A method for intraoperative lateral diffused wave shape recognition and vanishing point determination based on template state decoding and MVMD enhancement

Abstract

The application discloses an intraoperative lateral spread waveform recognition and disappearance point determination method based on template state decoding and MVMD enhancement, and relates to the technical field of intraoperative nerve monitoring.The technical scheme is as follows: after waveform data is preprocessed, the original signal, the MVMD enhanced signal or the fusion signal thereof can be selected as an analysis source;high-confidence abnormal waveforms are screened in the same file to construct individualized abnormal templates, and a baseline reference is constructed from early waveforms as a negative evidence model;the latency constraint matching is performed on each waveform, the multi-dimensional evidence is fused to calculate a comprehensive score;the state decoding is performed based on the score sequence to obtain a smooth waveform-by-waveform state sequence;finally, the disappearance point is determined according to the number of continuous normal states at the end and a preset threshold value.The application realizes the individualization, robustness of abnormal wave recognition and the quantifiability and interpretability of disappearance point determination through adaptive template construction in the same file, positive and negative evidence joint evaluation and sequence context decoding.

Description

Technical Field

[0001] This invention relates to the field of intraoperative neurological monitoring technology, and more specifically, to a method for intraoperative lateral diffusion waveform identification and vanishing point determination based on template state decoding and MVMD enhancement. Background Technology

[0002] Hemifacial spasm is a chronic neuromuscular functional disorder characterized by involuntary, paroxysmal twitching of the facial muscles on one side of the face, severely impacting patients' quality of life and social function. It is prevalent among middle-aged individuals, with epidemiological data showing that the incidence rate in women is approximately twice that of men. Hemifacial spasm includes two main types: primary and secondary, with primary cases accounting for over 90%. Its pathological basis is compression of the facial nerve root at the recurrent exocranial zone (REZ), causing repeated and continuous stimulation of the facial nerve. Regarding the pathogenesis of primary hemifacial spasm, two main theories have emerged: the central hypothesis and the peripheral hypothesis. Regardless of the hypothesis, vascular compression is the fundamental cause of hemifacial spasm. Currently, microvascular decompression (MVD) is the preferred surgical treatment for hemifacial spasm. Its core principle is to separate the blood vessel compressing the facial nerve and insert a special material to physically isolate the blood vessel from the nerve, thereby relieving the pressure of the responsible blood vessel on the facial nerve.

[0003] Intraoperative neurophysiological monitoring is a core means of ensuring the efficacy of facial nerve dissection (MVD) surgery and guiding intraoperative decision-making, and the lateral spread response (LSR) is the key to intraoperative neurophysiological monitoring in MVD. LSR is the abnormal muscle response (AMR) recorded on the muscles innervated by other branches of the facial nerve when a branch is stimulated; it is a characteristic pathophysiological signal of hemifacial spasm. Most current intraoperative neurophysiological monitoring techniques monitor physiological waves to reflect the integrity of the corresponding nerve conduction pathways and protect nerve function. During surgery, we aim to ensure the stable existence of this waveform, which can be called safety monitoring. LSR, on the other hand, monitors pathological waves, reflecting the specific manifestations of the disease itself. The goal of surgery is to make this waveform disappear, which can be called effectiveness monitoring. The core criterion for judging the efficacy of facial nerve MVD surgery is to observe the amplitude changes of this pathological waveform, its gradual weakening until it stably disappears, thereby determining whether vascular decompression is sufficient. Clinical studies have confirmed that the disappearance of LSR waveforms during surgery is highly correlated with long-term patient prognosis: when both zygomatic and mandibular marginal ramus LSRs are monitored simultaneously, if both LSR waveforms completely disappear at the end of surgery, the probability of spasm recurrence one year postoperatively is only 2.8%; if only one waveform disappears, the probability of poor prognosis rises to 23.5%; and if neither waveform disappears, the probability of poor postoperative prognosis is as high as 72.7%. Therefore, accurately identifying abnormal LSR / AMR waves within the continuous waveform stack during surgery and quantifying their disappearance is a crucial step in improving surgical efficacy, standardizing intraoperative decision-making, and objectively assessing prognosis.

