Debridement effectiveness evaluation method for debridement water jet scalpel

Abstract

The invention discloses a debridement efficiency evaluation method for a debridement water jet scalpel, relates to the technical field of debridement efficiency monitoring, and is used for solving the problem of lack of objective debridement evaluation. The method comprises the following steps: constructing a debridement efficiency evaluation closed loop from multi-modal measurement to strategy linkage: establishing a postoperative reference baseline, performing standardized collection, extracting fluid, surface and dynamic characteristics, and generating cleanliness, residual liquid load, tissue activity and repollution risk labels; according to the method, data quality, credibility and cross-modal, time and space constraints are combined to form confidence, multi-time scale thresholds and trends are combined to determine and output debridement insufficiency, observation maintenance and re-debridement suggestions, strategies such as treatment rhythms, parameter intervals and re-evaluation time windows are generated, track and stability and precipitation process abstracts are re-checked after execution, and a debridement result is obtained. Repollution and unnecessary dressing change are reduced, and evaluation objectivity, repeatability and traceability are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of debridement efficiency monitoring, and more particularly to a debridement efficiency evaluation method for a debridement water jet. BACKGROUND

[0002] The debridement water jet is widely used in complex wound treatment due to its selective removal of necrotic tissue and protection of healthy tissue. During the postoperative follow-up stage, the wound environment is continuously affected by exudation, dressing microenvironment, temperature and humidity, and pressure state, which easily leads to phenomena such as biofilm reattachment, local liquid retention and tissue activity fluctuation.

[0003] The existing technology has the following deficiencies: debridement follow-up evaluation is mostly dependent on naked eye observation, single-point image or exudation recording, lacks unified baseline and multi-modal fusion means, and does not establish a postoperative reference baseline and cross-batch calibration, resulting in that multi-time point data is not comparable, the evaluation results are greatly interfered by factors such as body position, illumination and dressing state, the evaluation results are highly subjective and have poor repeatability, it is difficult to identify debridement deficiency or recontamination risk in time, and nursing strategies cannot be dynamically adjusted according to quantitative results, thereby causing delay in debridement opportunity or unnecessary frequent dressing change. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the following scheme is provided to solve the problem of lack of objective evaluation of debridement in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical scheme:

[0006] A debridement efficiency evaluation method for a debridement water jet, comprising the following steps:

[0007] Establishing a postoperative reference baseline, collecting multi-modal initial data of the wound state at the time of leaving the room and forming a baseline parameter set;

[0008] Obtaining a multi-modal data set of the wound state according to a preset time table during the follow-up period, and forming a follow-up time series parameter set;

[0009] Pretreating and extracting features of the follow-up time series parameter set, and generating state labels of wound cleanliness, residual liquid load, tissue activity and recontamination risk according to cross-modal mapping;

[0010] Jointly determining the state labels under a multi-time scale integration window in combination with a threshold interval and a change trend, and outputting evaluation results of debridement deficiency, observation maintenance or debridement re-initiation;

[0011] Generating a set of nursing and follow-up strategies according to the evaluation results, and forming a process summary by gathering key parameters and time markers for quality tracing.

[0012] Further, the multi-modal dataset of wound state includes at least two of wound image data, drainage and exudation parameter data, tissue electrical parameter data, local temperature data, shallow layer ultrasound echo data, nursing operation and dressing change log data;

[0013] The follow-up period is divided according to the postoperative stage, and the collection rhythm of the early stage, the middle stage and the late stage is configured, and the collection frequency and the time interval are adaptively adjusted according to the previous evaluation result and the data quality.

[0014] Further, the follow-up timing parameter set is preprocessed and feature extracted, including the following steps:

[0015] The amplitude normalization and unit unification are performed on the multi-modal data to eliminate batch bias, and the noise suppression and artifact suppression are performed on the image and echo data;

[0016] Batch registration and time alignment are performed based on the postoperative reference baseline to form a spatial coordinate consistent with the baseline and a unified time axis;

[0017] The missing or intermittent data is interpolated and distinguished by quality marks;

[0018] Fluid-related features, tissue surface response features and dynamic change features are extracted by feature operators to form feature vectors; a mapping relationship between collection quality and feature credibility is established, and the features are down-weighted or removed;

[0019] When the registration residual or consistency deviation exceeds the threshold, the batch is frozen and temporarily excluded from the evaluation process, and a resampling prompt is issued.

[0020] Further, according to the cross-modal mapping, state labels of wound cleanliness, residual liquid load, tissue activity and recontamination risk are generated, and the specific steps include:

[0021] The follow-up timing parameter set is feature extracted to obtain a feature vector set, and feature-level fusion or decision-level fusion is used to form a fused feature;

[0022] Cross-modal consistency constraints are introduced to check different source features against each other, time continuity constraints are introduced to suppress non-physiological jumps of labels at adjacent time points, and spatial connectivity and regional topology constraints are introduced to correct local abnormalities of image features;

[0023] Based on the data quality score and the feature credibility, the label confidence is calculated, the conflicting labels are weighted and decided according to the confidence and the constraint satisfaction degree, the state labels of wound cleanliness, residual liquid load, tissue activity and recontamination risk are generated, and the confidence weight is attached.

