Adaptive correction and curve reconstruction method and system for long-term monitoring of tissue oxygen saturation

Abstract

The present application relates to the technical field of signal processing, and discloses an adaptive correction and curve reconstruction method and system for long-time monitoring of tissue oxygen saturation, comprising: obtaining original sampling data of tissue oxygen saturation and a real-time risk value of a monitoring object; determining a correction intensity coefficient negatively correlated with the real-time risk value according to the real-time risk value; performing adaptive signal correction on the original sampling data of tissue oxygen saturation based on the correction intensity coefficient and marking candidate abnormal peaks to obtain corrected monitoring data; performing credibility calculation on a statistical time window based on the marked candidate abnormal peaks to obtain a credibility index of the statistical time window; dividing a display area to obtain a historical display area and a real-time display area; performing nonlinear mapping on the corrected monitoring data in the historical display area according to the credibility index, and performing equal-time-interval linear mapping on the corrected monitoring data in the real-time display area to obtain a trend curve and output a display.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, specifically to an adaptive correction and curve reconstruction method and system for long-term monitoring of tissue oxygen saturation. Background Technology

[0002] Tissue oxygen saturation monitoring has been increasingly used in scenarios such as postoperative blood supply observation after free flap surgery, long-term ICU monitoring, and monitoring of blood supply to the limbs and microcirculation during wound recovery. The monitoring time has been extended from several hours during surgery to several days or even weeks postoperatively. Existing monitoring equipment typically uses fixed physiological thresholds to remove abnormal sampling values, median filtering with a fixed window to eliminate peak interference, and displays trend curves using equal time intervals or fixed-ratio compression.

[0003] Interference spikes caused by factors such as body movement, poor probe contact, and ambient light closely resemble the waveform of rapid tissue oxygen decline in the early stages of a vascular crisis. Currently, the equipment does not consider the patient's current risk status when processing signals, and the correction intensity is basically fixed. The real fluctuations in oxygen decline during high-risk periods are easily smoothed out as noise. In long-term displays, fixed-ratio compression makes stable and reliable periods and periods with frequent interference occupy the same width on the screen, making it impossible to discern the reliability of each data segment.

[0004] Vascular crisis risk assessment results are typically used alone, without signal correction or curve display, and the trend curve lacks intuitive indication of increased risk. These issues can affect the timely clinical judgment of early changes in the condition. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes an adaptive correction and curve reconstruction method and system for long-term monitoring of tissue oxygen saturation, in order to solve the aforementioned technical problems.

[0006] Firstly, an adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation is provided, including:

[0007] Acquire raw tissue oxygen saturation sampling data and real-time risk values ​​of the monitored subjects;

[0008] Determine the correction strength coefficient that is negatively correlated with the real-time risk value based on the real-time risk value;

[0009] Based on the correction intensity coefficient, adaptive signal correction is performed on the original tissue oxygen saturation sampling data, and candidate abnormal peaks are marked to obtain corrected monitoring data.

[0010] The confidence index of the statistical time window is obtained by calculating the confidence of the statistical time window based on the labeled candidate anomaly peaks.

[0011] The display area is divided into a historical display area and a real-time display area;

[0012] The corrected monitoring data is nonlinearly mapped in the historical display area based on the credibility index, and linearly mapped at equal time intervals in the real-time display area to obtain a trend curve, which is then output and displayed.

[0013] Furthermore, obtain the real-time risk value of the monitored object, including:

[0014] The data output by the risk assessment module is read and parsed according to a preset cycle to obtain the real-time risk value of the monitored object.

[0015] The real-time risk value and the correction strength coefficient are registered to obtain the risk value parameter and correction strength coefficient parameter stored in the shared memory.

[0016] Furthermore, based on the correction intensity coefficient, adaptive signal correction is performed on the original tissue oxygen saturation sampling data, and candidate abnormal peaks are labeled to obtain corrected monitoring data, including:

[0017] Read real-time risk values ​​and correction strength coefficients from shared memory;

[0018] Based on the preset basic physiological range, the effective data judgment threshold range is adjusted according to the real-time risk value. The original sampled data is marked with outliers and the positions of points exceeding the threshold are marked as missing values, resulting in threshold-processed data with missing value markings.

[0019] Based on the sliding window length that is positively correlated with the correction intensity coefficient, median filtering is performed on the effective sampling points of the threshold-processed data to obtain the local baseline of each sampling point.

[0020] Calculate the deviation between the effective sampling points and the local baseline, and filter and mark candidate abnormal peaks based on the peak judgment threshold that is positively correlated with the real-time risk value;

[0021] The candidate abnormal peaks are corrected to obtain corrected monitoring data.

[0022] Furthermore, the candidate abnormal peaks are corrected to obtain corrected monitoring data, including:

[0023] The candidate abnormal peaks are linearly interpolated using adjacent valid sampling points to obtain interpolation correction values.

[0024] The interpolated correction value and the original sampled value corresponding to the candidate abnormal peak are weighted and mixed based on the correction intensity coefficient to obtain the corrected output value. The larger the correction intensity coefficient, the higher the mixing weight of the interpolated correction value, and the smaller the correction intensity coefficient, the higher the mixing weight of the original sampled value.

[0025] When the real-time risk value is in the low-risk range, the candidate abnormal spike is marked as an interference spike;

[0026] When the real-time risk value is in the medium-risk range, and when the deviation value exceeds the peak determination threshold by a set multiple, the corresponding sampling point is invalidated to obtain missing value marked data, and the candidate abnormal peaks that are not invalidated are marked as interference peaks.

[0027] When the real-time risk value is in the high-risk range, the candidate abnormal peaks are subjected to characteristic wave morphology verification processing. Those that pass the verification are marked as confirmed crisis-type characteristic waves, and those that fail the verification are marked as peaks to be manually verified.

[0028] The correction results for each interval together constitute the corrected monitoring data.

[0029] Furthermore, when the real-time risk value is in the high-risk range, it also includes:

[0030] The original sampled data containing candidate abnormal peak periods are processed by cache priority configuration to obtain high-risk period data with the highest retention priority in the circular cache;

[0031] The data from the high-risk periods are stored at full sampling rate, while the monitoring data from other periods are stored at downsampling rate to obtain a hierarchically stored historical dataset.

[0032] The timestamps, risk values, and deviation amplitudes corresponding to the candidate abnormal peaks are packaged to obtain meta-data packets;

[0033] The metadata is stored in a non-volatile memory to obtain traceable feature wave annotation data.

[0034] Furthermore, the confidence level of the statistical time window is calculated based on the labeled candidate anomaly spikes, resulting in a confidence index for the statistical time window, including:

[0035] Using statistical time windows as statistical units, the total number of valid sampling points and the weighted number of candidate abnormal peaks within each time window are statistically processed. Among them, the interference peaks and the peaks to be manually reviewed are fully included in the weighted number, and the confirmed crisis characteristic waves are included in the weighted number according to the set discount weight.

[0036] The credibility index for the corresponding time window is calculated based on the statistical results. The credibility index is negatively correlated with the weighted proportion of candidate abnormal peaks.