[0004] However, the interpretation of intraoperative neuromonitoring waveforms still largely relies on manual observation. Operators typically make comprehensive judgments based on waveform amplitude, morphology, latency, and trends, which presents the following significant challenges:

[0005] Highly subjective and lacking consistency: Different operators may have inconsistent understandings of the boundaries of abnormal waves, weak abnormal waves, and the point at which they disappear. When there are many consecutive stacked waveforms (up to dozens or hundreds), strong noise fluctuations, or subtle waveform changes, manual methods are particularly prone to instability in judgment, leading to differences in interpretation results between different operators or even the same operator at different times for the same case.

[0006] Fixed threshold methods have insufficient adaptability: Existing automated interpretation methods commonly rely on anomaly detection based on absolute amplitude thresholds, relative thresholds, or simple statistical characteristics. While these methods are relatively simple to implement, they often depend on preset threshold parameters. Due to individual differences among patients, variations in noise levels under different acquisition conditions, and dynamic changes in waveform morphology during the same surgical procedure, a uniform threshold is difficult to maintain consistent effectiveness across different cases. When the threshold is set too high, it is easy to miss low-amplitude but clinically significant abnormal waves; when the threshold is set too low, it is easy to misinterpret noise or slight fluctuations as abnormal waves, resulting in false positive results.

[0007] Lack of context-based sequential interpretation: Existing methods often treat each waveform as an independent object for static interpretation, lacking utilization of the contextual relationships between the entire waveform stack. In reality, intraoperative monitoring waveforms are usually arranged in chronological or stimulus order, and there is a continuous evolutionary relationship between adjacent waveforms. The formation, persistence, weakening, and disappearance of abnormal waves are not isolated events, but rather follow a certain sequential state pattern—abnormal waves usually concentrate in a certain stage, then gradually weaken and transition to a sustained normal state. If this continuity of state is ignored, and independent interpretation is made based solely on the amplitude or local characteristics of a single waveform, it is easily affected by sporadic noise, resulting in frequent jumps between abnormal / normal states, thereby reducing the stability and clinical reliability of the overall results.

[0008] The lack of a unified quantitative mechanism for vanishing point determination is a significant issue. Intraoperative focus is not only on whether a waveform is abnormal, but also on whether the abnormal waveform has stably disappeared – this directly relates to the basis for determining the surgical endpoint. While existing manual observation methods can roughly determine the trend of waveform changes, they lack a unified, quantifiable, and repeatable vanishing point determination mechanism. Especially when there are slight fluctuations at the end of the waveform, minor latency shifts, or individual residual morphological abnormalities, operators often struggle to accurately determine whether vanishing has occurred, thus affecting the consistency of intraoperative decisions and the objectivity of surgical outcome assessment. Furthermore, the lack of a quantitative traceability mechanism for vanishing point determination also limits the reliability of intraoperative review and postoperative retrospective analysis.

[0009] In summary, existing technologies have key shortcomings in the intraoperative neuromonitoring scenarios of hemifacial spasm MVD surgery, particularly in terms of individualized adaptive recognition of abnormal waves, interpretation of the entire context sequence, and quantitative determination of the vanishing point of abnormal waves. This invention provides a method for intraoperative lateral diffusion waveform recognition and vanishing point determination based on template state decoding and MVMD enhancement. Summary of the Invention

[0010] The purpose of this invention is to provide a method for identifying intraoperative lateral diffusion waveforms and determining vanishing points based on template state decoding and MVMD enhancement, thereby solving the above-mentioned problems.

[0011] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0012] The first aspect of this invention provides a method for intraoperative lateral diffusion waveform identification and vanishing point determination based on template state decoding and MVMD enhancement, comprising the following steps:

[0013] Acquire intraoperative nerve monitoring waveform stack data, perform preprocessing, and obtain preprocessed waveform data;

[0014] Select intraoperative neuromonitoring waveform stack data, preprocessed waveform data, or a fusion of the two signals as the template analysis signal source;

[0015] Within a preset latency period, high-confidence anomalous candidate waveforms are screened from the template analysis signal source and one or more anomalous templates are constructed; at the same time, features of normal or low-response states are extracted from the early low-response waveforms of the template analysis signal source, and a baseline reference is constructed as a negative evidence model.

[0016] Latency constraint matching is performed on each waveform in the template analysis signal source to perform evidence scoring and calculate the comprehensive evidence score for each waveform.

[0017] Based on the waveform-by-wavelength comprehensive evidence score, the template analysis signal source is globally decoded to obtain the waveform-by-wavelength state sequence. The disappearance point of the abnormal wave is determined according to the number of consecutive normal states at the end and a preset threshold.