[0024] Further, the implementation of the cross-modal consistency constraint, the time continuity constraint, and the spatial connectivity and regional topology constraint includes the following steps:

[0025] Aligning the timeline and spatial reference system of different source characteristics, constructing a multi-source characteristic matrix and completing batch consistency processing;

[0026] Calculating the consistency score of each modal feature pair for each time point, de-weighting or triggering re-sampling for low-consistency features according to the score, and retaining and generating fusion confidence for high-consistency features;

[0027] Calculating the first-order difference and second-order change of the time sequence trajectory of the state label, setting the upper limit of physiological change and the smoothing window, performing smoothing, maintaining or reverting to the latest stable state for the super-limit jump to suppress non-physiological jump;

[0028] Performing connected component analysis and morphological filtering on image features, removing isolated small patches and noise, repairing holes and limiting boundary smoothness, and simultaneously constraining adjacent structures not to intersect and region numbers to be stable according to the region topological relationship;

[0029] The consistency constraint, time continuity constraint and spatial topology constraint are combined to form a decision rule with a preset weight, and the conflict label is weighted and decided to update the label and confidence;

[0030] When any constraint fails or the conflict degree between constraints exceeds the threshold, freeze the current frame result to enter the delayed judgment and generate an abnormal marker and a reason field to write the process summary.

[0031] Further, the conflict label is weighted and decided, including the following steps:

[0032] Obtain the quality score of each batch of multi-modal data and the reliability of the corresponding features, establish a score synthesis function to fuse the quality score and feature reliability into feature weight, and normalize the feature weight;

[0033] For the four label dimensions of wound cleanliness, residual load, tissue activity and recontamination risk, generate their respective candidate label values and basic confidence according to the fused features;

[0034] According to the cross-modal consistency, time continuity and spatial topology constraints, calculate the constraint satisfaction degree, and apply constraint weighting to the basic confidence to obtain the weighted confidence;

[0035] For the candidate labels that exist differences in the same label dimension, weighted decision is performed, and the one with the maximum weighted confidence is taken as the final label. When the maximum weighted confidence is lower than the threshold, delayed judgment is triggered or the last determination is maintained and marked as uncertain.

[0036] Further, under the multi-time scale integration window, the state label is jointly determined in combination with the threshold interval and the change trend, including the following steps:

[0037] Setting short-term, medium-term and long-term three integral windows and determining the respective integral mode and time span;

[0038] Weighted integration of the time sequence value of the state label in each integral window forms a window index;

[0039] Comparing the window index with the corresponding threshold interval generates a threshold determination result;

[0040] Calculate the first-order change rate and the second-order change rate of the state label in each integral window to obtain a change trend index;

[0041] According to the preset joint rule, the threshold determination result and the change trend index are combined to generate a preliminary determination;

[0042] Perform cross-window consistency check, and only when the preliminary determinations of at least two integral windows are consistent, output the joint determination result;

[0043] Hysteresis and hold time window are applied to the situation close to the threshold boundary to suppress frequent switching;

[0044] When the results of different integral windows conflict, the final determination result is output according to the preset priority and confidence weighting.

[0045] Further, according to the evaluation result, a set of nursing and follow-up strategies is generated, and key parameters and time markers are aggregated to form a process summary for quality traceability, including the following steps:

[0046] Receive the evaluation result and analyze the state label, confidence and time scale source to build a strategy generation input set;

[0047] Based on the strategy rule base, postoperative reference baseline and resource constraints, a set of nursing and follow-up strategies is calculated, and the treatment rhythm, parameter setting interval and reevaluation time window are determined;

[0048] Perform executability and consistency check on the generated strategy set, use hysteresis and minimum execution period to suppress frequent changes, output hierarchical prompts and execution priority;

[0049] Aggregating key parameters and time markers, recording time sequence parameter statistics, state label trajectory, determination and strategy change events and generating a process summary, and giving a version mark;

[0050] Write the process summary into the traceability storage to support audit playback, and trigger notification and reminder according to the preset object.

[0051] The technical effect and advantage of the debridement efficiency evaluation method for the debridement water jet:

[0052] The application realizes dynamic perception and quantitative determination of the wound state in the postoperative follow-up scene by constructing a debridement efficiency evaluation mechanism based on multi-modal measurement and strategy linkage, establishes a postoperative reference baseline and collects a multi-modal data set of the standardized wound state, extracts fluid-related features, tissue surface response features and dynamic change features, generates a state label containing wound cleanliness, residual liquid load, tissue activity and recontamination risk, combines data quality scores, feature reliability and cross-modal consistency, time continuity, spatial connectivity and regional topological constraints to form a fusion of confidence and label confidence, and outputs evaluation results of debridement deficiency, maintenance observation or re-debridement recommendations; accordingly, a set of nursing and follow-up strategies including treatment rhythm, parameter setting interval and re-evaluation time window is generated, resource constraints and hysteresis strategies are combined to construct execution stability rules, and executable review and minimum execution cycle control of the strategies are realized.

[0053] After the strategy is executed, the stability, drift direction and consistency of the state label trajectory and window indicators are reviewed, management scheduling features are extracted, evaluation nodes and execution priorities for process control are generated, and key parameters, confidence and time markers are aggregated into a process summary to support quality traceability and audit playback, reducing recurrent recontamination and unnecessary dressing frequency, improving the objectivity, repeatability and comparability of debridement efficiency evaluation and the transparency and traceability of postoperative management processes, promoting standardized landing and facilitating cross-department collaboration and compliance management. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 A flowchart of a debridement efficiency evaluation method for a debridement water jet according to the present application. DETAILED DESCRIPTION

[0055] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.