[0037] Furthermore, the corrected monitoring data is non-linearly mapped in the historical display area based on the credibility index, and linearly mapped at equal time intervals in the real-time display area, including:

[0038] The historical time periods corresponding to the historical display area are divided to obtain multiple consecutive statistical time windows of equal duration.

[0039] The confidence index of each statistical time window is multiplied by the corresponding duration and then summed to obtain the total confidence score.

[0040] When the confidence index of all statistical time windows is zero, the corrected monitoring data in the historical display area is subjected to uniform mapping at equal time intervals to obtain the right boundary pixel position of each statistical time window.

[0041] When there is a statistical time window with a non-zero credibility index, the display width is allocated according to the proportion of the credibility score of each statistical time window to the total credibility score, and the right boundary pixel position of each statistical time window is obtained.

[0042] The calibrated monitoring data in the real-time display area is processed by linear mapping at equal time intervals in chronological order to obtain the pixel position of each sampling point in the real-time area.

[0043] Furthermore, when there is a statistical time window where the credibility index is not zero, the first [time window] in the historical display area... The right boundary pixel position of each statistical time window satisfy:

[0044]

[0045] in, The total pixel width of the historical display area. The sum of credibility integrals. To determine the duration of the statistical time window, For the first The credibility index of each statistical time window.

[0046] Furthermore, it also includes:

[0047] The width is calculated based on the right boundary pixel position of each statistical time window to obtain the initial display width of each statistical time window;

[0048] Using a set proportion of the total width of the historical display area as the low confidence window marking threshold, statistical time windows with an initial display width less than the low confidence window marking threshold are marked to obtain low confidence windows;

[0049] The low-confidence windows that are consecutive in time are segmented and merged to obtain multiple independent low-confidence segments; a minimum guaranteed width is allocated to each low-confidence segment; the sum of the minimum guaranteed widths of all low-confidence segments does not exceed the upper limit ratio of the total width of the historical display area; if it exceeds this ratio, the minimum guaranteed width of the smaller segment is adaptively adjusted to obtain the adjusted minimum guaranteed width of each segment.

[0050] The difference between the adjusted minimum width and the initial display width for all low-confidence segments is summed to obtain the total number of pixels added for the minimum.

[0051] The total number of newly added pixels for the minimum guarantee is distributed and deducted, and the corresponding pixels are deducted from the statistical time windows that have never been marked as low confidence windows according to the proportion of their respective initial display widths;

[0052] Iteratively perform width allocation and pixel subtraction processing to obtain the final display width of each statistical time window after convergence.

[0053] Furthermore, it also includes:

[0054] When the real-time risk value is in the high-risk range, a red warning triangle is superimposed on the time period containing the confirmed crisis-type characteristic wave, and a yellow warning dot is superimposed on the time period corresponding to the peak to be manually reviewed, so as to obtain a high-risk trend curve with graded warning labels.

[0055] In response to the triggering operation of the warning sign, the raw monitoring data of the full sampling rate for the corresponding time period is retrieved from the non-volatile memory, and the raw monitoring data of the full sampling rate is displayed at equal time intervals to obtain an expanded detailed waveform window.

[0056] The invention employing the above technical solution has the following advantages:

[0057] 1. This invention uses a real-time risk value negative correlation correction intensity coefficient to uniformly adjust the effective data judgment threshold, median filter window length, and peak judgment threshold. Candidate abnormal peaks are uniformly corrected with weighted mixed correction across the entire range and differential additional processing is applied according to the risk range. This allows for strong filtering in low-risk periods to suppress interference from body movement, light, etc., while actively weakening correction, widening thresholds, and shrinking windows in high-risk periods to retain true oxygen drop fluctuations. This improves upon the problem of fixed filtering that easily treats pathological fluctuations as noise and removes them. It reduces false positive alarms in low-risk periods while lowering the risk of missed alarms in high-risk periods.

[0058] 2. This invention quantifies the credibility index of each time period by weighting the proportion of candidate abnormal peaks within a statistical time window. The screen width in the historical display area is allocated according to the credibility score proportion, so that stable and reliable time periods are automatically widened to facilitate the observation of baselines and slow trends, while time periods with frequent interference are automatically compressed to reduce the amount of invalid information occupying the screen. The real-time display area maintains a linear mapping with equal time intervals to ensure the accuracy of the time scale of the current data. When the credibility is all zero, it degenerates into uniform compression, and a minimum width is set for low credibility segments to prevent them from being compressed to the point of being unrecognizable, thus taking into account both long-term trends and recent changes within the same screen.

[0059] 3. The real-time risk value of this invention is simultaneously involved in the adjustment of correction intensity, switching of peak correction strategy and curve color mapping. The peak marking result output by the correction stage is involved in the confidence calculation. The confidence index determines the screen width allocation of each segment of the historical curve. The connection between each stage and the coupling of parameters enable the risk information to optimize the processing strategy and presentation of monitoring data in reverse. Compared with the conventional approach of risk assessment outputting separately and signal processing remaining unchanged, the cooperation between each stage is closer. Attached Figure Description

[0060] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0061] Figure 1 This is a flowchart illustrating the adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to the present invention.

[0062] Figure 2 This is a schematic diagram of the adaptive signal correction and graded correction process in the adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation of the present invention.

[0063] Figure 3 This is a schematic diagram of the reliability calculation and partition mapping process in the adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation of the present invention.

[0064] Figure 4 This is a schematic diagram of the process of iterative allocation of minimum width constraint in the adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation in this invention.

[0065] Figure 5 This is a schematic diagram of the trend curve display effect in the adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation of the present invention.

[0066] Figure 6 This is a schematic diagram of the architecture of the adaptive correction and curve reconstruction system for long-term monitoring of tissue oxygen saturation according to the present invention. Detailed Implementation

[0067] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0068] like Figures 1-6 As shown, the adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation of the present invention includes:

[0069] Step S01: Obtain raw tissue oxygen saturation sampling data and real-time risk values ​​of the monitored objects;

[0070] In this embodiment, obtaining the real-time risk value of the monitored object includes:

[0071] The data output by the risk assessment module is read and parsed according to a preset cycle to obtain the real-time risk value of the monitored object.

[0072] The real-time risk value and correction strength coefficient are registered to obtain the risk value parameter and correction strength coefficient parameter stored in the shared memory.

[0073] Specifically, the monitor collects raw tissue oxygen saturation sampling data at a sampling frequency of 1 Hz. Simultaneously, it connects to the dynamic risk assessment module for vascular crisis via Ethernet or Wi-Fi communication to obtain the dynamic risk value of the currently monitored subject according to a preset update cycle (every 5 minutes by default, with support for immediate updates triggered by abnormal vital sign data). , The value range is 0–100%. In practice, this risk assessment module can be a dynamic prediction model deployed on a server, such as a nomogram model for predicting vascular crisis risk based on clinical indicators like hemoglobin (Hb), albumin (ALB), packed red blood cells (PBRCs), and smoking status. The monitor receives the JSON-formatted risk value data packet output by this model, which is then parsed by the main processor to obtain… .

