Canalis spinalis debonding method and system for real-time state evaluation

Abstract

The invention relates to the technical field of intraoperative intelligent monitoring, and particularly discloses a spinal canal debonding method and system for real-time state evaluation, and the method comprises the steps: collecting neural structure microcirculation blood flow data in real time through integrating a micro-sensing unit at a working end part of a surgical instrument, and extracting blood flow oscillation frequency and capillary hematocrit parameters; performing multi-scale signal decomposition on the blood flow oscillation frequency to obtain an energy distribution parameter, and performing dynamic mode recognition on the hematocrit data to obtain a main variation characteristic value; fusing the two characteristic parameters to generate a nervous tissue vitality index; inputting the index into a pre-trained neural network for time sequence feature analysis, and outputting a graded ischemia risk early warning signal; the operation rhythm is automatically adjusted through a decision tree control mechanism based on the early warning signal, and self-adaptive regulation and control according to the state of the nervous tissue are achieved; the technical problem that the risk of neural ischemia cannot be monitored in real time in a traditional spinal canal release operation is solved, and occult nerve injury caused by asymptomatic neural ischemia can be prevented.

Description

Technical Field

[0001] This invention relates to the field of intraoperative intelligent monitoring technology, specifically to a method and system for real-time status assessment of spinal canal release. Background Technology

[0002] Spinal decompression is a key surgical procedure for treating spinal stenosis and related nerve compression diseases. Traditional methods rely heavily on preoperative imaging data and the surgeon's experience. In current techniques, surgeons use high-speed drills and curettes to remove hypertrophic bone and thickened ligaments, relying entirely on visual observation and tactile feedback. This subjective experience-based approach has significant limitations: surgeons cannot perceive the microscopic interaction between surgical instruments and nerve tissue in real time, nor can they quantify changes in the physiological state of nerve tissue. Current techniques lack the ability to monitor the microcirculation of nerve tissue in real time, making it difficult to accurately control the force and duration of manipulation during surgery. Postoperative neurological function assessments for verifying surgical effectiveness also exhibit significant lag.

[0003] During delicate procedures such as lumbar lateral recess decompression, the continuous compression of nerve roots and microvascular plexuses by surgical instruments may gradually reduce blood supply to nerve tissue. However, due to individual differences in ischemia tolerance thresholds, this decrease in blood supply may not cause obvious clinical symptoms or morphological changes before reaching a critical point. Especially for patients with microvascular disease, their nerve ischemia tolerance threshold is significantly reduced, and surgeons cannot perceive this insidious pathophysiological change using traditional methods. When postoperative tissue edema and other factors combine, nerve tissue already in a critical ischemic state may suffer irreversible damage, manifesting as delayed foot drop and other severe neurological dysfunctions. This damage is characterized by its insidious and delayed nature, and traditional techniques lack effective early warning and prevention methods. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for real-time status assessment of spinal canal release, in order to solve the problems mentioned above.

[0005] The objective of this invention can be achieved through the following technical solutions: A real-time status assessment method for spinal canal release, S1: During the spinal canal release operation, microcirculation blood flow data of the target nerve structure is continuously collected by a microsensor unit integrated into the working end of the surgical instrument, and two parameters, blood flow oscillation frequency and capillary hematocrit, are extracted from the microcirculation blood flow data; S2: Perform multi-scale signal decomposition on blood flow oscillation frequency and extract its energy distribution parameters in the characteristic frequency band; perform dynamic pattern recognition on capillary hematocrit and calculate its principal variation characteristic value over time; fuse the energy distribution parameters with the principal variation characteristic value to generate the neural tissue vitality index. S3: Input the neural tissue vitality index into the pre-trained neural ischemia model and output the neural ischemia risk level and corresponding early warning signal in real time; S4: Automatically adjust the pace of spinal canal release operation based on the warning signal. When a high-risk warning is output, immediately pause the current operation and continue the spinal canal release operation based on the updated safety parameters after the nerve tissue vitality index returns to the normal range.

[0006] As a further aspect of the present invention: the extraction of the two parameters, blood flow oscillation frequency and capillary hematocrit, from microcirculation blood flow data specifically includes: The collected microcirculation blood flow data were processed by multi-channel signal separation to eliminate tissue motion artifacts and obtain pure hemodynamic signals. Spectral feature mining was performed on the separated microcirculation blood flow data to identify characteristic oscillation modes in the signal, and the ratio of its fundamental frequency to its third harmonic frequency was extracted as the blood flow oscillation frequency. The time-frequency domain features of the preprocessed optical signal were analyzed. By analyzing the differences in absorption characteristics of light signals of different wavelengths and combining them with a blood optical parameter database, the hematocrit at the capillary level was calculated.

[0007] As a further aspect of the present invention: the process of obtaining the energy distribution parameters is as follows: Based on the time-frequency characteristics of blood flow oscillation signals, a set of basis functions with different oscillation characteristics are generated. The time-frequency characteristics of these basis functions match the physiological mechanism of neurovascular coupling. The blood flow oscillation signal is decomposed by projection onto the basis function to obtain signal components at different time scales; The energy value of each signal component in each characteristic frequency band is calculated, and the ratio of high-frequency band energy to low-frequency band energy is calculated as the energy distribution parameter.

