dimensional fusion assessment module, disease-specific adaptation of the parameter adaptation module, graded prompts of the early warning module, and personalized intervention of the anesthesia administration coordination suggestion module, a complete monitoring-assessment-intervention closed loop is constructed. This system accurately adapts to the physiological signal characteristics of epilepsy patients, solves the problems of insufficient specificity and black-box assessment in traditional monitoring, and ensures the safety of intraoperative anesthesia for epilepsy patients.
[0033] The specific implementation process is as follows:
[0034] Signal acquisition module
[0035] The signal acquisition module is the core of the system's data input. Its primary task is to accurately capture specific neural signals from the amygdala region, while also addressing acquisition deviations caused by intraoperative brain displacement. For example... Figure 1 As shown, the module consists of an array of sensors, a miniature intraoperative ultrasound probe, and a differential input acquisition circuit. The array of sensors is fabricated using MEMS technology, with a polylactic acid-based conductive polymer film coated on its surface. This film improves the stability of the sensor's fit with the scalp and enhances signal transmission performance, ensuring effective acquisition of microvolt-level neural electrical signals. The acquisition channels support independent adjustment, allowing medical staff to tailor the acquisition sensitivity of each channel based on the amygdala location determined by the patient's preoperative brain imaging, thus adapting to individual differences in the amygdala location among different patients.
[0036] A miniature intraoperative ultrasound probe is responsible for the dynamic calibration of brain displacement during surgery. Preoperatively, the initial position of the amygdala is determined using brain imaging. During surgery, the ultrasound probe acquires real-time images of the brain structure. Using an image feature point matching algorithm, the intraoperative images are registered with the preoperative images to accurately calculate the amygdala displacement. When the displacement reaches a set threshold, the system automatically triggers fine-tuning of the sensor's acquisition range and sensitivity, ensuring that even with brain displacement, the sensor accurately covers the amygdala region, avoiding signal acquisition deviations caused by displacement. The differential input acquisition circuit is specifically designed to suppress common-mode interference during the acquisition process. This interference often originates from external factors such as environmental electric and magnetic fields. The differential input method effectively cancels out the common-mode components in the two input signals, retaining only the target neural electrical signals and providing high-quality raw data for subsequent signal processing.
[0037] Signal preprocessing module
[0038] The core responsibility of the signal preprocessing module is to suppress noise and extract features from the acquired raw signals, providing standardized feature vectors for subsequent fusion evaluation. For example... Figure 2 As shown, this module employs a three-stage noise reduction scheme: adaptive power frequency filtering, wavelet threshold denoising, and electromyographic noise separation. First, adaptive power frequency filtering addresses power frequency interference in the environment by dynamically adjusting filtering parameters to accurately filter out interference signals in that frequency band, preventing them from masking neural signal characteristics. Next, wavelet threshold denoising decomposes the signal into different frequency components at multiple scales, setting thresholds based on the noise characteristics of each component to eliminate components with high noise levels and retain effective components containing the target signal characteristics, thus separating noise from the effective signal. Finally, independent component analysis separates electromyographic noise, which originates from facial muscle activity and whose signal characteristics differ significantly from neural electrical signals. Independent component analysis decomposes the signal into independent components, separating the components corresponding to electromyographic noise to further improve signal purity.
[0039] After noise suppression is complete, the module begins extracting specific frequency band signals from the amygdala and features of abnormal epileptic discharges. These specific frequency band signals from the amygdala are closely related to consciousness and seizure patterns, and are crucial for assessing the depth of anesthesia and brain function. The extraction of abnormal epileptic discharge features uses a judgment formula. Where A is the amplitude of the detected signal, and f is the frequency of the detected signal. The threshold for amplitude determination. A frequency threshold is set. Epilepsy abnormal discharge signals typically possess specific amplitude and frequency characteristics. By setting corresponding thresholds, abnormal discharge signals such as spikes and sharp waves can be accurately identified, and key parameters such as amplitude, frequency, and duration of these signals can be extracted. Finally, these characteristic parameters are integrated to generate a standardized feature vector, providing data support for subsequent fusion evaluation. Simultaneously, the module incorporates a temperature compensation module that automatically adjusts signal processing parameters when the ambient temperature changes, avoiding feature extraction deviations caused by temperature drift and ensuring signal processing stability.