[0074] The main processor will The data is written to a shared memory register for subsequent signal calibration and curve display. Between updates, the calibration algorithm uses the most recent valid risk value in the shared memory; if no valid risk value is received for more than three consecutive update cycles, the system retains the current value. The status remains unchanged, and a risk assessment offline message is displayed on the monitor interface. The correction strategy corresponding to the risk value continues to be used without automatic downgrading, ensuring that signal processing is uninterrupted and the correction intensity is not changed during network outages.

[0075] Step S02: Determine the correction strength coefficient that is negatively correlated with the real-time risk value based on the real-time risk value.

[0076] Specifically, to achieve continuous and smooth adjustment of the correction intensity with risk, based on real-time risk values... Define the signal correction intensity coefficient λ(t) and calculate it using the following formula:

[0077]

[0078] The value of λ(t) ranges from 0 to 1; the closer the value is to 1, the lower the current risk and the higher the signal correction strength; the closer the value is to 0, the higher the current risk and the lower the signal correction strength. This coefficient is involved in the updating of hardware register parameters for median filter window length, peak determination threshold, and corrective mixed weights throughout the process.

[0079] Step S03: Perform adaptive signal correction on the original tissue oxygen saturation sampling data based on the correction intensity coefficient and mark candidate abnormal peaks to obtain the corrected monitoring data;

[0080] In this embodiment, adaptive signal correction is performed on the raw tissue oxygen saturation sampling data based on the correction intensity coefficient, and candidate abnormal peaks are marked to obtain corrected monitoring data, including:

[0081] Read real-time risk values ​​and correction strength coefficients from shared memory;

[0082] Based on the preset basic physiological range, the effective data determination threshold range is adjusted according to the real-time risk value. The original sampled data is marked with outliers and the positions of points exceeding the threshold are marked as missing values, resulting in threshold-processed data with missing value markings.

[0083] Based on the sliding window length that is positively correlated with the correction intensity coefficient, median filtering is performed on the effective sampling points of the threshold-processed data to obtain the local baseline of each sampling point.

[0084] Calculate the deviation between the effective sampling points and the local baseline, and filter and mark candidate abnormal peaks based on the peak judgment threshold that is positively correlated with the real-time risk value;

[0085] The candidate abnormal peaks are corrected to obtain the corrected monitoring data.

[0086] Specifically, adaptive hard threshold anomaly removal: preset basic physiological threshold range for tissue oxygen saturation [ , (Default lower limit 40%, upper limit 100%), and set the maximum relaxation range for the upper and lower limits. , (The default lower limit can be relaxed by up to 10%, and the upper limit can be relaxed by up to 5%).

[0087] The threshold for determining valid data is dynamically adjusted based on real-time risk values.

[0088]

[0089]

[0090] The main processor filters the raw sampled data point by point: sampled values ​​below or higher The points were determined to be invalid data and were uniformly set as missing values.

[0091] In low-risk conditions, the threshold range is narrow, which can quickly eliminate extreme outliers introduced by factors such as sensor hardware malfunctions, probe irradiation by heat lamps, and data acquisition or transmission errors, thus completing the initial filtering of signal validity. In high-risk conditions, the threshold range is automatically widened to prevent true deep blood oxygenation from being misjudged as exceeding the range and discarded.

[0092] The upper limit is relaxed to a maximum of 105% to take into account the extremely small readings that may occur within the calibration tolerance range of the near-infrared spectral sensor; even if readings above 100% are retained within the relaxed threshold range, the system will still mark them as out-of-range suspicious and will not participate in subsequent trend statistics or in the median calculation of the median filter window.

[0093] Correspondingly, readings below 40% of the baseline lower limit but retained within the relaxed threshold range are marked with a suspected deep hypoxia marker. This type of data participates in median filtering and trend statistics, but is associated with confirmed crisis-type characteristic waves during high-risk periods to help clinicians determine whether it is true deep hypoxia or sensor abnormality.

[0094] Points marked as missing values ​​are not included in the median calculation of the subsequent median filtering window; when plotting in low-risk and medium-risk states, adjacent valid points are linearly connected to fill the gaps, while when plotting in high-risk states, gaps are left to accurately reflect missing data.

[0095] Adaptive candidate anomaly spike identification: The local baseline of the signal is extracted by a sliding window mid-range filter. The window length and spike judgment threshold are dynamically adjusted with the correction intensity coefficient λ(t), and the corresponding parameters are written to the control register of the FPGA coprocessor in real time.

[0096] The FPGA coprocessor is used to perform window caching and median calculation. This is because, under 1Hz sampling, all data in the window needs to be sorted and the median value needs to be calculated within each sampling period (1 second). Meanwhile, the main processor runs the network protocol stack and human-machine interface tasks at the same time. Operating system scheduling jitter may cause uncertain filtering delay. The FPGA executes the hardware logic to ensure that the baseline is updated within a certain time delay for each sampling point, which meets the real-time monitoring requirements.

[0097] The sliding window length N(t) is:

[0098]

[0099] in, The base window length (default 31 consecutive sampling points, an odd number is used to ensure the window has a clear center point). This is the minimum window length (default 5 sampling points). {} indicates taking the nearest odd number; if the rounding result is even, the larger odd number is taken. (For low-risk situations) A value close to 1 indicates a long window, high baseline smoothness, and strong filtering capability; at high risk... Approaching 0, the window shrinks to its minimum, and the baseline closely follows the original signal, avoiding smoothing out the true rapid downward trend.

[0100] For each sampling point, the FPGA coprocessor calculates the median of the valid sampled values ​​within its window as the local baseline. Calculate the original sampled values ​​point by point. Absolute deviation from local baseline Mt =| - |

[0101] The threshold for identifying spikes increases as the risk level rises.

[0102]

[0103] in, The baseline deviation threshold is set to 5% by default. The threshold for determining the peak value. This represents the absolute deviation between the sampling point and the local baseline. When > At that time, the sampling point was marked as a candidate anomalous spike.

[0104] The higher the risk, the larger the judgment threshold, thus reducing the probability of real pathological fluctuations being misjudged as spikes at the identification level.

[0105] Specifically, although the threshold for high-risk situations is relaxed, the signal fluctuations caused by vascular crises are usually much greater than the relaxed threshold. Early crises often manifest as an overall downward trend with large fluctuations, and some rapid decline segments can deviate from the local baseline by more than 10%, which will still be identified and marked. In contrast, the threshold is stricter in low-risk situations, and occasional spikes caused by body movement can also be effectively detected.

[0106] Differentiated correction strategy based on risk level: The basic correction of candidate abnormal peaks adopts a weighted mixture across the entire range. Differentiated additional processing is superimposed on each risk range to achieve negative feedback coupling where the higher the risk, the weaker the correction. Candidate abnormal peaks are classified into three categories: interference peaks, confirmed crisis characteristic waves, and peaks awaiting manual review.

[0107] In this embodiment, candidate abnormal peaks are corrected to obtain corrected monitoring data, including:

[0108] Candidate abnormal peaks are linearly interpolated using adjacent valid sampling points to obtain interpolated correction values;

[0109] The interpolated correction value and the original sampled value corresponding to the candidate abnormal peak are weighted and mixed based on the correction intensity coefficient to obtain the corrected output value. The larger the correction intensity coefficient, the higher the mixing weight of the interpolated correction value, and the smaller the correction intensity coefficient, the higher the mixing weight of the original sampled value.