[0018] A second aspect of the present invention also provides an apparatus / device / system for intraoperative lateral diffusion waveform identification and vanishing point determination based on template state decoding and MVMD enhancement, comprising a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above method.

[0019] A third aspect of the present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0020] A fourth aspect of the present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0021] In summary, the present invention has the following beneficial effects:

[0022] Personalized Adaptive Recognition: This invention constructs anomaly templates within the same file and performs adaptive recognition based on the waveform characteristics of the current case. It can automatically adapt to differences in waveform morphology between different patients, differences in noise levels under different acquisition conditions, and the temporal evolution of waveform characteristics during the same surgical procedure. Compared to fixed threshold or universal template methods, it significantly improves the individualized adaptability and cross-case stability of abnormal waveform recognition.

[0023] Joint determination of positive and negative evidence: This invention introduces a baseline reference negative evidence model, which incorporates "abnormal template similarity" (positive evidence) and "normal / low response similarity" (negative evidence) into the determination process at the same time, forming a more stable distinction mechanism between abnormal and normal waves, and effectively suppressing the false positive problem of noise waveforms or low-amplitude normal fluctuations being misjudged as abnormal waves.

[0024] Multidimensional evidence fusion under latency constraints: This invention uses a latency constraint matching mechanism to comprehensively evaluate and weight and fuse multidimensional evidence such as waveform morphology correlation, time position accuracy, local signal quality, and area correlation features. This overcomes the shortcomings of traditional single threshold methods in terms of insufficient information utilization. While taking into account waveform morphology features, it effectively utilizes their time position information for constraints, thereby improving the robustness and accuracy of abnormal wave identification.

[0025] Sequence Context State Decoding: This invention decodes the state of waveform-by-wave evidence sequences and utilizes the contextual relationships and state transition continuity of stacked waveforms to obtain the globally optimal state sequence through dynamic programming or equivalent optimization methods. This effectively suppresses isolated state jumps caused by occasional noise or transient interference in a single waveform, and improves the consistency and clinical reliability of waveform-by-wave judgment results.

[0026] Quantifiable vanishing point determination: This invention establishes a vanishing point determination mechanism through the rule of continuous normal states at the end, transforming the judgment of whether the abnormal wave has stably disappeared from subjective experience into a quantitative judgment process with clear mathematical constraints. It can output traceable results such as the vanishing point position, the number of continuous normal states at the end, and whether the judgment is valid. This reduces the uncertainty of judgment caused by micro-fluctuations at the end or individual residual abnormalities in manual observation, and improves the consistency of intraoperative review and the repeatability of postoperative review.

[0027] The entire process is interpretable and verifiable: This invention exports waveform-by-wave evidence indicators, template metadata, baseline reference information and vanishing point summaries, and provides an interactive verification interface. It supports recalculation and result comparison after parameter adjustment, which is beneficial for result interpretation, parameter tracking, manual confirmation of preoperative and intraoperative considerations in clinical applications, and methodological review in postoperative retrospective analysis.

[0028] No external dependencies and wide applicability: This invention only requires conventional waveform stack data as input, without the need for additional equipment, external databases or pre-defined general templates. It is easy to deploy and applicable to various neurosurgical scenarios that require continuous waveform interpretation for intraoperative nerve monitoring, including but not limited to intraoperative AMR / LSR waveform interpretation for hemifacial spasm MVD, intraoperative monitoring for trigeminal neuralgia MVD, and other intraoperative electrophysiological monitoring scenarios. Attached Figure Description

[0029] Figure 1 This is a flowchart of a method for intraoperative lateral diffusion waveform identification and vanishing point determination based on template state decoding and MVMD enhancement in an embodiment of the present invention.

[0030] Figure 2 This is a schematic diagram of the processing result of a single patient waveform stack in Embodiment 3 of the present invention. The left side is the imported original waveform stack, the right side is the waveform stack after MVMD enhancement, and the bands marked with different colors are the abnormal waves identified by the template algorithm and their respective templates. Abnormal waves disappear at the end of the sequence.

[0031] Figure 3 This is a confusion matrix diagram of the batch processing results in Embodiment 4 of the present invention. The left diagram is an abnormal wave determination confusion matrix based on each waveform, and the right diagram is a surgical outcome determination confusion matrix based on whether the abnormal wave in the final waveform stack disappears.