[0008] As a further aspect of the present invention: the process of obtaining the principal mutation feature value is as follows: A fixed time window is used to segment continuously collected capillary hematocrit data to construct a data matrix that includes time and spatial distribution dimensions; Feature pattern analysis is performed on the data matrix to identify the mutation patterns with the greatest contribution, and the projection coefficients of the corresponding mutation patterns on the time axis are extracted as the main mutation trajectories. The main variant trajectory is subjected to change characteristic analysis, the mean and variance of its trajectory curvature are calculated, and the product of the two is used as the main variant characteristic value.

[0009] As a further aspect of the present invention: the neural tissue vitality index specifically includes: Construct a nonlinear transformation function between energy distribution parameters and principal variability eigenvalues; By mapping energy distribution parameters and principal variability eigenvalues ​​to a two-dimensional feature space, the baseline region under normal physiological conditions is determined. The neural tissue vitality index is generated by calculating the relative distance of the current energy distribution parameters and the combination of principal variability eigenvalues ​​in the feature space to the center point of the baseline region. The neural tissue vitality index was normalized so that its value ranged from zero to one.

[0010] As a further aspect of the present invention: the construction process of the neural ischemia model is as follows: We collected clinical surgical data from multiple medical centers, including neural tissue viability index sequences, intraoperative physiological parameter changes, and postoperative neurological function assessment results. The collected neural tissue viability index sequences were time-aligned and standardized, and the corresponding ischemic risk levels were labeled according to the postoperative neurological function assessment results. Construct a deep neural network with temporal feature extraction capabilities; The deep learning network was trained using a preprocessed training dataset, and the network parameters were adjusted through cross-validation to ensure the accuracy of the neural ischemia model in identifying different types of ischemic events.

[0011] As a further aspect of the present invention: the output of the nerve ischemia risk level and the corresponding early warning signal specifically includes: The continuously collected neural tissue vitality indices are arranged into an input sequence in chronological order to preserve their dynamic change characteristics; The input sequence is fed into a pre-trained neural ischemia model, and the multi-layer feature extraction network in the neural ischemia model identifies feature patterns related to ischemia risk. Based on the feature pattern matching results output by the neural ischemia model, the corresponding neural ischemia risk level and early warning signal are output according to the preset risk level classification standard.

[0012] As a further aspect of the present invention: the automatic adjustment of the spinal canal release operation rhythm based on the warning signal specifically includes: A real-time decision tree is constructed based on the risk level of nerve ischemia, and corresponding operation control strategies are triggered according to different risk levels. When a medium-risk warning is detected, the frequency of instrument operation is automatically reduced and the duration of a single operation is shortened; when a high-risk warning is detected, the current operation is immediately interrupted and safety protection is activated. During the operation pause, the recovery trend of the neural tissue vitality index is continuously monitored, and the safe time window and operation intensity for subsequent operations are dynamically calculated based on the recovery rate. Once the nerve tissue vitality index has stabilized within a safe range, the release procedure is restarted with gradually increasing intensity, and the stability of the vitality index is verified in real time.

[0013] A spinal canal release system for real-time status assessment, comprising: The real-time monitoring and feature extraction module continuously collects microcirculation blood flow data of the target nerve structure through a microsensor unit integrated into the working end of the surgical instrument, and extracts two parameters, blood flow oscillation frequency and capillary hematocrit, from the microcirculation blood flow data. The multi-dimensional feature fusion analysis module performs multi-scale signal decomposition on blood flow oscillation frequency and extracts its energy distribution parameters within the characteristic frequency band; performs dynamic pattern recognition on capillary hematocrit and calculates its principal variation characteristic value over time; and fuses the energy distribution parameters with the principal variation characteristic value to generate a neural tissue vitality index. The intelligent early warning decision module inputs the neural tissue vitality index into the pre-trained neural ischemia model and outputs the neural ischemia risk level and corresponding early warning signal in real time. An adaptive safety operation module automatically adjusts the rhythm of spinal canal release operation based on warning signals. When a high-risk warning is output, the current operation is immediately paused, and the spinal canal release operation continues based on updated safety parameters after the nerve tissue vitality index returns to the normal range.

[0014] The beneficial effects of this invention are: (1) By extracting two deep physiological parameters, blood flow oscillation frequency and capillary hematocrit, from microcirculatory blood flow data, and using multi-scale signal decomposition and dynamic pattern recognition technology to generate a neural tissue vitality index, an intelligent early warning system based on a neural network model was established. This system can identify latent precursors of nerve ischemia that cannot be detected by traditional methods, and issue graded warnings before irreversible damage to nerve function occurs, enabling the surgical team to adjust their operating strategies in a timely manner, transforming the traditional "post-operative treatment" into "pre-operative prevention," and reducing the risk of postoperative neurological dysfunction.

[0015] (2) The decision tree control system based on real-time risk levels can automatically adjust the operation frequency, duration, and intensity according to the state of the neural tissue. When there is a high risk, the operation is immediately interrupted and safety protection is activated. After the neural tissue recovers, the operation is resumed according to a gradual strategy. This closed-loop control mechanism effectively solves the problem of adjustment lag caused by the reliance on the doctor's subjective judgment in traditional methods. It not only ensures the safety of the neural tissue but also maintains the efficiency of the operation, providing standardized and quantifiable safe operation specifications for spinal canal release surgery. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart of the system of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Please see Figure 1 As shown, this invention is a method for real-time status assessment and spinal canal release, comprising the following steps: S1: During the spinal canal release procedure, the microcirculation blood flow data of the target nerve structure is continuously collected by the microsensor unit integrated into the working end of the surgical instrument, and two parameters, blood flow oscillation frequency and capillary hematocrit, are extracted from the microcirculation blood flow data. S2: Perform multi-scale signal decomposition on blood flow oscillation frequency and extract its energy distribution parameters in the characteristic frequency band; perform dynamic pattern recognition on capillary hematocrit and calculate its principal variation characteristic value over time; fuse the energy distribution parameters with the principal variation characteristic value to generate the neural tissue vitality index. S3: Input the neural tissue vitality index into the pre-trained neural ischemia model and output the neural ischemia risk level and corresponding early warning signal in real time; S4: Automatically adjust the pace of spinal canal release operation based on the warning signal. When a high-risk warning is output, immediately pause the current operation and continue the spinal canal release operation based on the updated safety parameters after the nerve tissue vitality index returns to the normal range.