[0110] When the real-time risk value is in the low-risk range, the candidate abnormal spikes are marked as interference spikes;

[0111] When the real-time risk value is in the medium risk range, and the deviation value exceeds the peak judgment threshold by a set multiple, the corresponding sampling point is invalidated to obtain missing value marked data. Candidate abnormal peaks that are not invalidated are marked as interference peaks.

[0112] When the real-time risk value is in the high-risk range, the candidate abnormal peaks are subjected to characteristic wave morphology verification. Those that pass the verification are marked as confirmed crisis-type characteristic waves, and those that fail the verification are marked as peaks to be manually verified.

[0113] The correction results for each interval together constitute the corrected monitoring data.

[0114] Specifically, the differentiated correction strategy based on risk level: the basic correction of candidate abnormal peaks uniformly adopts a weighted hybrid strategy across the entire range:

[0115]

[0116] in, The correction value is obtained by linear interpolation of adjacent valid sampling points. These are the original sampled values ​​corresponding to the candidate abnormal spikes. This is the corrected output value after weighted mixing. When the value approaches 1 (low risk), the mixed result degenerates into a complete interpolation replacement; when When the value approaches 0 (high risk), the mixed result degenerates into the original value. The correction strength varies across the entire range. Continuous changes, no gear shifts.

[0117] Based on this, differentiated additional processing is applied to each risk range:

[0118] when <50% (low-risk range): If the value is >0.5, the interpolation weight is higher than 50%, and candidate abnormal spikes are mainly corrected by interpolation. At this stage, the patient's condition is stable, and the spikes are most likely caused by external factors such as body movement or poor probe contact. These are defined as interference spikes, and strong correction can maximize the suppression of interference, which is consistent with the logic of conventional correction.

[0119] When 50%≤ ≤75% (medium risk range): 0.25≤ ≤0.5, the interpolation weight transitions smoothly between 25% and 50%, achieving a continuous connection between correction and retention. Only when the deviation exceeds 2 times ΔS_th is it judged as extreme interference and directly set as a missing value.

[0120] when >75% (high-risk range): <0.25, interpolation weight is less than 25%, weighted mixing approximates the original value preservation, and candidate abnormal peaks likely correspond to drastic fluctuations in blood oxygenation under pathological conditions. Characteristic wave morphology verification is also performed in this interval.

[0121] Candidate abnormal spikes retained in high-risk intervals need to be graded and processed after characteristic wave morphology verification in order to distinguish between pathological fluctuations and interference such as probe loosening.

[0122] For each retained candidate abnormal peak, its temporal morphological feature parameters are extracted, including: the slope of the peak abrupt change (the larger absolute value of the rising or falling edge, representing the change in tissue oxygen saturation per unit time), the duration of the peak (the time interval from deviation from the local baseline to the return to the baseline or the entry into the next peak), and the proportion of the same direction of change of consecutive sampling points before and after the peak.

[0123] The same-direction change ratio is defined as follows: taking the abrupt change of the candidate abnormal peak as the reference, take 3 consecutive sampling points before and after it (a total of 6 sampling points), and count the number of points whose sampling value change direction is consistent with the direction of the abrupt change. The ratio of this number to 6 is the same-direction change ratio.

[0124] A candidate abnormal spike is deemed to have passed morphological verification when all three of the following conditions are met: the slope of the mutation does not exceed the human physiological limit threshold (default 5 percentage points / second, i.e., the change does not exceed 5 percentage points per second); the proportion of changes in the same direction before and after the spike is ≥2 / 3; and the duration of the spike is not less than 3 seconds. A spike that passes verification is marked as a confirmed crisis-type characteristic wave, triggering a red warning triangle; a spike that fails verification is marked as a spike awaiting manual verification, retaining the original sampled value but not triggering an automatic red warning, and is only marked with a yellow warning dot for manual judgment by medical staff.

[0125] Interference such as probe loosening usually manifests as instantaneous jumps (the slope of the sudden change is far beyond the physiological limit, the single-point jump is immediately recovered, and the duration is less than 1 second), which can be effectively intercepted by morphological verification; the early tissue oxygenation in vascular crisis manifests as a gradual and continuous homologous change, with the slope within the physiological limit, which can be verified.

[0126] The empirical parameters in this embodiment, including a baseline deviation threshold of 5%, a peak determination threshold amplification factor of 0.5, a maximum relaxation range of 10% for the lower limit of the hard threshold and an upper limit of 5%, a mutation slope threshold of 5 percentage points / second for morphological verification, a unidirectional change ratio threshold of 2 / 3, and a duration threshold of 3 seconds, are derived from statistical analysis of clinical tissue oxygenation monitoring data at a sampling rate of 1 Hz. Specifically, after labeling the signals during the body motion interference period and the early period of vascular crisis, the distribution of characteristics such as fluctuation amplitude, mutation slope, and duration of the two types of periods is statistically analyzed. The separation point of the two distributions is taken as the default value of each threshold to maximize the distinction between interference and pathological fluctuations. Those skilled in the art can make adaptive adjustments according to the actual sampling rate and clinical scenario.

[0127] In this embodiment, when the real-time risk value is in the high-risk range, the method further includes:

[0128] The original sampled data containing candidate abnormal peak periods are processed by cache priority configuration to obtain high-risk period data with the highest retention priority in the circular cache;

[0129] Data from high-risk periods are stored at full sampling rate, while monitoring data from other periods are downsampled and stored to obtain a hierarchically stored historical dataset.

[0130] The timestamps, risk values, and deviation ranges corresponding to candidate anomaly peaks are packaged to obtain metadata packets;

[0131] The metadata is stored in a non-volatile memory to obtain traceable feature wave annotation data.

[0132] Specifically, all candidate abnormal spikes (including confirmed crisis-type characteristic waves and spikes awaiting manual review) within this time period undergo triple processing in the data storage and caching stage: First, the original sampled data for this time period is given the highest retention priority in the circular cache and will not be overwritten by new data; second, for long-term storage, the original data of the time period at a full sampling rate of 1Hz is retained, while the remaining regular historical data is downsampled at 0.5Hz after storage for more than 24 hours, and the storage density of high-risk periods is twice that of regular historical data, which is equivalent to double the storage weight; third, each candidate abnormal spike (including confirmed crisis-type characteristic waves and spikes awaiting manual review) is accompanied by metadata such as timestamp, risk value, and deviation amplitude, and stored in non-volatile memory to support subsequent backtracking analysis.

[0133] Through the above preprocessing steps, a real-time tissue oxygen fluctuation curve with complete time sequence, continuous trend, and fidelity matching the risk level can be obtained.