[0056] To achieve the above-mentioned purpose, Figure 1 A structural diagram of a debridement efficiency evaluation method for a debridement water jet according to the present application is given, which specifically includes the following steps:

[0057] A postoperative reference baseline is established, and multi-modal initial data of the wound state at the time of leaving the room are collected to form a baseline parameter set;

[0058] In the follow-up period, multi-modal data sets of the wound state are obtained according to a preset time table, and a follow-up time series parameter set is formed;

[0059] The follow-up time series parameter set is preprocessed and feature extracted, and the state label of wound cleanliness, residual liquid load, tissue activity and repollution risk is generated according to the cross-modal mapping;

[0060] The state label is jointly determined under the multi-time scale integral window combined with the threshold interval and the change trend, and the evaluation results of debridement deficiency, maintenance observation or re-debridement proposal are output;

[0061] According to the evaluation results, the nursing and follow-up strategy set is generated, and the process summary is formed by gathering key parameters and time markers for quality tracing.

[0062] Step 1, establish the postoperative reference baseline, collect the initial data of wound state multi-modal at the time of leaving the room and form the baseline parameter set, which is specifically implemented as:

[0063] After the patient completes the debridement water jet operation and leaves the room, the postoperative reference baseline is immediately established in the ward environment, and the initial data of wound state multi-modal is collected under the condition of uniform shooting distance and constant color temperature lighting, including at least two of wound image data, drainage and exudation parameter data, tissue electrical parameter data, local temperature data, shallow ultrasonic echo data and nursing operation and dressing change log data; Each data is collected at the same time point and given a uniform time marker to form a baseline parameter set, wherein: the wound image data is used to provide the wound surface morphology and texture reference; The drainage and exudation parameter data are converted into standard readings according to the drainage container calibration; The tissue electrical parameter data are obtained by surface electrode and the electrode contact state is recorded; The local temperature data are obtained by contact or non-contact body temperature measurement and the probe self-checking result is recorded; The shallow ultrasonic echo data are collected by bedside shallow ultrasound and the preset gain and depth are recorded; The nursing operation and dressing change log data record the key points of the current treatment and the batch of consumables.

[0064] The above data is confirmed as the postoperative reference baseline after the collection is completed, and all subsequent follow-up data is compared with the baseline parameter set as the spatial coordinate and time marker.

[0065] Step 2, obtain the multi-modal data set of wound state according to the preset time table in the follow-up period, and form the follow-up time series parameter set, which is specifically implemented as:

[0066] In the follow-up period, the postoperative stage is divided into early stage, middle stage and late stage, and the collection frequency and time interval are configured to form a preset time table; The rhythm of early stage is relatively dense, the rhythm of middle stage is moderate, and the rhythm of late stage is relatively sparse.

[0067] At each follow-up, the multi-modal data set of wound state is obtained according to the preset time table, and all data at the same time point are aggregated to form the follow-up time series parameter set;

[0068] The follow-up time-series parameter set includes at least two of the paired wound image data, drainage and exudation parameter data, tissue electrical parameter data, local temperature data, shallow layer ultrasound echo data, and nursing operation and dressing change log data under the same time label.

[0069] Step 3, preprocessing and feature extraction are performed on the follow-up time-series parameter set, and state labels of wound cleanliness, residual load, tissue activity and recontamination risk are generated according to cross-modal mapping, which is implemented as follows:

[0070] The follow-up time-series parameter set is preprocessed in batches. First, amplitude normalization and unit unification are performed: white balance and brightness standardization are performed on wound image data using a reference color card to eliminate differences in light between follow-ups; drainage and exudation parameter data are converted to a unified unit of measurement according to the drainage container calibration coefficient; tissue electrical parameter data are corrected according to the electrode contact self-check record to reduce the reading deviation caused by contact impedance changes; local temperature data are corrected for zero and range according to the temperature probe calibration record; and shallow layer ultrasound echo data are intensity standardized according to a unified gain and depth setting.

[0071] Subsequently, noise suppression and artifact suppression are implemented: dark field correction is used in image data to reduce random noise while preserving edge details, and neighborhood consistency smoothing is used in shallow layer ultrasound echo data to suppress shot noise and locally correct stripes caused by slight changes in probe angle, and short window smoothing is used in tissue electrical parameter data and local temperature data to remove occasional jumps but preserve slow trends.

[0072] After standardization, batch registration and time alignment are performed with the postoperative reference baseline as a control. Image data is aligned to the reference frame by aligning the current frame spatial coordinates to the reference frame using the wound margin, marker stickers or disposable positioning grid as the alignment constraint; shallow layer ultrasound echo data is aligned by comparing the reference layer according to the probe position marker, scanning path and depth configuration to ensure consistent echo sources; and the time labels of all modalities are uniformly mapped to a unified time axis consistent with the baseline. When there is missing or intermittent data, interpolation is performed according to two-level rules and quality labels are added: if the modality shows a stable trend at adjacent time points, the current value is estimated from the adjacent stable segment; if there is a significant trend change, the current value is estimated from the nearest stable segment of the same modality; if it cannot be reliably estimated, it is marked as not segmented and does not participate in subsequent weight calculation.

[0073] Registration residual and consistency deviation are defined: registration residual is the maximum deviation between the aligned key anatomical landmarks and the reference position; consistency deviation is the difference between images or echoes taken repeatedly in the same body position, when the registration residual or consistency deviation exceeds the preset threshold, the batch is frozen and does not enter the evaluation process, and a resampling prompt is generated, and the abnormal reason is recorded in the nursing operation and dressing change log data to ensure that the data entering the feature extraction has stable comparability.