[0020] In S1, during the spinal canal release procedure, microcirculatory blood flow data of the target nerve structure is continuously collected by a microsensor unit integrated into the working end of the surgical instrument. Two parameters, blood flow oscillation frequency and capillary hematocrit, are extracted from the microcirculatory blood flow data, specifically including: During spinal canal release, microcirculatory blood flow data of the target nerve structure is continuously acquired via a microsensor unit integrated into the working end of the surgical instrument. This microsensor unit comprises a laser Doppler flowmeter and a dual-wavelength optical sensing unit. The laser Doppler flowmeter transmits data to the nerve tissue surface via optical fiber to receive backscattered light signals. The dual-wavelength optical sensing unit is used to detect the absorption characteristics of oxyhemoglobin and deoxyhemoglobin. During acquisition, the sensing unit continuously records optical signals at a sampling frequency of 100 times per second to ensure the capture of rapid changes in microcirculatory blood flow. All acquired data is transmitted to the signal processing unit via a shielded cable, and digital filtering technology is used to eliminate electromagnetic interference during transmission.

[0021] Multi-channel signal separation processing was performed on the acquired microcirculation blood flow data to eliminate tissue motion artifacts. First, a signal separation system with three input channels was established to receive the original optical signal, the instrument motion signal, and the tissue contact pressure signal, respectively. An adaptive filtering algorithm was used to separate interference signals generated by instrument motion and tissue movement from the original optical signal. The specific processing steps included: first, calculating the correlation coefficient between the motion signal and the optical signal; when the correlation coefficient exceeded 0.8, significant motion interference was identified; then, a recursive least squares algorithm was used to dynamically update the filter coefficients to achieve real-time elimination of motion artifacts; finally, phase correction was performed on the processed signal to ensure the temporal consistency of the signals in each channel. After these processing steps, a pure hemodynamic signal was obtained, which accurately reflects the true state of the neural tissue microcirculation.

[0022] Spectral feature mining was performed on the separated microcirculatory blood flow data to identify characteristic oscillation modes in the signal. First, the power spectral density of the preprocessed blood flow signal was estimated using a modified periodogram method, segmented using a 5-second Hanning window with 50% overlap. In the spectral analysis, the myogenic oscillation band (0.05 Hz to 0.15 Hz) and the neurogenic oscillation band (0.15 Hz to 0.4 Hz) were focused on. The dominant frequency components within each band were identified using a peak detection algorithm, and their amplitude and phase information were calculated. Then, harmonic analysis was used to extract the amplitude ratio of the fundamental frequency component and the third harmonic component, which was used as a blood flow oscillation frequency parameter. This parameter reflects the integrity of vascular regulation; when there is a risk of ischemia in nerve tissue, this ratio will exhibit characteristic changes.

[0023] The preprocessed optical signal was analyzed in the time-frequency domain to calculate the capillary hematocrit. First, the light absorption signal was differentially processed to eliminate the influence of tissue background absorption. Then, according to the Lambert-Beer law, the ratio of light absorbance at two wavelengths was calculated, and this ratio has a definite correspondence with hematocrit. To improve calculation accuracy, a database containing optical parameters under different physiological states was established, covering optical characteristics under different ages, sexes, and pathological conditions. The preliminary calculation results were corrected by querying this database to obtain accurate capillary hematocrit values. The influence of tissue scattering coefficient was also considered during the calculation, and an iterative algorithm was used to gradually approximate the true value to ensure the reliability of the calculation results.

[0024] In S2, multi-scale signal decomposition is performed on the blood flow oscillation frequency to extract its energy distribution parameters within the characteristic frequency band; dynamic pattern recognition is performed on capillary hematocrit to calculate its principal variance characteristic value over time; the energy distribution parameters and principal variance characteristic values ​​are fused to generate a neural tissue vitality index, specifically including: Multi-scale signal decomposition was performed on the blood flow oscillation frequency to extract its energy distribution parameters within the characteristic frequency band. First, based on the time-frequency characteristics of the blood flow oscillation signal, a set of basis functions with different oscillation characteristics were generated. The construction process of these basis functions considered the physiological mechanism of neurovascular coupling, and the oscillation frequency range of the basis functions was set between 0.01 Hz and 0.5 Hz, covering major physiological oscillation modes such as endothelial cell-dependent oscillations, myogenic oscillations, and neurogenic oscillations. Each basis function has specific time-frequency localization characteristics, and its duration is adjustable between 3 and 30 seconds to adapt to blood flow regulation activities at different time scales. The number of basis functions was set to eight, each corresponding to a different physiological regulatory mechanism. These basis functions were generated through an iterative optimization process to ensure effective characterization of the characteristic components of the blood flow oscillation signal.