[0134] Step S04: Calculate the confidence level of the statistical time window based on the labeled candidate anomaly spikes to obtain the confidence index of the statistical time window;

[0135] In this embodiment, the confidence index of the statistical time window is obtained by calculating the confidence level of the statistical time window based on the labeled candidate anomaly spikes, including:

[0136] Using statistical time windows as statistical units, the total number of valid sampling points and the weighted number of candidate abnormal peaks within each time window are statistically processed. Among them, interference peaks and peaks awaiting manual review are fully included in the weighted number, and confirmed crisis characteristic waves are included in the weighted number according to the set discount weight.

[0137] The credibility index for the corresponding time window is calculated based on the statistical results. The credibility index is negatively correlated with the weighted proportion of candidate abnormal peaks.

[0138] Specifically, using a statistical time window (default 10 minutes, which can be adaptively adjusted according to the total monitoring duration) as the statistical unit, the main processor counts the number of valid sampling points within that time window. (Excluding missing values ​​removed by hard thresholding), and the number of candidate abnormal spikes marked by the aforementioned adaptive signal correction step. .

[0139] First, the total number of abnormal points identified as deviating from the local baseline is counted. Then, they are weighted according to the following peak types. Regardless of whether the peak is ultimately corrected or retained, its existence itself indicates that the signal fluctuations during that period are more severe, more likely to be interfered with, or more likely to have pathological mutations. Therefore, its reliability as a basis for clinical baseline judgment is lower.

[0140] Define the credibility index for this time window:

[0141]

[0142] when When the value is 0, the reliability index of this statistical time window is... Recorded as 0, The value ranges from 0 to 1. The closer the value is to 1, the more stable the signal is, the more stable the baseline is, and the higher the clinical reliability of the data. The closer the value is to 0, the more drastic the signal fluctuation is, the more interference or mutations there are, and the data needs to be interpreted with caution.

[0143] Candidate abnormal spikes are classified into two categories based on their final correction results: interference spikes, including spikes that are replaced by interpolation in low-risk intervals, spikes that are invalidated in medium-risk intervals, candidate abnormal spikes that are not invalidated in medium-risk intervals, and spikes that are awaiting manual review in high-risk intervals that have not passed morphological review; and pathological spikes, namely confirmed crisis-type characteristic waves that have passed morphological review in high-risk intervals.

[0144] When calculating the credibility index, all interfering spikes are included in the outlier statistics. Pathological spikes are weighted at 0.3, meaning each pathological spike is equivalent to 0.3 abnormal points. This weighting design reflects the following considerations: although pathological fluctuations cause signal instability and reduce baseline reference value, the data itself has clinical significance and should not be compressed in the same way as interference from body movement, probe loosening, etc. After weighting adjustment, the confidence index of high-risk periods will not be excessively lowered due to dense pathological fluctuations, and the compression of this period in the historical display area will be correspondingly reduced. Combined with the red triple marker and click-to-expand function, this ensures that clinicians pay attention to high-value periods.

[0145] Step S05: Divide the display area to obtain the historical display area and the real-time display area.

[0146] Specifically, let the total pixel width of the screen's horizontal axis be W, and divide it into two fixed parts: the left side is the historical display area, with a width denoted as W. (By default, it occupies 60% of the total width); the right side is the real-time display area, with a width of [missing information]. (By default, it occupies 40% of the total width), satisfying... .

[0147] The two zones employ independent coordinate mapping rules, ensuring continuous coordinates and smooth curve transitions at the boundary point, thus guaranteeing visual continuity. The width ratio of the historical to the real-time zone can be configured by the user according to clinical preferences. This applies when the total monitoring duration T does not exceed the real-time window length. At that time, there is no historical data, so the historical display area does not display data. The real-time display area expands to the entire width of the screen, and all data is displayed in the real-time area at equal time intervals. When T exceeds... Afterwards, data earlier than the real-time window enters the historical area and is mapped according to the following rules.

[0148] Real-time window length display Supports switching between 2 / 4 / 6 / 8 / 12 hours.

[0149] Step S06: Perform nonlinear mapping on the corrected monitoring data in the historical display area based on the credibility index, and perform equal-time linear mapping on the corrected monitoring data in the real-time display area to obtain the trend curve and output it for display.

[0150] In this embodiment, the corrected monitoring data is non-linearly mapped in the historical display area based on the confidence index, and linearly mapped at equal time intervals in the real-time display area, including:

[0151] The historical time periods corresponding to the historical display area are divided to obtain multiple consecutive statistical time windows of equal duration.

[0152] The confidence index of each statistical time window is multiplied by the corresponding duration and then summed to obtain the total confidence score.

[0153] When the confidence index of all statistical time windows is zero, the corrected monitoring data in the historical display area is uniformly mapped at equal time intervals to obtain the right boundary pixel position of each statistical time window.

[0154] When there is a statistical time window with a non-zero credibility index, the display width is allocated according to the proportion of the credibility score of each statistical time window to the total credibility score, and the right boundary pixel position of each statistical time window is obtained.

[0155] The calibrated monitoring data in the real-time display area is processed by linear mapping at equal time intervals in chronological order to obtain the pixel position of each sampling point in the real-time area.

[0156] Specifically, historical periods are divided into A series of consecutive statistical time windows, each with a duration of [duration missing]. .

[0157] First, calculate the sum of the credibility integrals over the entire historical period:

[0158]

[0159] To avoid the denominator being zero, when the total time window... When all values ​​are 0 (in extreme interference situations), the system automatically degenerates into uniform compression with equal time intervals, that is, all... Participate in the calculation.

[0160] The real-time display area corresponds to the latest monitoring period and is used by bedside medical staff to observe changes in current tissue oxygen saturation in real time. Therefore, a strict equidistant linear mapping is adopted to ensure the accuracy of the time scale and consistency with clinical reading habits. Perform linear coordinate mapping in the latest time period:

[0161]

[0162] in, The pixel width of the history display area The pixel width of the real-time display area. To display the window length in real time, To monitor the total duration.

[0163] Within this area, the timeline is strictly equidistant at a 1:1 scale and is unaffected by credibility-weighted compression. All time scales and numerical annotations are drawn according to standard time intervals. Credibility background color bands, risk curve colors, alarm markers, and other elements are displayed normally overlaid within this area.

[0164] In this embodiment, when there is a statistical time window with a confidence index that is not zero, the first [time window name] in the historical display area... The right boundary pixel position of each statistical time window satisfy:

[0165]

[0166] in, The total pixel width of the history display area. The sum of credibility integrals. To determine the duration of the statistical time window, For the first The credibility index of each statistical time window.

[0167] Specifically, the general mapping formula for the continuous time domain is:

[0168]

[0169] Among them, C It is a continuous function of the credibility index. This is the time variable for integration.

[0170] The physical meaning of this mapping is that each pixel in the history region corresponds to a fixed confidence integral, rather than a fixed duration.

[0171] During stable periods with high credibility, the credibility score per unit of time is higher, so they are automatically stretched on the screen to occupy more display space, making it easier for doctors to observe baseline levels and slow trends. During fluctuating / interference periods with low credibility, the credibility score per unit of time is lower, so they are automatically compressed on the screen to reduce visual interference from invalid information.

[0172] When the credibility index of each historical period is approximately equal, the mapping degenerates into uniform linear compression, that is, after monitoring the corresponding duration, it is compressed by a fixed ratio and the most recent display span is displayed at equal intervals. This method is fully compatible with this display method.