[0040] Whole brain signal synchronization module
[0041] The core function of the whole-brain signal synchronization module is to supplement the monitoring data across the entire brain, forming a two-dimensional data support with the amygdala-specific signals, while ensuring the temporal consistency of the two types of signals. This module includes multi-channel EEG acquisition electrodes deployed in key brain regions such as the prefrontal, parietal, and temporal lobes. The EEG signals from these regions reflect the overall state of consciousness, and the extracted EEG-related assessment parameters will serve as another important data source for the two-dimensional fusion assessment.
[0042] To ensure the effectiveness of the fusion of nucleus and whole-brain signals, the module employs a timestamp synchronization mechanism to calibrate signal timing. Since nucleus and whole-brain signals are acquired synchronously by different acquisition components, slight time discrepancies may exist, which directly affect the accuracy of the fusion assessment. The timestamp synchronization mechanism adds a unique timestamp to each acquired signal. By comparing the timestamps of different signals, the system adjusts the temporal relationship of the signals, ensuring complete alignment of nucleus and whole-brain signals in the time dimension. The synchronized signals are directly transmitted to the two-dimensional fusion assessment module as input data, laying the timing foundation for two-dimensional fusion analysis.
[0043] Dual-dimensional fusion evaluation module
[0044] The dual-dimensional fusion assessment module is the core analysis unit of the system. Through a fusion architecture of CNN, LSTM, and SHAP, it achieves accurate assessment and interpretable output of anesthesia depth and brain functional status. For example... Figure 3 As shown, the CNN network contains multiple convolutional layers and pooling layers. The role of the convolutional layers is to extract the spatial specific features of the nucleus signal. The nucleus signal has obvious spatial distribution characteristics. The convolutional layers capture the local spatial correlation information of the signal through convolution operations. The pooling layers perform dimensionality reduction processing on the convolutional features, reducing the amount of computation while retaining key features. The LSTM network contains multiple hidden layers and fully connected layers. It is mainly used to analyze the temporal dynamic changes of the whole brain signal. The consciousness-related features of the whole brain signal fluctuate dynamically over time. The LSTM network effectively captures the long-term temporal dependencies of the signal and extracts key features in the temporal dimension through memory units and gating mechanisms.
[0045] After the two types of features are extracted, the weight allocation formula is used. To integrate, among which, To integrate the evaluation weights, For dynamic adjustment coefficients, For the basic weights of the nucleus signal, The baseline weights for whole-brain signals are determined by the amygdala region. Signals in the amygdala region of epileptic patients are closely related to their disease state and anesthetic response, therefore requiring a higher baseline weight. In contrast, signals in ordinary patients are allocated a more balanced weight distribution. The value of is determined by the intraoperative displacement calibration parameter, because intraoperative displacement affects the acquisition quality of the nucleus signal. When displacement causes a decrease in the reliability of the nucleus signal, the system adjusts . Reduce the fusion weight of nucleus signals and increase the weight of whole-brain signals to ensure the stability and accuracy of the fusion assessment results.
[0046] After fusion evaluation, a signal contribution heatmap is generated using the SHAP algorithm, and the proportion of feature contributions is determined by the formula. Calculate, where, The contribution percentage of the i-th feature. Let be the SHAP value of the i-th feature. The sum of the SHAP values for all features involved in the assessment is used. This formula quantifies the impact of each feature on the assessment results, visually presenting the contribution ratio of specific frequency band signals in the amygdala and abnormal epileptic discharge features through a heatmap. This solves the black-box problem of traditional AI models, allowing medical staff to clearly understand the logic behind the assessment results. Ultimately, the module outputs a quantitative value of consciousness depth and an assessment result of brain function status, providing a core basis for subsequent early warning and medication recommendations.
[0047] Parameter adaptation module
[0048] The core objective of the parameter adaptation module is to provide epilepsy patients with customized monitoring parameter configurations, ensuring that the system is adapted to disease specificity. The module pre-stores dedicated parameter templates for epilepsy and other brain diseases. These templates are generated based on a large amount of clinical data from patients with corresponding diseases and cover key aspects such as signal acquisition thresholds, feature extraction parameters, and fusion weight benchmark values, enabling precise matching of the physiological signal response patterns of epilepsy patients.