[0074] The process of feature extraction, feature vector formation and feature credibility assignment is as follows:

[0075] After registration, time alignment and quality control are completed, three types of features are extracted around the follow-up timing parameter set and the feature vector is formed, the fluid-related features are obtained from the drainage and exudation parameter data, including the change direction per unit time, the change amplitude stability and the consistency between adjacent follow-ups, which are used to reflect the residual liquid cleaning trend; the tissue surface response features are obtained from the wound image data and tissue electrical parameter data, including the wound edge continuity, texture uniformity, color consistency and electrical reading stability, which are used to reflect the surface tissue state; the dynamic change features are obtained from the local temperature data and shallow ultrasonic echo data, including the slow change amplitude of local temperature within the follow-up window and the echo intensity and echo interface continuity, which are used to reflect the reaction of tissues at different depths.

[0076] In order to reasonably assign weights in subsequent cross-modal mapping, the embodiment establishes a mapping relationship between acquisition quality and feature credibility: taking the data quality score as input, combining the acquisition condition record of this modality (such as whether the electrode contact is stable, whether the image is clear, and whether the echo level is consistent with the baseline), the feature credibility of each type of feature is calculated; when the feature credibility is lower than the preset threshold, the weight of this type of feature is reduced or eliminated, and the remaining all features and their feature credibility jointly constitute the feature vector and quality label set, which provides controlled input for cross-modal mapping.

[0077] The feature vector set is input into the cross-modal mapping process and the state label is generated. First, select between feature-level fusion and decision-level fusion according to data integrity and data quality score: when multiple modalities are available and the data quality score is high, use feature-level fusion to directly form fused features; when there is a non-segmentation label or the credibility of the features of a certain modality is low, use decision-level fusion to reduce the influence of single modality distortion.

[0078] Then, three types of constraints are introduced in turn and the constraint satisfaction degree is calculated: the cross-modal consistency constraint is used for mutual verification between different source features, the time continuity constraint is used to suppress non-physiological jumps of the state label at adjacent follow-up times, and the spatial connectivity and regional topology constraint is used to correct local anomalies of image features (such as isolated small patches, holes or unreasonable boundary fluctuations), each type of constraint gives a qualitative judgment of whether it is satisfied and a quantitative grading of the satisfaction degree.

[0079] Based on the data quality score and feature credibility, the label confidence of each state label dimension (wound cleanliness, residual load, tissue activity, and recontamination risk) is calculated: first, the candidate label value and basic confidence are obtained according to the fused features, and then the basic confidence is weighted according to the satisfaction degree of the three types of constraints to obtain the label confidence. If there are multiple candidate label values for the same dimension, the label confidence and constraint satisfaction degree are weighted to determine the final state label of the dimension, and the highest weighted value is taken as the final state label of the dimension; when the highest weighted value is still lower than the uncertainty threshold, the state label of the last follow-up is maintained and marked as uncertain, and the next follow-up is reviewed.

[0080] Before entering the constraint and decision-making process, the wound image data, drainage and exudation parameter data, tissue electrical parameter data, local temperature data, and shallow ultrasound echo data obtained at the same follow-up time point are aligned in the time axis and spatial reference system. The time axis alignment takes the unified time marker as the reference to ensure that the data of each modality corresponds to the same follow-up time; the spatial reference system alignment takes the postoperative reference baseline wound margin, positioning grid, or anatomical reference point as the anchor point to map the current image and echo coordinates to the same coordinate system as the postoperative reference baseline.

[0081] Subsequently, batch consistency processing is carried out: for the offset affected by illumination, gain, and probe angle, adjust according to the preset standardization sequence (first light and gain, then angle and layer), to ensure comparability of different batches. On this basis, a multi-source feature matrix is constructed, including image texture, edge definition, color uniformity, echo intensity and interface continuity, drainage and exudation change, tissue electrical stability, local temperature change amplitude, etc.

[0082] The steps for generating cross-modality consistency constraints and fusion confidence are:

[0083] For each follow-up time point, the consistency score of each pair of features from different sources in the multi-source feature matrix is calculated. The definition of consistency score is: according to the three rules of direction consistency, change amplitude comparability, and abnormality in the same direction, the consistency score is calculated by weighting the three rules with preset weights and summing them up.

[0084] For all pairs of features at the same time point, the median value of the consistency score distribution is taken and combined with the quantile interval to form the overall evaluation of cross-modality consistency at that time point.

[0085] When the consistency score of a pair of features is in the inconsistent level, the feature credibility corresponding to the pair of features is reduced by one level.

[0086] When most of the pair features belong to consistent or substantially consistent, the fusion confidence of the time point is calculated, and the fusion confidence is obtained in the following way: taking the data quality score as the basis weight, taking the pair feature consistency score as the adjustment weight, weighing and collecting the confidence of each modal feature participating in fusion to obtain the fusion confidence for subsequent label confidence correction; if the fusion confidence is lower than the preset threshold, mark the time point as needing review, and only participate in the conservative judgment of time continuity constraint subsequently.