[0173] In this embodiment, it also includes:

[0174] The width is calculated based on the right boundary pixel position of each statistical time window to obtain the initial display width of each statistical time window;

[0175] Using a set proportion of the total width of the historical display area as the low confidence window marking threshold, statistical time windows with an initial display width less than the low confidence window marking threshold are marked to obtain low confidence windows;

[0176] Segment merging is performed on time-continuous low-confidence windows to obtain multiple independent low-confidence segments; a minimum guaranteed width is allocated to each low-confidence segment; the sum of the minimum guaranteed widths of all low-confidence segments does not exceed the upper limit ratio of the total width of the historical display area, and if it does, the minimum guaranteed width of the smaller segment is adaptively adjusted to obtain the adjusted minimum guaranteed width of each segment.

[0177] The difference between the adjusted minimum width and the initial display width for all low-confidence segments is summed to obtain the total number of pixels added for the minimum.

[0178] The total number of newly added pixels is distributed and deducted, and the corresponding pixels are deducted from the statistical time windows that have never been marked as low confidence windows according to the proportion of their respective initial display widths.

[0179] Iteratively perform width allocation and pixel subtraction processing to obtain the final display width of each statistical time window after convergence.

[0180] Specifically, to prevent high-risk periods from being excessively compressed to the point of being undetectable to the naked eye due to dense candidate anomaly peaks and low reliability, thus causing high-risk alerts to fail, a minimum guaranteed width constraint is set. This takes into account the number of statistical time windows in long-term monitoring. The probability is high (e.g., monitoring for a week, or one window every 10 minutes). (Over a thousand), if a separate safety net is provided for each low-confidence window, the total safety net width may exceed the total width of the historical region, making convergence impossible. Therefore, this method adopts a strategy of merging consecutive low-confidence segments for safety net coverage:

[0181] Calculate the initial display width of all time windows using the confidence integral formula described above;

[0182] The initial width of the marker is less than the low confidence window marker threshold. ( default (Time window)

[0183] Temporally consecutive marked windows are merged into a single low-confidence segment, and each independent low-confidence segment is allocated a minimum segment width. ( default ), with each window within the interval having an equal width;

[0184] The sum of the minimum widths of all low-reliability segments must not exceed the total width of the historical segment. That is, let the total number of low-confidence segments be n, when hour, Adaptive adjustment This ensures algorithm convergence;

[0185] The total increase in pixels due to the minimum guarantee is deducted from all unmarked high-confidence windows by an equal proportion of their initial widths;

[0186] If, after deduction, the width of a high-confidence window is less than... If the value is low, it is merged into the adjacent low-confidence segment, and the above allocation is repeated until convergence.

[0187] The convergence of the above iterative process is guaranteed by two points: First, the total number of statistical time windows marked as low confidence in each iteration is monotonically non-decreasing. The merging operation only assigns high confidence windows to low confidence segments. There is no reverse operation that re-marks low confidence windows as high confidence. The total number of low confidence windows is bounded by the total number of statistical time windows N in the historical period. Therefore, the number of iterations does not exceed N and will terminate within a finite number of steps.

[0188] Furthermore, the sum of the guaranteed widths of all low-confidence segments is constrained by an upper limit of 50% of the total width of the historical region. If this limit is exceeded, the width is adaptively reduced, ensuring that the total number of newly added pixels never exceeds this upper limit. After deducting from the high-confidence window, the total pixel width of the historical region remains constant at W. h .

[0189] These two points together ensure that the iterative allocation process is free from dead loops and overflows.

[0190] This constraint ensures that, regardless of the monitoring duration or the number of low-confidence windows, each low-confidence segment retains at least one visually identifiable and clickable warning strip on the screen, and the total pixel width of the historical area always equals [the specified value]. No overflow, no offset.

[0191] Furthermore, when the proportion of low-confidence segments is too high within the monitoring period, and the total width of the high-confidence window after iterative convergence is less than 50% of the total width of the historical display area, the confidence-weighted differentiated display effect weakens. The system automatically degrades to a uniform mapping with equal time intervals, and marks the entire time period with large signal fluctuations above the historical area, suggesting that each segment be expanded for viewing. This degradation only occurs when the patient is continuously at high risk with frequent pathological fluctuations or prolonged body movement interference. It is a reasonable degradation strategy in boundary scenarios and does not affect the differentiated display effect in normal monitoring scenarios.

[0192] In this embodiment, it also includes:

[0193] The horizontal axis region below the trend curve is colored according to the credibility index to obtain the basic trend curve with credibility background color band.

[0194] The basic trend curve is color-mapped according to the risk level range of the real-time risk value to obtain the trend curve with the corresponding risk level color.

[0195] When the real-time risk value is in the high-risk range, a red warning triangle is superimposed on the time period containing the confirmed crisis-type characteristic wave, and a yellow warning dot is superimposed on the time period corresponding to the peak to be manually reviewed, so as to obtain a high-risk trend curve with graded warning labels.

[0196] In response to the triggering operation of the warning sign, the raw monitoring data of the full sampling rate for the corresponding time period is retrieved from the non-volatile memory, and the raw monitoring data of the full sampling rate is displayed at equal time intervals to obtain an expanded detailed waveform window.

[0197] Specifically, in the horizontal axis area below the curve, a background color band corresponding to the statistical time window is simultaneously plotted. This coloring rule is applied uniformly to both the real-time and historical areas to visually represent the data signal quality. The correspondence rule between the background color band and the confidence index is as follows:

[0198] Green background indicates high-reliability data with stable signals and reliable baselines, which can serve as a stable basis for clinical decision-making.

[0199] Yellow background indicates data of medium confidence, which may fluctuate slightly and should be interpreted in the context of clinical practice.

[0200] Red background indicates low-reliability data with drastic signal fluctuations, which may contain a lot of interference or pathological mutations. Interpretation should be done with caution.

[0201] The curve color represents the patient's pathological risk level, and the background color represents the data signal quality; the two are displayed superimposed on the screen.

[0202] The system reconstructs the color of the monitoring curve based on the real-time risk value: when When the oxygen saturation level is >75%, the tissue oxygen monitoring curve turns continuously red from that time point, indicating a high risk. As the dynamic risk assessment is updated, when... When the level drops below 75%, the monitoring curve returns to its normal color.

[0203] The color status changes based on the timestamp returned by the risk assessment module, and the last color status is retained during network interruptions. When a patient is at high risk and the data has low reliability, a red curve and a red background form a double red alarm, which can enhance the high-risk warning effect.

[0204] High-risk periods marked as containing confirmed crisis-type characteristic waves are indicated by a red warning triangle above the historical curve. Periods containing spikes requiring manual review but without confirmed crisis-type characteristic waves are indicated by a yellow warning dot, aligned with the center of the corresponding low-confidence segment. Healthcare professionals can trigger interaction via touch clicks, mouse hover, or physical buttons. Upon triggering, the monitor's display controller retrieves the full sampling rate raw data for that period from non-volatile memory and displays a separate details window on the screen. This window displays the raw monitoring curve for that period at standard equidistant time intervals and simultaneously shows metadata such as the risk value change curve, the time of occurrence of candidate abnormal spikes, deviation amplitude, and hard threshold status. Because a double-weighted duplicate storage strategy is applied to high-risk periods, the full sampling rate raw data for those periods is fully preserved, resulting in no loss of accuracy in the expanded detailed curves, allowing healthcare professionals to perform detailed waveform analysis.