[0049] Preoperatively, medical staff input the patient's preoperative brain imaging information. The system automatically matches corresponding epilepsy-specific parameter templates based on the location and extent of the lesions reflected in the images. During the procedure, as signal acquisition and processing progress, the module dynamically fine-tunes the template parameters based on the real-time signal characteristics. For example, when frequent occurrences of abnormal epileptic discharge characteristics are detected, the system automatically adjusts the feature extraction threshold to ensure the accuracy of feature recognition. Simultaneously, the module allows medical staff to manually fine-tune parameters based on individual patient information such as age, weight, and type of surgery, balancing standardization and personalization needs. Furthermore, the basic parameters in the template are directly linked to the weight allocation formula of the two-dimensional fusion assessment module, ensuring that parameter adjustments directly impact the fusion assessment process and improve the adaptability of the assessment results.
[0050] Early warning module
[0051] The early warning module, based on the dual-dimensional fusion evaluation results and the feature extraction results from the signal preprocessing module, provides tiered alerts for abnormal situations. The module's early warning trigger logic is consistent with the epileptic abnormal discharge judgment threshold of the signal preprocessing module; when a threshold is detected... When the epileptic abnormal discharge characteristics or the quantification value of the depth of consciousness exceeds the preset range, the system will automatically trigger a graded warning.
[0052] The tiered early warning system is designed with three levels: Level 1, indicated by a green indicator light, alerts medical staff to suspicious abnormalities requiring close monitoring; Level 2, indicated by a yellow indicator light and a specific frequency sound, suggests a mild abnormality requiring intervention; and Level 3, indicated by a red indicator light, a high-frequency sound, and a pop-up window on the screen, suggests a severe abnormality requiring immediate intervention. The early warning lead time is set within a reasonable range to allow sufficient intervention time for medical staff, and the warning sensitivity can be adjusted as needed to adapt to the monitoring requirements of different surgical scenarios. The early warning module operates closely in conjunction with other modules; the source and key contributing characteristics of abnormal signals can be traced through the output results of the dual-dimensional fusion evaluation module, assisting medical staff in quickly locating problems.
[0053] Anesthesia Dosing Coordination Recommendation Module
[0054] The core of the anesthetic dosing coordination recommendation module is to generate personalized anesthetic dosing recommendations based on monitoring results, thus constructing a closed loop of monitoring and intervention. The module contains a large database of clinical cases, linking monitoring results with anesthetic drug dosages. All this data is validated through clinical practice, ensuring the reliability of the recommendations.
[0055] Dosing recommendations are generated using a gradient boosting tree algorithm, which takes a dual-dimensional fusion assessment result, patient age, weight, surgical type, and type of brain disease as input features. The gradient boosting tree algorithm optimizes model parameters through iterative training, fully exploring the non-linear relationship between input features and anesthetic dosage to accurately match the dosing needs of different patients. The algorithm's output dosing recommendations clearly define the direction and magnitude of adjustments. Healthcare professionals can manually revise the recommendations based on actual clinical conditions, and the revised data is fed back to the database in real time. Incremental training continuously optimizes model parameters, improving the accuracy of subsequent recommendations. This module is directly linked to the dual-dimensional fusion assessment module; changes in the assessment results affect the generation of dosing recommendations in real time, ensuring that the recommendations are synchronized with the patient's real-time anesthesia status.
[0056] This embodiment achieves precise monitoring and personalized intervention of the depth of consciousness during anesthesia in epilepsy patients through the coordinated operation of various modules. The targeted capture and shift calibration of the signal acquisition module ensure the reliability of amygdala-specific signals; the three-level noise reduction and feature extraction of the signal preprocessing module provide high-quality data for evaluation; the interpretability analysis of the two-dimensional fusion evaluation module solves the black-box evaluation problem; and the parameter adaptation module and the anesthesia administration coordination suggestion module ensure the system's disease adaptability and clinical applicability.
[0057] Example 2
[0058] This embodiment provides an application of anesthesia and depth of consciousness monitoring system in intraoperative anesthesia monitoring of brain tumor patients. Focusing on the characteristics of abnormal brain structure and specific physiological signals in brain tumor patients, the system achieves targeted acquisition of signals in the amygdala region, noise suppression, dual-dimensional fusion assessment, disease-specific parameter adaptation, graded early warning, and personalized drug administration recommendations through precise coordination of various modules. This solves the problem of insufficient adaptability of traditional monitoring for brain disease patients and ensures the accuracy of intraoperative anesthesia depth monitoring and the timeliness of clinical intervention in brain tumor patients.