[0025] The blood flow oscillation signal is decomposed by projection onto basis functions to obtain signal components at different time scales. The projection decomposition process employs point-by-point matching to calculate the similarity coefficient between the original signal and each basis function. For a 10-second signal segment, the projection coefficients with each of the eight basis functions are calculated, resulting in eight sets of decomposed signal components. Each signal component represents the contribution of a specific physiological mechanism to the total blood flow oscillation. During the decomposition process, an overlapping segmentation method is used, with 50% overlap between adjacent signal segments to ensure the continuity of the decomposition. For each signal segment, its projection coefficients on different basis functions are recorded; these coefficients constitute the characteristic representation of the blood flow oscillation signal within that time segment. The decomposition process is continuous, with the results updated every 0.1 seconds, achieving real-time decomposition of the blood flow oscillation signal throughout the entire surgical procedure.

[0026] The energy values ​​of signal components within each characteristic frequency band are calculated, and the ratio of high-frequency band energy to low-frequency band energy is calculated as the energy distribution parameter. Energy calculation employs the square integral method, summing the squares of the amplitude of each signal component within a 1-second time window. The eight basis functions are divided into two groups: the signal components corresponding to the four basis functions with frequencies below 0.1 Hz are classified as the low-frequency band, and the signal components corresponding to the four basis functions with frequencies above 0.1 Hz are classified as the high-frequency band. The sum of the energy of all signal components in the low-frequency band and the sum of the energy of all signal components in the high-frequency band are calculated separately. Then, the high-frequency band energy value is divided by the low-frequency band energy value to obtain the energy distribution parameter. This parameter is updated every 0.5 seconds to reflect the real-time changes in the energy distribution of different frequency components during blood flow oscillations. When there is a risk of ischemia in nerve tissue, the high-frequency band energy shows a characteristic decrease relative to the low-frequency band energy, resulting in a decrease in the energy distribution parameter.

[0027] Dynamic pattern recognition was performed on capillary hematocrit to calculate its principal variation features over time. First, continuously acquired capillary hematocrit data were segmented into segments within fixed time windows, constructing a data matrix containing both temporal and spatial dimensions. The time window length was set to 5 seconds, with each window containing 500 sampling points. Rows in the data matrix represent the time series, and columns represent measurements at different spatial locations. For each time window, a 500-row, 10-column data matrix was constructed, where the 10 columns represent hematocrit data collected from 10 different sensor points at the instrument tip. Each element of the matrix represents the hematocrit measurement at a specific location at a specific time. The data matrix was updated every 0.5 seconds, retaining all hematocrit data from the most recent 5 seconds, forming a continuously sliding time window.

[0028] Feature pattern analysis was performed on the data matrix to identify the most contributing variation patterns. The projection coefficients of these patterns on the time axis were extracted as the principal variation trajectory. The feature pattern analysis employed singular value decomposition (SVD) to decompose the data matrix into the product of three matrices. By calculating the contribution of each feature pattern, the feature pattern with the highest contribution was selected as the dominant variation pattern. The projection coefficients of the dominant variation pattern constitute the principal variation trajectory, which is 500 points long and corresponds to the variation pattern over a 5-second time period. The principal variation trajectory reflects the main temporal and spatial variation characteristics of hematocrit, eliminating the influence of random fluctuations and measurement noise. During the calculation process, feature pattern analysis was performed once a new data matrix was obtained to ensure that the principal variation trajectory was updated in real time and accurately reflected the dynamic trend of hematocrit.

[0029] The principal variant trajectory was analyzed for variation characteristics, and the mean and variance of its trajectory curvature were calculated. Their product was used as the principal variant eigenvalue. Curvature calculation employed a difference geometry method, calculating the instantaneous curvature of the trajectory at each time point. First, cubic spline interpolation was performed on the principal variant trajectory to obtain a smooth curve representation. Then, the first and second derivatives of the curve were calculated, and the curvature value at each point was calculated using the curvature formula. After obtaining 500 curvature values, the arithmetic mean of these curvature values ​​was calculated as the curvature mean, and the variance of the curvature values ​​was calculated as the curvature variance. Finally, the curvature mean and curvature variance were multiplied to obtain the principal variant eigenvalue. This eigenvalue reflects both the degree and consistency of curvature of the principal variant trajectory, effectively characterizing the regularity of changes in hematocrit. When the microcirculation state changes, the principal variant eigenvalue will show corresponding changes.

[0030] A neural tissue viability index is generated by fusing energy distribution parameters with principal variability eigenvalues. First, a nonlinear transformation function between the energy distribution parameters and principal variability eigenvalues ​​is constructed. This function is built using a piecewise linear approximation method, employing a steeper slope in regions where the energy distribution parameter is less than 0.5 and a gentler slope in regions where the energy distribution parameter is greater than 0.5. The specific parameters of the transformation function are determined through experimental data to ensure accurate reflection of the physiological correlation between the two parameters. The output of the transformation function is an intermediate variable that integrates information from both the energy distribution parameters and principal variability eigenvalues, providing a foundation for subsequent feature space mapping.