[0087] For example, at the follow-up moment, the wound image data, shallow layer ultrasound echo data, tissue electrical parameter data and local temperature data are obtained, and their data quality scores and feature confidence are 0.90 / 0.85, 0.80 / 0.75, 0.70 / 0.80 and 0.60 / 0.65 respectively. The consistency score of each modality is calculated, and the result is mostly in the interval of 0.75-0.90, with a median value of about 0.80, indicating that the overall consistency is high. Then, the data quality score of each modality is multiplied by the feature confidence to obtain the weight contribution (image 0.765, ultrasound 0.600, electrical 0.560, temperature 0.390), the total is 2.315, and then combined with the consistency coefficient 0.80 to obtain 1.852, which is finally normalized to 0.78 according to the batch standard, which is used as the fusion confidence, indicating that the overall reliability of the multi-modal at the follow-up moment is high, which can be used for subsequent state label judgment.

[0088] The construction steps of the time continuity constraint, the spatial connectivity and the regional topological constraint are as follows:

[0089] For each state label dimension (wound cleanliness, residual load, tissue activity and recontamination risk), a state label time sequence trajectory is established, and the change amount (as a first-order change) between adjacent follow-up time points and the difference size (as a second-order change) between adjacent change amounts are calculated, which are as follows:

[0090] The difference size between the current value and the previous follow-up value is compared, and if it exceeds the upper limit of the physiological change prepared according to the postoperative reference baseline and the clinical reasonable range, it is determined as an out-of-limit jump. Further, the difference size between the current change amount and the last change amount is compared, and if the difference size is sharply enlarged in a short time, it is determined as an incoherent mutation. When the out-of-limit jump occurs, the smoothing, maintaining or backtracking rules are applied in sequence: the current value is preferentially smoothed by a fixed length smoothing window; if the smoothed value still exceeds the upper limit of the physiological change, the label value of the last follow-up is kept unchanged; if the incoherent mutation still occurs, the label value of the last time marked as a stable state is returned.

[0091] For the spatial connectivity constraint, first, the connected domain analysis and morphological screening are performed in the wound image data: small area patches isolated from the main region are removed and the boundary is smoothed; the region with holes is closed and repaired morphologically to make the region boundary continuous;

[0092] Finally, the area topology constraint is applied, i.e. adjacent structures cannot intersect, the area number remains stable between adjacent follow-ups, and the area number is only allowed to change when it is jointly supported by the consistency score and the data quality score. The satisfaction degree of the spatial connectivity constraint is recorded as complete satisfaction, basic satisfaction, and non-satisfaction, and is input as the constraint satisfaction degree for subsequent adjudication.

[0093] The cross-modality consistency constraint, the temporal coherence constraint, and the spatial connectivity and area topology constraint are combined into an adjudication rule with preset weights. The determination method of the preset weights is as follows:

[0094] According to the stability of historical follow-up data in the target population, a higher weight is given to the temporal coherence to preferentially suppress non-physiological jumps. When the wound image data quality is high, the weights of the spatial connectivity and area topology constraints are increased. When the overall data quality score is low, the weight of the cross-modality consistency constraint is increased to avoid single modality misdirection. In specific implementation, the candidate labels with differences are compared by weighting based on the joint value of the label confidence and the constraint satisfaction degree;

[0095] The joint value is obtained as follows: first, the label confidence is corrected by the fusion confidence, then the satisfaction degrees of the three types of constraints are continuously adjusted by the weights, and finally a comparable joint value is formed according to the preset weight order. The candidate label with the highest joint value is taken as the preliminary label at the current time point.

[0096] If any of the three types of constraints fails or the conflict degree between the constraints exceeds the preset threshold, the freezing mechanism is triggered, i.e. the preliminary label of the current frame is frozen and marked as delayed determination. At the same time, an abnormal marker and a reason field are generated and written into the process summary, and the time point and the next time point are bundled and reviewed at the next follow-up time point.

[0097] The weighting adjudication and uncertainty threshold processing process for conflicting labels are as follows:

[0098] When multiple candidate labels appear in the same state label dimension, the weighting adjudication is performed. First, the data quality score and the feature confidence of each feature of the batch are obtained, and the feature weight is generated by the score synthesis function. The specific implementation of the score synthesis function is as follows: first, the data quality score and the feature confidence are respectively mapped into uniform level scores, then the two are added according to the preset proportion to form the original weight, and finally the normalization processing is performed to limit the sum of all feature weights to the same reference value to ensure the comparability between different time points. Subsequently, for the four dimensions of wound cleanliness, residual load, tissue activity, and recontamination risk, the respective candidate label values and basic confidence are generated according to the fusion features.

[0099] According to the satisfaction degrees of the three types of constraints, a constraint satisfaction degree is calculated, and the constraint satisfaction degree is used to weight adjust the basic confidence to obtain a weighted confidence.

[0100] In the same dimension, the candidate label with the maximum weighted confidence is taken as the final state label; if the maximum weighted confidence is still lower than the uncertainty threshold, two measures are performed: one is to maintain the final state label of the last follow-up unchanged and mark it as uncertain, and the other is to trigger a delayed decision to enter the bundled review of the next follow-up time point. After completing the decision, the final state label, weighted confidence, constraint satisfaction degree, feature weight and abnormal reason field are written into the process summary to support subsequent quality tracing and audit playback.