[0059] The specific implementation process is as follows:
[0060] Signal acquisition module
[0061] The signal acquisition module addresses the structural abnormalities and potential nucleus displacement in brain tumor patients, emphasizing the synergy between targeted acquisition and intraoperative displacement calibration. The module comprises an array of sensors, a miniature intraoperative ultrasound probe, and a differential input acquisition circuit. The array of sensors is fabricated using MEMS technology, and its surface-coated polylactic acid-based conductive polymer film adapts to potential scalp abnormalities in brain tumor patients, ensuring stable contact between the sensor and the scalp and preventing signal distortion due to poor fit. The acquisition channels support independent adjustment. Preoperatively, based on the amygdala location determined by brain imaging and the nucleus displacement caused by tumor compression, the acquisition range of each channel is adjusted to ensure the sensor array accurately covers the core amygdala region.
[0062] A miniature intraoperative ultrasound probe acquires real-time images of the patient's brain structure during surgery. Since brain tumor patients may experience brain displacement during surgery due to tumor location and surgical procedures, the ultrasound probe uses an image feature point matching algorithm to register intraoperative images with preoperative images, accurately calculating the displacement of the amygdala. When the displacement reaches a set threshold, the system automatically fine-tunes the sensor's acquisition range and sensitivity to ensure that the sensor can effectively capture amygdala signals even under displacement, avoiding acquisition deviations caused by tumor compression and intraoperative displacement. A differential input acquisition circuit is used to suppress common-mode interference in the surgical environment. By canceling external interference signals, it preserves pure amygdala neural electrical signals, laying the foundation for subsequent signal processing.
[0063] Signal preprocessing module
[0064] The signal preprocessing module addresses the potential for abnormal signal fluctuations and increased noise interference in brain tumor patients by employing a three-stage denoising and precise feature extraction process to ensure signal quality and feature effectiveness. This module utilizes a three-stage denoising scheme: adaptive power frequency filtering, wavelet threshold denoising, and electromyographic noise separation. Adaptive power frequency filtering dynamically adjusts filtering parameters based on the intensity of power frequency interference in the environment, effectively filtering out such interference. Wavelet threshold denoising separates noise and effective components from the signal through multi-scale decomposition, eliminating invalid noise and retaining signal components containing amygdala features. Electromyographic noise separation uses independent component analysis to separate electromyographic noise and neural electrical signals that may arise during surgery due to changes in patient position and muscle tension, further improving signal purity.
[0065] In the feature extraction stage, the focus is on extracting specific frequency band signals from the amygdala and potential accompanying abnormal discharge characteristics. The determination of abnormal discharge characteristics uses a formula... ,in, f is the amplitude of the detected signal, and f is the frequency of the detected signal. The threshold for amplitude determination. This serves as the frequency threshold. Brain tumor patients may experience abnormal brain nerve discharges due to tumor compression. This formula accurately extracts abnormal discharge features such as spikes and sharp waves by identifying signals with specific amplitude and frequency characteristics, and records parameters such as amplitude, frequency, and duration. Simultaneously, the module incorporates a temperature compensation module. When the surgical environment temperature changes, it automatically calibrates the signal processing parameters to avoid the impact of temperature drift on the feature extraction results. Finally, the extracted feature parameters are integrated into a standardized feature vector and transmitted to subsequent modules.
[0066] Whole brain signal synchronization module
[0067] The whole-brain signal synchronization module provides whole-brain data support for the two-dimensional fusion assessment of brain tumor patients, while ensuring the temporal consistency between nucleus signals and whole-brain signals. The module includes multi-channel EEG acquisition electrodes deployed in key brain regions such as the prefrontal, parietal, and temporal lobes. EEG signals from these regions can reflect the overall state of consciousness of the whole brain. The extracted EEG-related assessment parameters complement the amygdala-specific signals, avoiding the limitations of single-nucleus signal assessment.
[0068] Due to the structural abnormalities in the brains of brain tumor patients, there may be a timing discrepancy between the acquisition of nucleus signals and whole-brain signals. The module calibrates the signal timing through a timestamp synchronization mechanism. The system adds a unique timestamp to each acquired signal. By comparing the timestamps of different signals, the transmission and storage timing of the signals are adjusted to ensure complete temporal alignment between nucleus signals and whole-brain signals, with timing deviations controlled within a reasonable range. The synchronized whole-brain signals are directly transmitted to the dual-dimensional fusion assessment module, serving as input data for the fusion assessment along with amygdala-specific signals. This provides dual-dimensional data support for accurately assessing the depth of anesthesia and brain functional status in brain tumor patients.