[0031] Energy distribution parameters and principal variability eigenvalues ​​are mapped onto a two-dimensional feature space to determine the baseline region under normal physiological conditions. The horizontal axis of the feature space represents the energy distribution parameters, and the vertical axis represents the principal variability eigenvalues. By analyzing a large amount of data under normal physiological conditions, the coordinates of the center point and the boundary range of the baseline region are determined. The baseline region is elliptical, with its major axis along the direction of the energy distribution parameters and its minor axis along the direction of the principal variability eigenvalues. The coordinates of the center point are obtained by calculating the mean of the two parameters under normal conditions, and the boundary range is determined by calculating the standard deviation of the two parameters. The baseline region is updated every 30 minutes to adapt to the slow changes in individual physiological parameters. In the feature space, points closer to the center point of the baseline region indicate better neural tissue vitality.

[0032] A neural tissue viability index is generated by calculating the relative distance between the current energy distribution parameters and the main variability eigenvalue combination in the feature space and the center point of the baseline region. The relative distance is calculated using the Mahalanobis distance method, considering the correlation and variability of the two parameters. First, the differences between the current parameter combination and the center point of the baseline region along the horizontal and vertical axes are calculated, and then weighted according to the covariance matrix of the baseline region. The calculated Mahalanobis distance is converted into a viability index through a nonlinear function. The conversion function ensures that the smaller the distance, the closer the viability index is to 1, and the larger the distance, the closer the viability index is to 0. The viability index is updated every 0.5 seconds, reflecting the functional state of neural tissue in real time. When the viability index is below 0.3, it indicates a high risk in the neural tissue.

[0033] The neural tissue viability index is normalized to a value between zero and one. The normalization process uses a minimum-maximum method, setting the theoretical minimum value to 0 and the theoretical maximum value to 1. In actual calculations, the value is 0 when the viability index is less than 0 and 1 when it is greater than 1. The normalized viability index is easier to interpret and apply clinically; a value closer to 1 indicates better neural tissue viability, while a value closer to 0 indicates worse viability. The normalization process also includes a moving average filter, using a 3-second moving window to smooth the viability index, eliminating the influence of instantaneous fluctuations and providing stable output results. The processed viability index is displayed in real-time through a graphical interface, providing surgeons with intuitive feedback on the neural state.

[0034] In S3, the neural tissue viability index is input into the pre-trained neural ischemia model, which outputs the neural ischemia risk level and corresponding early warning signals in real time, specifically including: The construction of the neuroischemia analysis system began with data collection. Clinical surgical data from multiple medical centers were collected, including continuously recorded neural tissue viability index sequences, intraoperative arterial blood pressure, blood oxygen saturation, and other physiological parameter changes, as well as neurological function assessments performed within 24 hours postoperatively. Data collection lasted 12 months, acquiring complete surgical records for 300 cases. Each record included neural tissue viability index data throughout the entire surgery, with a sampling interval of 0.5 seconds. Intraoperative physiological parameter recordings included arterial blood pressure, heart rate, and blood oxygen saturation values ​​recorded every minute. Postoperative neurological function assessments used a standardized scoring scale, conducted at 6, 12, and 24 hours postoperatively, evaluating motor, sensory, and reflex functions. All data was anonymized during collection, removing patient-identified information and retaining only surgically relevant medical data.

[0035] The collected neural tissue viability index sequences underwent time alignment and standardization. Time alignment used the start of surgery as the baseline, unifying the time axes of all physiological parameters to the same time coordinate system. For data with different sampling frequencies, cubic spline interpolation was used to resample the data onto a unified time grid with a grid interval of 0.5 seconds. Standardization involved two steps: first, z-score standardization was performed on the neural tissue viability index sequence for each case, subtracting the mean and dividing by the standard deviation; then, min-max standardization was performed, mapping the values ​​to the range of 0 to 1. The corresponding ischemic risk level was labeled according to the postoperative neurological function assessment results: cases with a postoperative neurological function score decrease of more than 30% were labeled as high-risk, decreases of 10% to 30% as medium-risk, and decreases of less than 10% as low-risk. The labeling work was completed independently by three senior neurosurgeons, and cases with differing opinions were discussed to reach a consensus.

[0036] A data processing network with temporal feature extraction capabilities was constructed. This network employs a hierarchical structure with four processing layers. The first layer is the input layer, receiving a 60-second sequence of neural tissue viability index data, containing 120 data points. The second layer is the feature extraction layer, containing 32 processing units, each connected to a 5-second sliding window, extracting local temporal features through convolutional operations. The third layer is the feature integration layer, which reduces the dimensionality of the features output from the second layer, retaining the 16 most important feature dimensions. The fourth layer is the classification output layer, mapping the integrated features to probability distributions for three risk levels. The connection weights between layers are determined during training, and each processing unit undergoes a non-linear transformation using the ReLU activation function. The network structure contains approximately 5000 adjustable parameters, which are progressively optimized during training.

[0037] The data processing network was trained using a preprocessed training dataset. Stochastic gradient descent was employed during training, with each training iteration using 32 samples per batch. The number of training iterations was set to 1000, with an initial learning rate of 0.01, halved every 200 iterations. During training, the cross-entropy loss function was used to evaluate the difference between the network output and the true label, and the network parameters were updated using backpropagation. A five-fold cross-validation method was used, randomly dividing the 300 cases into five groups. Four groups were used for training each time, with the remaining group used for validation. Cross-validation was used to adjust the network parameters, ensuring the network's accuracy in identifying different types of ischemic events. After training, the parameter combination that performed best on the validation set was selected as the final network parameters.