[0101] For example, at a certain follow-up time, three candidate labels are generated for the residual load dimension: high (0.62), medium (0.58) and low (0.52). After adjustment of the three types of constraints, “high” is supported by ultrasound and drainage trend and is added to 0.70, “medium” is added to 0.60, and “low” is reduced to 0.48 due to inconsistency and space constraints. Combined with the fusion confidence 0.78 at this time, it can be confirmed that the basic modal data is stable, and the weighted confidence can be directly compared. The highest value is 0.70, corresponding to the candidate “high”, and higher than the uncertainty threshold 0.60, so the final output state label is “residual load is high”. This conclusion will enter the subsequent joint decision and nursing strategy generation process to guide whether to prompt the debridement.

[0102] Step 4: Joint decision of state label under multi-time scale integration window combined with threshold interval and change trend, output evaluation results of debridement deficiency, maintenance observation or debridement creation proposal, specific implementation is:

[0103] Before entering the joint decision, set short-term integration window, medium-term integration window and long-term integration window for each state label (wound cleanliness, residual load, tissue activity and recontamination risk), and clearly define their respective integration methods and time span; for the same state label, process the time sequence value of the label in each integration window in chronological order: multiply the label value of each follow-up with the data quality score and the corresponding feature reliability, then accumulate, and at the same time, accumulate the weights, finally get the window index of the integration window by dividing the accumulated value by the total weight.

[0104] Subsequently, according to the threshold interval determined by the postoperative reference baseline, historical sample distribution and rule base, the window index is compared one by one in the interval and the threshold decision result is given; among them, the wound cleanliness and tissue activity are usually more beneficial when they are high in the threshold interval, the residual load and recontamination risk are usually more beneficial when they are low in the threshold interval, if the window index falls into the risk interval, it forms an unfavorable decision, falls into the safe interval, it forms a favorable decision, and between them is a critical decision.

[0105] The trend indicator of the state label is calculated simultaneously in each integration window, with the first-order change rate reflecting the change direction and amplitude of the current value relative to the last follow-up value, and the second-order change rate reflecting the amplification or convergence of the current change relative to the last change.

[0106] The first-order change rate is used to distinguish between improvement, deterioration, and basic stability; the second-order change rate is used to identify sudden increases and decreases and slow changes, and the isolated abnormality is suppressed by means of the holding time window: when the single change is inconsistent with the adjacent period, it is observed in the holding time window first, and only after the same change occurs continuously for multiple times is the trend confirmed. Then, in each integration window, the threshold determination result and the change trend indicator are combined according to the preset joint rule to obtain the preliminary determination: for example, when the window indicator is in the safe interval and the trend is improvement or basic stability, the preliminary determination tends to maintain observation; when the window indicator is in the risk interval and the trend is deterioration or deterioration is amplifying, the preliminary determination tends to clear the creation proposal; when the window indicator is in the critical interval, if the trend converges to the favorable direction, it tends to maintain observation, and if the trend diverges to the unfavorable direction, it tends to clear the insufficient creation. The above joint rule is based on the threshold and corrected by the trend, ensuring that both the current level and the trend and stability are considered.

[0107] To obtain the final evaluation result, the cross-window consistency check is performed on the preliminary determinations of the three integration windows, and only when the preliminary determinations of at least two integration windows are consistent, the joint determination result is output; if it is close to the threshold boundary, a hysteresis strategy is introduced to set the upper and lower limit difference and the minimum holding time, avoiding frequent switching between adjacent two types of conclusions.

[0108] When there is a conflict among the preliminary determinations of the three integration windows, the decision is made according to the preset priority and confidence weighting: the priority is sorted according to the closeness to the current time, the data integrity in the window, and the stability of the data quality score; the confidence weighting is obtained by synthesizing the stability of the window indicator, the consistency of the trend indicator in the holding time window, and the fusion confidence. The decision-making process is as follows: first, compare the two windows with high priority, if there is still a conflict, introduce the third window as a weighting factor to form the final determination result.

[0109] When the final determination result is clear and insufficient, maintain observation or clear and create a proposal, the window indicator used for determination, the threshold interval position, the trend category, the trigger condition of hysteresis and holding time window, and the decision-making reason of priority and confidence weighting are recorded simultaneously, and the process summary is written for quality traceability and subsequent review.

[0110] Step 5, generate the set of nursing and follow-up strategies according to the evaluation results, and gather the key parameters and time markers to form the process summary for quality traceability, the specific implementation is as follows:

[0111] After obtaining the evaluation results, first parse the state label, confidence and time scale source in the evaluation results, together with the postoperative reference baseline, the last cycle strategy and resource constraints to form the strategy generation input set. The strategy rule base is composed of rule entries, each containing trigger conditions (state label combination, confidence hierarchy, time scale source and postoperative stage), output items (treatment rhythm, parameter setting interval and reevaluation time window) and applicable scope (wound site, dressing type, follow-up scenario) and entry priority, resource constraints include personnel shift, consumable inventory, equipment available period and bedside examination capacity.

[0112] The strategy calculation process is: matching rule entries according to state label severity and confidence; comparing the output items with the postoperative reference baseline to make individualized fine-tuning (such as tightening the parameter setting interval or shortening the treatment rhythm within the baseline tolerable range); when multiple rules are satisfied at the same time and point to different conclusions, prioritize the label severity, then the time scale, and finally the entry priority order; the treatment rhythm is defined as the execution frequency and minimum interval of dressing change and follow-up, the parameter setting interval is defined as the upper and lower limits of the safety range of fluid management and monitoring items, and the reevaluation time window is defined as the earliest and latest time boundary of the next reevaluation. The above three types of output form the nursing and follow-up strategy set, accompanied by the generation reason and the cited rule identifier.