[0069] Dual-dimensional fusion evaluation module
[0070] The dual-dimensional fusion assessment module, tailored to the disease characteristics of brain tumor patients, achieves accurate and interpretable assessments through a fusion architecture and dynamic weight allocation. The CNN network, comprising multiple convolutional and pooling layers, extracts spatially specific features of the amygdala signal. In brain tumor patients, the amygdala signal may exhibit abnormal spatial distribution due to tumor compression. The convolutional layers capture these anomalous spatial features through convolutional operations, while the pooling layers perform dimensionality reduction, preserving key features. The LSTM network, comprising multiple hidden and fully connected layers, analyzes the temporal dynamic changes of whole-brain signals, capturing the temporal evolution of whole-brain consciousness states.
[0071] The fusion of the two types of features uses a weighting formula. ,in, To integrate the evaluation weights, For dynamic adjustment coefficients, For the basic weights of the nucleus signal, This represents the baseline weighting of whole-brain signals. Because signals from the lesion area in brain tumor patients are of higher reference value for assessing anesthesia status, the baseline weighting of nucleus signals is higher than that in ordinary patients. The calibration parameters are adjusted in real time based on intraoperative displacement. When intraoperative displacement causes changes in the signal acquisition quality of the nucleus, Adjustments will be made accordingly to ensure that the results of the integrated assessment are always based primarily on the most reliable signals.
[0072] The SHAP algorithm is used to generate a heatmap of signal contributions, and the proportion of feature contributions is determined by the formula. Calculate, where, The contribution percentage of the i-th feature. Let be the SHAP value of the i-th feature. The sum of the SHAP values for all features involved in the assessment is calculated. This formula quantifies the contribution of specific frequency band signals from the amygdala, abnormal discharge characteristics, and various parameters of whole-brain signals to the assessment results. Presented visually through a heatmap, it allows medical staff to clearly understand the logic behind the assessment results, particularly the impact of signals from lesion areas. The module ultimately outputs a quantitative value for depth of consciousness and an assessment of brain functional status, providing core data for early warning and medication recommendations.
[0073] Parameter adaptation module
[0074] The parameter adaptation module provides customized parameter configurations for brain tumor patients, ensuring the accuracy of system monitoring. The module pre-stores dedicated parameter templates for brain diseases such as brain tumors. These templates are generated based on extensive clinical data from brain tumor patients and cover key aspects such as signal acquisition thresholds, feature extraction parameters, and fusion weight benchmark values. This allows for precise matching of the physiological signal characteristics of brain tumor patients caused by tumor compression and structural abnormalities.
[0075] Preoperatively, medical staff input the patient's preoperative brain imaging information. The system automatically matches a corresponding brain tumor-specific parameter template based on the tumor's location, size, and compression of the amygdala in the images. During the operation, the module dynamically fine-tunes the template parameters based on real-time acquired signal characteristics. For example, when a change in the amplitude or frequency of abnormal discharge characteristics is detected, the system automatically adjusts the feature extraction threshold to ensure accurate feature recognition. Simultaneously, the module allows medical staff to manually fine-tune parameters based on individual patient information such as age, weight, and type of surgery. The basic parameters in the template are directly linked to the weight allocation formula of the two-dimensional fusion assessment module. Parameter adjustments are applied in real-time to the fusion assessment process, ensuring that the assessment results accurately match the real-time state of the brain tumor patient.
[0076] Early warning module
[0077] The early warning module provides tiered early warning services for brain tumor patients based on the dual-dimensional fusion assessment results and the feature extraction results from the signal preprocessing module. The module's early warning trigger threshold is consistent with the abnormal discharge judgment threshold of the signal preprocessing module. When a condition is detected that meets the criteria... When abnormal discharge characteristics or a quantified value of consciousness depth exceeds a preset range, the system automatically triggers a graded warning.