[0038] During the real-time analysis phase, continuously acquired neural tissue viability indices were assembled into an input sequence in chronological order. The input sequence length was fixed at 60 seconds and contained 120 continuously acquired neural tissue viability index values. A sliding window method was used for sequence construction, updating every 0.5 seconds by removing the oldest data point and adding the newest one. Quality checks were performed concurrently during sequence construction; missing values ​​were imputed using linear interpolation. The completed input sequence preserved the dynamic characteristics of the neural tissue viability index, including trend, fluctuation, and periodic features. These features are valuable for identifying early ischemic changes. Each input sequence was accompanied by a timestamp to ensure temporal synchronization with other intraoperative monitoring data.

[0039] The input sequence is fed into a trained data processing network for feature analysis. The network first preprocesses the input sequence, including detrending and standardization. Detrending uses a first-order differencing method to eliminate linear trend components in the sequence. Standardization adjusts the sequence to a distribution with a mean of 0 and a standard deviation of 1. The preprocessed sequence then enters the feature extraction stage, where processing units in the network sequentially transform the sequence, progressively extracting feature representations at different time scales. During feature extraction, the network focuses on abrupt changes, trend inflection points, and periodic fluctuation patterns in the sequence. These feature patterns are closely related to the early manifestations of neural ischemia. Through cascaded operations of a multi-layered processing structure, the network finally obtains a feature representation of the input sequence, which comprehensively reflects the sequence's characteristics in both the time and frequency domains.

[0040] Based on the feature analysis results of the network output, warning signals are generated according to a preset risk level classification standard. The risk level classification is based on the probability distribution of the network output, dividing the output results into three levels: low risk level corresponds to a low-risk probability exceeding 0.7, medium risk level corresponds to a medium-risk probability exceeding 0.6, and high risk level corresponds to a high-risk probability exceeding 0.5. The warning signals employ a multi-level coding method: low risk level corresponds to a green signal, medium risk level corresponds to a yellow signal, and high risk level corresponds to a red signal. Warning signals are generated every 0.5 seconds to ensure real-time reflection of changes in the vitality of neural tissue. The warning signals are displayed on a visual display device in the operating room, and an audible alert is provided as well. All warning signals and their corresponding timestamps are recorded in the surgical record system for postoperative analysis and summary.

[0041] In S4, the pace of spinal canal release is automatically adjusted based on warning signals. When a high-risk warning is issued, the current operation is immediately paused, and the spinal canal release operation continues based on updated safety parameters after the neural tissue vitality index returns to the normal range. Specifically, this includes: A real-time decision tree is constructed based on the risk level of neuronal ischemia, triggering corresponding operational control strategies according to different risk levels. The decision tree contains three main branches, corresponding to low-risk, medium-risk, and high-risk states, respectively. The decision tree executes once per second to ensure real-time response to changes in neuronal state. When the system detects a low-risk state, the decision tree executes the first branch, maintaining the current operating parameters unchanged, with an operation frequency of twice per second and a single operation duration of 0.8 seconds. When the system detects a medium-risk state, the decision tree executes the second branch, initiating a first-level control strategy. At this time, the system automatically records the current operation position and reduces the operation frequency to once per second, shortening the single operation duration to 0.5 seconds. When the system detects a high-risk state, the decision tree executes the third branch, initiating a second-level control strategy, immediately terminating the current operation cycle and automatically retracting the device to a safe position. Each decision node in the decision tree is equipped with time delay protection to prevent misjudgments caused by signal fluctuations. The system continuously monitors changes in risk state, updating the decision state every 0.2 seconds to ensure timely adjustment of control strategies.

[0042] When a medium-risk warning is detected, the system automatically reduces the frequency of instrument manipulation and shortens the duration of each manipulation session. The frequency adjustment is gradual: in the first cycle after a medium-risk warning, the frequency is reduced from the normal 2 times per second to 1.5 times per second, and further reduced to 1 time per second in the second cycle. The duration of each manipulation session is adjusted simultaneously, gradually shortening from the normal 0.8 seconds to 0.6 seconds, and finally stabilizing at 0.5 seconds. During frequency adjustment, the system simultaneously monitors the trend of changes in the neural tissue vitality index. If the index continues to improve, the current parameters remain unchanged; if the index continues to deteriorate, the system immediately escalates to a high-risk treatment strategy. The adjustment parameters for frequency and duration are stored in the system cache, and a timestamp and parameter values ​​before and after each adjustment are recorded for subsequent analysis and optimization. The system also sets a minimum operation frequency limit to ensure that the surgical progress is not severely affected.

[0043] When a high-risk warning is issued, the system immediately interrupts the current operation and activates the safety protection mechanism. The response time of the interruption command is controlled within 0.1 seconds, directly cutting off the output of the current operation command through hardware-level signals. The safety protection mechanism consists of three steps: First, the surgical instrument is automatically retracted 2 mm via a precision servo control system, detaching it from direct contact with the nerve tissue; second, local perfusion monitoring is activated, continuously monitoring the microcirculation status of the retracted area through built-in micro-sensors; finally, the safety locking device is activated to prevent any accidental operation from causing the instrument to advance. Throughout the safety protection process, the system continuously monitors the nerve tissue vitality index, recording the index value every 0.5 seconds and calculating its trend. If the index does not begin to recover within 5 seconds, the system will issue a secondary alarm, prompting the surgeon to intervene. The safety protection state is maintained for at least 10 seconds to ensure sufficient recovery time for the nerve tissue.