[0113] To ensure the strategy is executable and stable, subsequent executability and consistency checks are performed: the executability check checks whether the resource constraints meet the treatment rhythm and parameter setting interval, if not, the lower limit measures related to safety are retained first, and non-critical items are relaxed according to the pre-set degradation table; the consistency check compares whether the new and old strategies have conflicting frequencies or ranges, if there is a conflict, the safe side is used and the conflict pair is recorded.

[0114] To suppress frequent changes, hysteresis and minimum execution period can also be introduced: keep the current strategy unchanged before crossing the upgrade or downgrade threshold; only when multiple consecutive follow-up points meet the upgrade or downgrade conditions and reach the minimum execution period, switch.

[0115] According to the state label severity and confidence output classification prompts (observation level, intervention level, reevaluation creation proposal level) and execution priority, and gather key parameters and time markers into a process summary, which at least contains timing parameter statistics, state label trajectory, joint decision record, strategy change event and cited rule identifier, and is assigned a version marker (including strategy version, rule base version and baseline version). Finally, the process summary is written into the trace storage to support audit playback, when the evaluation results reach the reevaluation creation proposal level or the early trigger condition appears within the reevaluation time window, immediately send a notification to the nursing team and the responsible doctor, at the same time, push a reminder to the patient or caregiver, the rest of the levels are sent according to the pre-set batch strategy, to ensure information accessibility and process closure.

[0116] It should be noted that the threshold information related in this embodiment is set by professionals in advance, and is not explained too much here. When used, different meanings are explained, and are not explained one by one here.

[0117] The application realizes dynamic perception and quantitative determination of wound state in postoperative follow-up scene by constructing a debridement efficiency evaluation mechanism based on multi-modal measurement and strategy linkage. By establishing a postoperative reference baseline and collecting a standardized wound state multi-modal data set, fluid-related features, tissue surface response features and dynamic change features are extracted, a state label containing wound cleanliness, residual liquid load, tissue activity and recontamination risk is generated, and an evaluation result of debridement deficiency, maintenance observation or re-cleaning recommendation is output in combination with data quality score, feature credibility and cross-modal consistency, time continuity, spatial connectivity and regional topological constraints, forming a fusion of confidence and label confidence. A set of nursing and follow-up strategies including treatment rhythm, parameter setting interval and reevaluation time window is generated accordingly, and execution stability rules are constructed in combination with resource constraints and hysteresis strategies to realize executable review and minimum execution cycle control of the strategies.

[0118] After the strategy is executed, the stability, drift direction and consistency of the state label trajectory and window indicators are reviewed, management scheduling features are extracted, evaluation nodes and execution priorities for process control are generated, and key parameters, confidence and time markers are aggregated into a process summary to support quality traceability and audit playback, reducing recurrent recontamination and unnecessary dressing frequency, and improving the objectivity, repeatability and comparability of debridement efficiency evaluation and the transparency and traceability of postoperative management processes. Promote standardized landing and facilitate cross-department collaboration and compliance management.

[0119] The above embodiments can be realized wholly or partially by software, hardware, firmware or any combination thereof. When realized by software, the above embodiments can be realized in the form of a computer program product, wholly or partially.

[0120] Those of ordinary skill in the art can realize that the modules and algorithm steps of the examples described in conjunction with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solutions. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0121] In addition, each functional module in each embodiment of the present application can be integrated in one processing module, or each module can exist physically alone, or two or more modules can be integrated in one module.

[0122] The above merely provides the specific implementation of the present application, but the protection scope of the present application is not limited to this. Any modification or replacement within the technical range disclosed by the present application can be easily thought by any person skilled in the art, and should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0123] Finally, the above merely provides the preferred embodiments of the present application, and is not used to limit the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for evaluating the debridement efficacy of water jet debridement, characterized in that: Includes the following steps: Establish a postoperative reference baseline, collect multimodal initial data on wound status upon exiting the operating room, and form a baseline parameter set; During the follow-up period, a multimodal dataset of wound status is acquired according to a preset schedule, and a follow-up time series parameter set is formed. The follow-up time series parameter set was preprocessed and features were extracted. Based on cross-modal mapping, status labels of wound cleanliness, residual fluid load, tissue viability and risk of recontamination were generated. Under a multi-timescale integration window, the status label is jointly determined by combining the threshold range and the trend of change, and the assessment results of insufficient debridement, maintenance of observation, or re-debridement are output. Based on the assessment results, a set of nursing and follow-up strategies is generated, and key parameters and time stamps are combined to form a process summary for quality traceability.

2. The method for evaluating the debridement efficacy of a water jet debridement device according to claim 1, characterized in that: The multimodal dataset of wound status includes at least two of the following: wound imaging data, drainage and exudation parameter data, tissue electrical parameter data, local temperature data, superficial ultrasound echo data, and nursing operation and dressing change log data. The follow-up period is divided according to the postoperative stage, and the collection rhythms for the early, middle and late stages are configured. The collection frequency and time interval are adaptively adjusted based on the previous assessment results and data quality.