[0078] The tiered early warning system includes three forms: Level 1 alerts are indicated by a green indicator light, informing medical staff of a suspected abnormality; Level 2 alerts are indicated by a yellow indicator light and a specific frequency sound, indicating a mild abnormality; and Level 3 alerts are indicated by a red indicator light, a high-frequency sound, and a pop-up window on the screen, indicating a severe abnormality. The early warning lead time is set within a reasonable range to allow sufficient intervention time for medical staff, and the alert sensitivity can be adjusted as needed based on the progress of the surgery and the patient's condition. The early warning module is linked with the dual-dimensional fusion assessment module, which can identify the source of abnormal signals and key influencing factors through feature contribution heatmaps, assisting medical staff in quickly determining the cause of the abnormality and taking targeted intervention measures.
[0079] Anesthesia Dosing Coordination Recommendation Module
[0080] The anesthesia dosing recommendation module generates personalized dosing suggestions, taking into account the potential impact of anesthetic drug metabolism on brain tumor patients. The module includes a large database of clinical cases, linking monitoring results with anesthetic drug dosages across different brain tumor types and surgical scenarios. This data has been clinically validated to ensure the safety and effectiveness of the recommendations.
[0081] Dosing recommendations are generated using a gradient boosting tree algorithm, which takes a dual-dimensional fusion assessment result, patient age, weight, surgical type, and brain tumor type as input features. The gradient boosting tree algorithm optimizes model parameters through iterative training, fully considering individual differences and disease specificity in brain tumor patients, uncovering the complex relationship between input features and anesthetic dosage, and generating accurate dosing recommendations with clear adjustment directions and magnitudes. Healthcare professionals can manually revise the recommendations based on actual clinical conditions; the revised data is fed back into the database for incremental training to continuously optimize the model and improve the recommendations' adaptability. This module is directly linked to the dual-dimensional fusion assessment module; real-time changes in the assessment results drive dynamic adjustments to the dosing recommendations, ensuring that the recommendations remain synchronized with the patient's anesthesia status.
[0082] This embodiment, tailored to the specific characteristics of brain tumor patients, achieves precise monitoring and personalized intervention of intraoperative anesthesia level of consciousness through the precise synergy of various modules and disease-specific adaptation. The targeted acquisition and displacement calibration of the signal acquisition module address the impact of brain structural abnormalities and intraoperative displacement; the noise suppression and feature extraction of the signal preprocessing module ensure signal quality; the interpretability analysis of the dual-dimensional fusion assessment module enhances the reliability of the assessment results; and the parameter adaptation module and the anesthesia administration coordination suggestion module ensure the system's suitability for brain tumor patients and its clinical applicability.
[0083] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. An anesthesia and depth of consciousness monitoring system, characterized in that, The system includes: Signal acquisition module: includes an array of sensors fabricated using MEMS technology, a miniature intraoperative ultrasound probe, and a differential input acquisition circuit. The sensor surface is coated with a polylactic acid-based conductive polymer film, and the acquisition channels are independently adjustable. Signal preprocessing module: It adopts a three-level noise reduction scheme of adaptive power frequency filtering, wavelet threshold noise reduction and electromyographic noise separation to extract specific frequency band signals of amygdala and epileptic abnormal discharge characteristics, generate standardized feature vectors, and has a built-in temperature compensation module. Whole brain signal synchronization module: includes multi-channel EEG acquisition electrodes, with a sampling rate consistent with the signal acquisition module, extracts relevant EEG assessment parameters, and calibrates signal timing through a timestamp synchronization mechanism; Dual-dimensional fusion assessment module: It adopts a fusion architecture of CNN, LSTM and SHAP, and assigns weights of nucleus signals and whole brain signals based on the patient's brain images and disease type, and outputs quantitative values of consciousness depth, brain functional status assessment results and signal contribution heatmaps. Early warning module: When a specific epileptic abnormal discharge characteristic or a quantified value of the depth of consciousness exceeds a preset range, a graded early warning is triggered; Parameter adaptation module: Pre-stores parameter templates specific to brain diseases, which can be automatically matched based on the patient's preoperative brain imaging information; Anesthesia Dosing Coordination Recommendation Module: Includes a clinical case database, outputs dosing recommendations through gradient boosting tree algorithm, and supports manual correction and feedback optimization of the model.
2. The anesthesia and depth of consciousness monitoring system according to claim 1, characterized in that, The MEMS sensor array is arranged in a 3×3 configuration and adopts a flexible design with a contact resistance of ≤5kΩ. The frequency of the miniature intraoperative ultrasound probe is ≥10MHz. When the amygdala displacement is detected to be ≥2mm, the sensor acquisition range and sensitivity are finely adjusted. The displacement calibration is achieved by a registration algorithm based on image feature point matching between preoperative brain images and intraoperative ultrasound images.