[0044] During the pause in operation, the system continuously monitors the recovery trend of the neural tissue vitality index and dynamically calculates the safe time window and operation intensity for subsequent operations based on the recovery rate. The recovery trend analysis uses linear regression to calculate the slope of the neural tissue vitality index change over the past 10 seconds. Based on the slope value, the recovery rate is divided into three levels: rapid recovery (slope greater than 0.05 / second), moderate recovery (slope between 0.02 and 0.05 / second), and slow recovery (slope less than 0.02 / second). The safe time window is calculated based on the recovery rate level: a 5-second window for rapid recovery, an 8-second window for moderate recovery, and a 12-second window for slow recovery. The operation intensity calculation considers both the recovery rate and the current vitality index value, using a two-dimensional interpolation method to determine the optimal operation intensity. Intensity levels are divided into 50%, 70%, and 100%. These parameters are updated every 2 seconds to ensure synchronization with the actual recovery state of the neural tissue.

[0045] Once the neural tissue viability index stabilizes within a safe range, the system restarts the release procedure with gradually increasing intensity. The safe range is defined as a neural tissue viability index remaining above 0.7 for 30 seconds. The restart employs a phased recovery strategy, consisting of three stages: Stage 1 is the initial recovery period, with the intensity set to 50% of the normal value and lasting at least 15 seconds; this stage focuses on observing the neural tissue's response to the procedure. Stage 2 is the gradual enhancement period, with the intensity increased to 70% of the normal value and lasting at least 20 seconds; this stage gradually establishes the operational rhythm. Stage 3 is the complete recovery period, with the intensity restored to 100%, entering normal operating mode. At each stage transition, the system performs stability verification, requiring the neural tissue viability index to fluctuate by no more than 0.1 within that stage. If the index drops by more than 0.2 in any stage, the system immediately reverts to the previous stage and restarts the stability verification. The entire process ensures that the neural tissue can smoothly adapt to the increased intensity of the procedure, avoiding the risk of recurring ischemia.

[0046] Please see Figure 2 As shown, a spinal canal release system for real-time status assessment includes: The real-time monitoring and feature extraction module continuously collects microcirculation blood flow data of the target nerve structure through a microsensor unit integrated into the working end of the surgical instrument, and extracts two parameters, blood flow oscillation frequency and capillary hematocrit, from the microcirculation blood flow data. The multi-dimensional feature fusion analysis module performs multi-scale signal decomposition on blood flow oscillation frequency and extracts its energy distribution parameters within the characteristic frequency band; performs dynamic pattern recognition on capillary hematocrit and calculates its principal variation characteristic value over time; and fuses the energy distribution parameters with the principal variation characteristic value to generate a neural tissue vitality index. The intelligent early warning decision module inputs the neural tissue vitality index into the pre-trained neural ischemia model and outputs the neural ischemia risk level and corresponding early warning signal in real time. An adaptive safety operation module automatically adjusts the rhythm of spinal canal release operation based on warning signals. When a high-risk warning is output, the current operation is immediately paused, and the spinal canal release operation continues based on updated safety parameters after the nerve tissue vitality index returns to the normal range.

[0047] The working principle of this invention is as follows: Micro-sensing units are integrated into the end of surgical instruments to collect real-time microcirculatory blood flow data of neural structures, extracting blood flow oscillation frequency and capillary hematocrit parameters. Multi-scale signal decomposition of the blood flow oscillation frequency yields energy distribution parameters, while dynamic pattern recognition of the hematocrit data obtains principal variation features. These two feature parameters are fused into a neural tissue viability index through nonlinear mapping and feature space distance calculation. This index is input into a pre-trained neural network for temporal feature analysis, outputting graded ischemic risk warning signals. Based on the warning signals, a decision tree control mechanism automatically adjusts the operation rhythm, including dynamically adjusting the operation frequency, duration, and intensity according to the risk level. In cases of high risk, the operation is immediately interrupted and safety protection is activated. Once neural tissue viability recovers, the operation is resumed using a gradual strategy, forming a complete monitoring-assessment-early warning-regulation closed-loop system.

[0048] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for spinal canal release based on real-time status assessment, characterized in that, Includes the following steps: S1: During the spinal canal release procedure, the microcirculation blood flow data of the target nerve structure is continuously collected by the microsensor unit integrated into the working end of the surgical instrument, and two parameters, blood flow oscillation frequency and capillary hematocrit, are extracted from the microcirculation blood flow data. S2: Perform multi-scale signal decomposition on blood flow oscillation frequency and extract its energy distribution parameters in the characteristic frequency band; perform dynamic pattern recognition on capillary hematocrit and calculate its principal variation characteristic value over time; fuse the energy distribution parameters with the principal variation characteristic value to generate the neural tissue vitality index. S3: Input the neural tissue vitality index into the pre-trained neural ischemia model and output the neural ischemia risk level and corresponding early warning signal in real time; S4: Automatically adjust the pace of spinal canal release operation based on the warning signal. When a high-risk warning is output, immediately pause the current operation and continue the spinal canal release operation based on the updated safety parameters after the nerve tissue vitality index returns to the normal range.