[0205] The yellow warning dot also supports touch clicks, mouse hovering, or physical button triggering. When expanded, it displays the original waveform of the full sampling rate and morphological verification parameters (mutation slope, duration, and proportion of change in the same direction) for that period. Clinicians can then manually determine whether it is a pathological fluctuation by combining it with the patient's vital signs data.

[0206] The interactive expansion function is only executed when actively triggered by the user and does not affect the normal display of the long-term trend chart, balancing display efficiency and detail accuracy. The red high-risk area also follows the compression rules in the curve reconstruction and is compressed and displayed along with the historical data. However, because it has a triple visual identifier of red curve color, red low confidence background and red warning triangle, it will not be ignored by doctors even when compressed to a narrow width.

[0207] In other embodiments, an adaptive correction and curve reconstruction system for long-term tissue oxygen saturation monitoring is provided, based on any of the preceding adaptive correction and curve reconstruction methods for long-term tissue oxygen saturation monitoring, including:

[0208] The real-time risk value module is configured to acquire raw tissue oxygen saturation sampling data and real-time risk values ​​of the monitored objects;

[0209] The correction strength coefficient module is configured to determine a correction strength coefficient that is negatively correlated with the real-time risk value based on the real-time risk value.

[0210] The monitoring data module is configured to perform adaptive signal correction on the raw tissue oxygen saturation sampling data based on the correction intensity coefficient and mark candidate abnormal peaks to obtain the corrected monitoring data.

[0211] The credibility index module is configured to calculate the credibility of a statistical time window based on labeled candidate anomaly spikes, and obtain the credibility index of the statistical time window.

[0212] The display area module is configured to divide the display area into a historical display area and a real-time display area;

[0213] The trend curve module is configured to perform non-linear mapping on the corrected monitoring data in the historical display area based on the confidence index, and perform linear mapping on the corrected monitoring data in the real-time display area at equal time intervals to obtain the trend curve and output it for display. Specific Implementation Example 1:

[0215] The monitor acquires raw tissue oxygen saturation data at a frequency of 1 Hz and retrieves risk values ​​from a vascular crisis risk assessment tool via the hospital network every 5 minutes. .

[0216] Median Filtering Base Window =31, minimum window =5, Basic Deviation Threshold =5%, credibility statistics window t=10 minutes, real-time display window =2 hours; Total width of screen horizontal axis =1000 pixels, including the history area =600 pixels, real-time area =400 pixels; Low confidence window marking coefficient α=3% (18 pixels), segment guarantee coefficient β=5% (30 pixels).

[0217] From 0 to 24 hours post-surgery, the patient's condition remained stable, with R(t) maintained between 20% and 35% (low-risk range), λ(t) approximately 0.65 to 0.8, and an odd number of points (approximately 21 to 25) in the median filtering window. The peak detection threshold was approximately 5.5% to 5.88%. Occasional patient movements generated short-term interfering peaks, which were identified and corrected using a weighted hybrid method (interpolation weight approximately 65% ​​to 80%), resulting in a smooth curve.

[0218] There were very few candidate abnormal spikes within each 10-minute window. Approximately 0.92~0.98, the background is displayed in green; in the historical area, this stable segment occupies a wider display space due to its high reliability, making it easy to observe the baseline trend.

[0219] At 30 hours post-surgery, the risk assessment module returned R(t) = 80% (high-risk range), λ(t) = 0.2, the median filter window shrank to 7 points (close to the minimum window length), the peak determination threshold was increased to 7% (1.4 times the baseline value), and the lower limit of the hard threshold was widened to 32% and the upper limit to 104%. The patient exhibited early signs of vascular crisis, with tissue oxygen signal oscillating downward from 65% to the 45% range within approximately 5 minutes. During this downward movement, continuous fluctuations occurred, with the rate of decline accelerating at certain times, reaching a minimum of 32%–35%, and the amplitude of signal fluctuations significantly increased.

[0220] Because the window was shrunk to 7 points, the threshold was magnified to 7%, and interpolation was disabled, the entire downward trend and rapid decline were fully preserved. Among them, the rapid decline that deviated from the local baseline by more than 7% was morphologically verified (the abrupt change was along a slope of about 2 percentage points / second, with 5 consecutive points changing in the same direction and lasting for about 8 seconds, all of which met the verification conditions) and was determined to be a confirmed crisis-type characteristic wave. The data in this period was not overwritten in the circular cache and was stored for a long time with a full sampling rate of 1Hz (the conventional historical data was downsampled to 0.5Hz). From this point in time, the curve turns continuously red. Within this 10-minute window, approximately 300 points are marked as candidate abnormal peaks. Of these, about 60 are confirmed as crisis-type characteristic waves (pathological type) after morphological review, while the remaining approximately 240 are peaks awaiting manual review (interference type). According to weighted statistics, the weighted number of abnormal points = 240 × 1 + 60 × 0.3 = 258, and the confidence index ≈ 1 − 258 / 600 ≈ 0.57. The background is displayed in yellow (medium confidence). Although there are many pathological fluctuations during this period, the confidence is not excessively lowered because the weight of pathological peaks is relatively low. The red warning is reflected by the continuous red curve and the red warning triangle. During this period, the initial mapping width in the historical area is lower than the low confidence window marking threshold, triggering the minimum constraint. It is forcibly raised and merged with the adjacent low confidence window into a low confidence segment, with an overall minimum width of 30 pixels. A red warning triangle is superimposed on the curve, making it visually clear.

[0221] The doctor observed a red warning band at the 30th hour on the long-term trend chart. Clicking to expand it, a pop-up window displayed the original 1Hz waveform of that period at 1-second intervals. It was evident that tissue oxygen was declining in oscillations with multiple rapid drops. Combined with the red high-risk indicator, it was determined to be an early manifestation of vascular crisis, and timely treatment was initiated.

[0222] From 40 hours post-surgery, the risk value decreased to 60% (intermediate risk range), λ(t) = 0.4, the median filtering window was approximately 13 points, and the system employed a weighted mixed correction. The curve gradually smoothed out as the risk decreased. 48 hours post-surgery, the risk value dropped below 30%, and the low-risk strong correction strategy was reverted. When the value rises above 0.9, the background returns to green, the curve returns to its normal color, and it occupies a wider space on the screen, making it easier to observe the trend of blood supply recovery in the flap.

[0223] The latest 2-hour data is always displayed in the real-time area on the right with equal time intervals, unaffected by compression, allowing bedside medical staff to observe current changes in real time.