3. The anesthesia and depth of consciousness monitoring system according to claim 1, characterized in that, The differential input acquisition circuit of the signal acquisition module is used to suppress common-mode interference. The acquisition channel adjustment range is adapted to the individual differences in the amygdala position of different patients. After adjustment, the acquisition parameters are adapted to the feature extraction process of the signal preprocessing module.
4. The anesthesia and depth of consciousness monitoring system according to claim 1, characterized in that, The signal preprocessing module separates electromyographic noise through independent component analysis, and the determination of epileptic abnormal discharge characteristics is based on the following formula: ,in, To detect the amplitude of the signal, To detect the frequency of the signal, The threshold for amplitude determination. The frequency determination threshold is used to extract parameters including the amplitude, frequency, and duration of spikes and sharp waves. The temperature compensation module is adapted to ambient temperature changes of ±5℃.
5. The anesthesia and depth of consciousness monitoring system according to claim 1, characterized in that, The multi-channel EEG acquisition electrodes of the whole-brain signal synchronization module have 8 channels and are deployed in the prefrontal, parietal, and temporal lobe regions respectively. The timestamp synchronization mechanism ensures that the acquisition time deviation between the nucleus signals and the whole-brain signals is ≤1ms. The synchronized signals are directly used as input data for the dual-dimensional fusion evaluation module.
6. The anesthesia and depth of consciousness monitoring system according to claim 1, characterized in that, The weighting of nucleus signals and whole-brain signals in the dual-dimensional fusion assessment module is based on the following formula: ,in, To integrate the evaluation weights, For dynamic adjustment coefficients, For the basic weights of the nucleus signal, As the basic weight of whole-brain signals, for ordinary patients Values range from 30% to 40%, for patients with brain diseases. Values range from 40% to 50%. Adjust in real time according to the intraoperative displacement calibration parameters.
7. The anesthesia and depth of consciousness monitoring system according to claim 1, characterized in that, The dual-dimensional fusion evaluation module's CNN network contains 3 convolutional layers and 2 pooling layers, while the LSTM network contains 2 hidden layers and 1 fully connected layer. The signal contribution heatmap generated by the SHAP algorithm calculates the feature contribution ratio using the following formula: ,in, The contribution percentage of the i-th feature. Let be the SHAP value of the i-th feature. The sum of SHAP values for all features involved in the evaluation is used to indicate the contribution percentage of the 4-12Hz theta wave, the 13-30Hz beta wave, and the epileptic abnormal discharge features.
8. The anesthesia and depth of consciousness monitoring system according to claim 1, characterized in that, The parameter adaptation module's exclusive parameter template is generated based on clinical data from over 500 patients with corresponding diseases, covering epilepsy, brain tumors, and traumatic brain injury, as well as different severity levels. It supports manual fine-tuning based on patient age, weight, and surgical type. The template's basic parameters are directly linked to the weight allocation formula of the dual-dimensional fusion assessment module.
9. The anesthesia and depth of consciousness monitoring system according to claim 1, characterized in that, The warning module provides three tiered warning systems: Level 1 warnings are indicated by a green indicator light, Level 2 warnings by a yellow indicator light and a 1Hz sound, and Level 3 warnings by a red indicator light, a 2Hz high-frequency sound, and a pop-up window on the screen. The warning lead time is ≥5 seconds, the warning sensitivity can be adjusted as needed, and the warning trigger threshold is consistent with the epileptic abnormal discharge judgment threshold of the signal preprocessing module.
10. The anesthesia and depth of consciousness monitoring system according to claim 1, characterized in that, The clinical case database of the anesthetic drug administration collaborative recommendation module contains 1000+ clinical cases. The database correlates the monitoring results with the anesthetic drug dosage, and the output drug administration recommendations clearly indicate the direction and extent of adjustment. The clinical validation accuracy is ≥90%. The gradient boosting tree algorithm takes the dual-dimensional fusion evaluation results, patient age, weight, surgical type, and brain disease type as input features, and generates personalized drug administration recommendations by iteratively training and optimizing model parameters. These recommendations are directly related to the output results of the dual-dimensional fusion evaluation module.