2. The method for real-time status assessment and spinal canal release according to claim 1, characterized in that, The extraction of two parameters from microcirculatory blood flow data—blood flow oscillation frequency and capillary hematocrit—specifically includes: The collected microcirculation blood flow data were processed by multi-channel signal separation to eliminate tissue motion artifacts and obtain pure hemodynamic signals. Spectral feature mining was performed on the separated microcirculation blood flow data to identify characteristic oscillation modes in the signal, and the ratio of its fundamental frequency to its third harmonic frequency was extracted as the blood flow oscillation frequency. The time-frequency domain features of the preprocessed optical signal were analyzed. By analyzing the differences in absorption characteristics of light signals of different wavelengths and combining them with a blood optical parameter database, the hematocrit at the capillary level was calculated.

3. The method for real-time status assessment and spinal canal release according to claim 1, characterized in that, The process of obtaining the energy distribution parameters is as follows: Based on the time-frequency characteristics of blood flow oscillation signals, a set of basis functions with different oscillation characteristics are generated. The time-frequency characteristics of these basis functions match the physiological mechanism of neurovascular coupling. The blood flow oscillation signal is decomposed by projection onto the basis function to obtain signal components at different time scales; The energy value of each signal component in each characteristic frequency band is calculated, and the ratio of high-frequency band energy to low-frequency band energy is calculated as the energy distribution parameter.

4. The method for real-time status assessment and spinal canal release according to claim 1, characterized in that, The process for obtaining the principal mutation feature value is as follows: A fixed time window is used to segment continuously collected capillary hematocrit data to construct a data matrix that includes time and spatial distribution dimensions; Feature pattern analysis is performed on the data matrix to identify the mutation patterns with the greatest contribution, and the projection coefficients of the corresponding mutation patterns on the time axis are extracted as the main mutation trajectories. The main variant trajectory is subjected to change characteristic analysis, the mean and variance of its trajectory curvature are calculated, and the product of the two is used as the main variant characteristic value.

5. The method for real-time status assessment and spinal canal release according to claim 1, characterized in that, The neural tissue vitality index specifically includes: Construct a nonlinear transformation function between energy distribution parameters and principal variability eigenvalues; By mapping energy distribution parameters and principal variability eigenvalues ​​to a two-dimensional feature space, the baseline region under normal physiological conditions is determined. The neural tissue vitality index is generated by calculating the relative distance of the current energy distribution parameters and the combination of principal variability eigenvalues ​​in the feature space to the center point of the baseline region. The neural tissue vitality index was normalized so that its value ranged from zero to one.

6. The method for real-time status assessment and spinal canal release according to claim 1, characterized in that, The process of constructing the neural ischemia model is as follows: We collected clinical surgical data from multiple medical centers, including neural tissue viability index sequences, intraoperative physiological parameter changes, and postoperative neurological function assessment results. The collected neural tissue viability index sequences were time-aligned and standardized, and the corresponding ischemic risk levels were labeled according to the postoperative neurological function assessment results. Construct a deep neural network with temporal feature extraction capabilities; The deep learning network was trained using a preprocessed training dataset, and the network parameters were adjusted through cross-validation to ensure the accuracy of the neural ischemia model in identifying different types of ischemic events.

7. The method for real-time status assessment and spinal canal release according to claim 1, characterized in that, The output nerve ischemia risk level and corresponding early warning signal specifically include: The continuously collected neural tissue vitality indices are arranged into an input sequence in chronological order to preserve their dynamic change characteristics; The input sequence is fed into a pre-trained neural ischemia model, and the multi-layer feature extraction network in the neural ischemia model identifies feature patterns related to ischemia risk. Based on the feature pattern matching results output by the neural ischemia model, the corresponding neural ischemia risk level and early warning signal are output according to the preset risk level classification standard.

8. The method for real-time status assessment and spinal canal release according to claim 1, characterized in that, The automatic adjustment of the spinal canal release operation rhythm based on the early warning signal specifically includes: A real-time decision tree is constructed based on the risk level of nerve ischemia, and corresponding operation control strategies are triggered according to different risk levels. When a medium-risk warning is detected, the frequency of instrument operation is automatically reduced and the duration of a single operation is shortened; when a high-risk warning is detected, the current operation is immediately interrupted and safety protection is activated. During the operation pause, the recovery trend of the neural tissue vitality index is continuously monitored, and the safe time window and operation intensity for subsequent operations are dynamically calculated based on the recovery rate. Once the nerve tissue vitality index has stabilized within a safe range, the release procedure is restarted with gradually increasing intensity, and the stability of the vitality index is verified in real time.

9. A spinal canal release system for real-time status assessment, characterized in that, A spinal canal release method for real-time status assessment as described in any one of claims 1-8, comprising: The real-time monitoring and feature extraction module continuously collects microcirculation blood flow data of the target nerve structure through a microsensor unit integrated into the working end of the surgical instrument, and extracts two parameters, blood flow oscillation frequency and capillary hematocrit, from the microcirculation blood flow data. The multi-dimensional feature fusion analysis module performs multi-scale signal decomposition on blood flow oscillation frequency and extracts its energy distribution parameters within the characteristic frequency band; performs dynamic pattern recognition on capillary hematocrit and calculates its principal variation characteristic value over time; and fuses the energy distribution parameters with the principal variation characteristic value to generate a neural tissue vitality index. The intelligent early warning decision module inputs the neural tissue vitality index into the pre-trained neural ischemia model and outputs the neural ischemia risk level and corresponding early warning signal in real time. An adaptive safety operation module automatically adjusts the rhythm of spinal canal release operation based on warning signals. When a high-risk warning is output, the current operation is immediately paused, and the spinal canal release operation continues based on updated safety parameters after the nerve tissue vitality index returns to the normal range.

Images