[0224] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. An adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation, characterized in that, include: Acquire raw tissue oxygen saturation sampling data and real-time risk values ​​of the monitored subjects; Determine the correction strength coefficient that is negatively correlated with the real-time risk value based on the real-time risk value; Based on the correction intensity coefficient, adaptive signal correction is performed on the original tissue oxygen saturation sampling data, and candidate abnormal peaks are marked to obtain corrected monitoring data. The confidence index of the statistical time window is obtained by calculating the confidence of the candidate anomaly peaks based on the labels. The display area is divided into a historical display area and a real-time display area; The corrected monitoring data is nonlinearly mapped in the historical display area based on the credibility index, and linearly mapped at equal time intervals in the real-time display area to obtain a trend curve, which is then output and displayed.

2. The adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to claim 1, characterized in that, Obtain the real-time risk value of the monitored object, including: The data output by the risk assessment module is read and parsed according to a preset cycle to obtain the real-time risk value of the monitored object. The real-time risk value and the correction strength coefficient are registered to obtain the risk value parameter and correction strength coefficient parameter stored in the shared memory.

3. The adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to claim 2, characterized in that, Based on the correction intensity coefficient, adaptive signal correction is performed on the original tissue oxygen saturation sampling data, and candidate abnormal peaks are marked to obtain corrected monitoring data, including: Read real-time risk values ​​and correction strength coefficients from shared memory; Based on the preset basic physiological range, the effective data judgment threshold range is adjusted according to the real-time risk value. The original sampled data is marked with outliers and the positions of points exceeding the threshold are marked as missing values, resulting in threshold-processed data with missing value markings. Based on the sliding window length that is positively correlated with the correction intensity coefficient, median filtering is performed on the effective sampling points of the threshold-processed data to obtain the local baseline of each sampling point. Calculate the deviation between the effective sampling points and the local baseline, and filter and mark candidate abnormal peaks based on the peak judgment threshold that is positively correlated with the real-time risk value; The candidate abnormal peaks are corrected to obtain corrected monitoring data.

4. The adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to claim 3, characterized in that, The candidate abnormal peaks are corrected to obtain corrected monitoring data, including: The candidate abnormal peaks are linearly interpolated using adjacent valid sampling points to obtain interpolation correction values. The interpolated correction value and the original sampled value corresponding to the candidate abnormal peak are weighted and mixed based on the correction intensity coefficient to obtain the corrected output value. The larger the correction intensity coefficient, the higher the mixing weight of the interpolated correction value, and the smaller the correction intensity coefficient, the higher the mixing weight of the original sampled value. When the real-time risk value is in the low-risk range, the candidate abnormal spike is marked as an interference spike; When the real-time risk value is in the medium-risk range, and when the deviation value exceeds the peak determination threshold by a set multiple, the corresponding sampling point is invalidated to obtain missing value marked data, and the candidate abnormal peaks that are not invalidated are marked as interference peaks. When the real-time risk value is in the high-risk range, the candidate abnormal peaks are subjected to characteristic wave morphology verification processing. Those that pass the verification are marked as confirmed crisis-type characteristic waves, and those that fail the verification are marked as peaks to be manually verified. The correction results for each interval together constitute the corrected monitoring data.

5. The adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to claim 4, characterized in that, When the real-time risk value is in the high-risk range, it also includes: The original sampled data containing candidate abnormal peak periods are processed by cache priority configuration to obtain high-risk period data with the highest retention priority in the circular cache; The data from the high-risk periods are stored at full sampling rate, while the monitoring data from other periods are stored at downsampling rate to obtain a hierarchically stored historical dataset. The timestamps, risk values, and deviation amplitudes corresponding to the candidate abnormal peaks are packaged to obtain meta-data packets; The metadata is stored in a non-volatile memory to obtain traceable feature wave annotation data.

6. The adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to claim 4, characterized in that, The confidence index of a statistical time window is calculated based on labeled candidate anomaly spikes, including: Using statistical time windows as statistical units, the total number of valid sampling points and the weighted number of candidate abnormal peaks within each time window are statistically processed. Among them, the interference peaks and the peaks to be manually reviewed are fully included in the weighted number, and the confirmed crisis characteristic waves are included in the weighted number according to the set discount weight. The credibility index for the corresponding time window is calculated based on the statistical results. The credibility index is negatively correlated with the weighted proportion of candidate abnormal peaks.

7. The adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to claim 6, characterized in that, The corrected monitoring data is non-linearly mapped in the historical display area according to the credibility index, and linearly mapped at equal time intervals in the real-time display area, including: The historical time periods corresponding to the historical display area are divided to obtain multiple consecutive statistical time windows of equal duration. The confidence index of each statistical time window is multiplied by the corresponding duration and then summed to obtain the total confidence score. When the confidence index of all statistical time windows is zero, the corrected monitoring data in the historical display area is subjected to uniform mapping at equal time intervals to obtain the right boundary pixel position of each statistical time window. When there is a statistical time window with a non-zero credibility index, the display width is allocated according to the proportion of the credibility score of each statistical time window to the total credibility score, and the right boundary pixel position of each statistical time window is obtained. The calibrated monitoring data in the real-time display area is processed by linear mapping at equal time intervals in chronological order to obtain the pixel position of each sampling point in the real-time area.

8. The adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to claim 7, characterized in that, When a statistical time window with a non-zero credibility index exists, the first [time window] in the historical display area... The right boundary pixel position of each statistical time window satisfy: in, The total pixel width of the historical display area. The sum of credibility integrals. To determine the duration of the statistical time window, For the first The credibility index of each statistical time window.

9. The adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to claim 7, characterized in that, Also includes: The width is calculated based on the right boundary pixel position of each statistical time window to obtain the initial display width of each statistical time window; Using a set proportion of the total width of the historical display area as the low confidence window marking threshold, statistical time windows with an initial display width less than the low confidence window marking threshold are marked to obtain low confidence windows; The time-continuous low-confidence windows are segmented and merged to obtain multiple independent low-confidence segments; a minimum segment width is allocated to each low-confidence segment. The sum of the minimum widths of all low-confidence segments shall not exceed the upper limit ratio of the total width of the historical display area. If it exceeds this limit, the minimum width of the smaller segment shall be adaptively adjusted to obtain the adjusted minimum width of each segment. The difference between the adjusted minimum width and the initial display width for all low-confidence segments is summed to obtain the total number of pixels added for the minimum. The total number of newly added pixels for the minimum guarantee is distributed and deducted, and the corresponding pixels are deducted from the statistical time windows that have never been marked as low confidence windows according to the proportion of their respective initial display widths; Iteratively perform width allocation and pixel subtraction processing to obtain the final display width of each statistical time window after convergence.

10. The adaptive correction and curve reconstruction method for long-term monitoring of tissue oxygen saturation according to claim 5, characterized in that, Also includes: When the real-time risk value is in the high-risk range, a red warning triangle is superimposed on the time period containing the confirmed crisis-type characteristic wave, and a yellow warning dot is superimposed on the time period corresponding to the peak to be manually reviewed, so as to obtain a high-risk trend curve with graded warning labels. In response to the triggering operation of the warning sign, the raw monitoring data of the full sampling rate for the corresponding time period is retrieved from the non-volatile memory, and the raw monitoring data of the full sampling rate is displayed at equal time intervals to obtain an expanded detailed waveform window.

Images