3. The method for evaluating the debridement efficacy of a water jet debridement device according to claim 2, characterized in that: The preprocessing and feature extraction of the follow-up time-series parameter set includes the following steps: Amplitude normalization and unit unification are performed on multimodal data to eliminate batch bias, and noise and artifact suppression are implemented on image and echo data. Batch registration and time alignment were performed based on the postoperative reference baseline to form spatial coordinates and a unified time axis consistent with the baseline. Imput missing or discontinuous data and distinguish them with quality markers; Fluid-related features, tissue surface response features, and dynamic change features are extracted using feature operators to form feature vectors; a mapping relationship between acquisition quality and feature reliability is established, and features are deweighted or eliminated. When the registration residual or consistency deviation exceeds the threshold, the batch is frozen, temporarily suspended from the evaluation process, and a re-sampling prompt is issued.

4. The method for evaluating the debridement efficacy of a water jet debridement device according to claim 3, characterized in that: The process of generating status labels for wound cleanliness, residual fluid load, tissue viability, and risk of recontamination based on cross-modal mapping includes the following steps: The follow-up time series parameter set is used to extract features to obtain a set of feature vectors, and feature-level fusion or decision-level fusion is used to form fused features; Cross-modal consistency constraints are introduced to cross-verify features from different sources, temporal coherence constraints are introduced to suppress non-physiological jumps in labels at adjacent time points, and spatial connectivity and regional topology constraints are introduced to correct local anomalies in image-like features. The confidence level of the labels is calculated based on the data quality score and feature credibility. Conflicting labels are weighted and adjudicated according to confidence level and constraint satisfaction. Status labels of wound cleanliness, residual fluid load, tissue viability and recontamination risk are generated and attached with confidence level weights.

5. The method for evaluating the debridement efficacy of a water jet debridement device according to claim 4, characterized in that: The implementation of cross-modal consistency constraints, temporal coherence constraints, and spatial connectivity and regional topology constraints includes the following steps: Align features from different sources with the time axis and spatial reference frame, construct a multi-source feature matrix and complete batch consistency processing; For each time point, calculate the consistency score of each modal feature pair, reduce the weight of low consistency features or trigger resampling based on the score, and retain high consistency features and generate fusion confidence. Calculate the first-order difference and second-order change of the temporal trajectory of the status label, set the upper limit of physiological change and smoothing window, and perform smoothing, maintenance or regression to the most recent stable state for the over-limit jump to suppress non-physiological jumps. Connectivity analysis and morphological screening are performed on image features to remove isolated small patches and noise, repair holes and limit boundary smoothness, while constraining adjacent structures to not intersect and region numbering to be stable based on regional topological relationships. Consistency constraints, temporal coherence constraints, and spatial topology constraints are used to form adjudication rules with preset weights. Conflicting labels are then weighted and the labels and confidence levels are updated. When any constraint fails or the degree of conflict between constraints exceeds the threshold, the current frame result is frozen, and a delayed judgment is performed, and an exception marker and cause field are generated and written into the process summary.

6. The method for evaluating the debridement efficacy of a water jet debridement device according to claim 5, characterized in that: The weighted decision-making process for conflicting labels includes the following steps: Obtain the quality score and corresponding feature credibility of each batch of multimodal data, establish a score synthesis function to integrate the quality score and feature credibility into feature weights, and normalize the feature weights. For the four label dimensions of wound cleanliness, residual fluid load, tissue viability and risk of recontamination, candidate label values ​​and basic confidence scores are generated for their respective dimensions based on the fusion characteristics. The constraint satisfaction is calculated based on cross-modal consistency, temporal coherence, and spatial topological constraints. The weighted confidence is obtained by applying constraints to the basic confidence score. For candidate labels that differ within the same label dimension, a weighted decision is made, and the label with the highest weighted confidence score is selected as the final label. When the maximum weighted confidence score is lower than the threshold, a delayed decision is triggered or the previous decision is maintained and marked as uncertain.

7. The method for evaluating the debridement efficacy of a water jet debridement device according to claim 6, characterized in that: The joint determination of state labels by combining threshold intervals and changing trends under multi-timescale integration windows includes the following steps: Set up three integration windows: short-term, medium-term, and long-term, and determine their respective integration methods and time spans; The time-series values ​​of the status labels are weighted and integrated within each integration window to form a window index; The threshold determination result is generated by comparing the window index with the corresponding threshold range. Within each integration window, the first-order and second-order rates of change of the status labels are calculated to obtain the trend index. The threshold judgment result is combined with the trend indicator according to the preset joint rules to generate a preliminary judgment; Perform cross-window consistency checks and output the joint decision result only if the preliminary decisions of at least two integration windows are consistent; Hysteresis and hold time windows are applied to cases approaching the threshold boundary to suppress frequent switching; When results from different integration windows conflict, they are weighted according to preset priority and confidence level, and the final judgment result is output.

8. The method for evaluating the debridement efficacy of a water jet debridement device according to claim 7, characterized in that: Based on the assessment results, a set of nursing and follow-up strategies is generated, and key parameters and time stamps are combined to form a process summary for quality traceability, including the following steps: Receive the evaluation results and parse the source of state labels, confidence levels and time scales to construct the policy generation input set; Based on the strategy rule base, postoperative reference baseline and resource constraints, the nursing and follow-up strategy set is calculated to determine the treatment rhythm, parameter setting interval and re-evaluation time window; Perform executability and consistency checks on the generated strategy set, use hysteresis and minimum execution cycle to suppress frequent changes, and output hierarchical prompts and execution priorities; It aggregates key parameters and time stamps, records time-series parameter statistics, status label trajectories, judgment and strategy change events, generates process summaries, and assigns version tags; Write process summaries to traceability storage to support audit playback, and trigger notifications and alerts for preset objects.

Images