[0032] Figure 4 This is a comparison chart of the macro average and micro average evaluation indices of the abnormal wave determination results in Embodiment 4 of the present invention;

[0033] Figure 5 This is a comparison chart of the macro average and micro average evaluation indicators of the surgical outcome determination results in Embodiment 4 of the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] In the embodiments provided in this application, it should be understood that the disclosed methods and systems can be implemented in other ways. The system embodiments described below are merely illustrative. For example, the division of units and modules is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or modules can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, and can be electrical, mechanical, or other forms.

[0036] In addition, each functional unit in the various embodiments of this application can be integrated into a single processor, or each unit can be a separate device, or two or more units can be integrated into a single device; each functional unit in the various embodiments of this application can be implemented in hardware or in the form of hardware plus software functional units.

[0037] Those skilled in the art will understand that all or part of the steps of the following method embodiments can be implemented by program instructions and related hardware. The aforementioned program instructions can be stored in a computer-readable storage medium. When the program instructions are executed, they perform the steps of the following method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), magnetic disks, or optical disks.

[0038] It should be understood that the use of terms such as "system," "device," "unit," and / or "module" in this application is merely one method of distinguishing different components, elements, parts, sections, or assemblies at different levels. However, if other terms can achieve the same purpose, they may be replaced by other expressions.

[0039] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "a plurality of" or "several" means two or more, unless otherwise explicitly specified.

[0040] If a flowchart is used in this application, it is used to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, the steps can be processed in reverse order or simultaneously. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0041] It should also be noted that, in this document, terms such as “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the article or device that includes the aforementioned element.

[0042] Example 1:

[0043] A method for intraoperative lateral diffusion waveform identification and vanishing point determination based on template state decoding and MVMD enhancement, such as... Figure 1 As shown, it includes the following steps:

[0044] S1. Acquire intraoperative neuromonitoring waveform stack data, perform preprocessing, and obtain preprocessed waveform data;

[0045] S2. Select intraoperative neuromonitoring waveform stack data, preprocessed waveform data, or a fusion of the two signals as the template analysis signal source;

[0046] S3. Within a preset latency period, select high-confidence anomalous candidate waveforms from the template analysis signal source and construct one or more anomalous templates; at the same time, extract features of normal or low-response states from the early low-response waveforms of the template analysis signal source and construct a baseline reference as a negative evidence model.

[0047] S4. Perform latency constraint matching on each waveform in the template analysis signal source, perform evidence scoring, and calculate the comprehensive evidence score for each waveform.

[0048] S5. Based on the waveform-by-wavelength comprehensive evidence score, perform global state decoding on the template analysis signal source to obtain the waveform-by-wavelength state sequence, and determine the disappearance point of the abnormal wave according to the number of consecutive normal states at the end and the preset threshold.

[0049] In step S1, the intraoperative neuromonitoring waveform stack data is acquired. The waveform data is preferably stored in comma-separated file (CSV) format. Each file includes several waveform data entries arranged in chronological or stimulus order, with each waveform corresponding to different time points, different stimulus cycles, or different acquisition sequences. The system can read a single file or read multiple files in batches according to a directory structure.

[0050] During the reading process, the system identifies waveform columns according to preset data column rules. If a waveform data column prefix is ​​set, columns matching the prefix are selected first; if no prefix is ​​set, the waveform columns can be automatically identified based on the numerical characteristics of the column content. Subsequently, the identified waveform columns are organized into a two-dimensional waveform matrix. ,in Indicates the waveform row number (total) (waveform) This indicates the number of sampling points for each waveform on the time axis.

[0051] This step converts the original comma-separated data into a standardized waveform stack structure, providing a unified input for subsequent template construction, waveform-by-wave scoring, and vanishing point determination.

[0052] The preprocessing includes time window trimming and sampling parameter determination.

[0053] After constructing the waveform matrix, the sampling parameters and analysis time range are further determined. Specifically, the sampling rate can be derived based on the total waveform duration, the total number of sampling points, or the time range information corresponding to each column. :

[0054] ;

[0055] in The total number of sampling points. This represents the total duration of the waveform. If a start time is specified beforehand... and termination time The system then performs a cropping operation on the waveform matrix based on the time range, retaining only the target analysis interval. Sampling points within.

[0056] This step preserves not only the effective waveform matrix after cropping, but also information such as the total duration of the original waveform, the requested cropping range, the actual loading range, and the corresponding sampling point index. This information can be used to restore the correspondence between the analysis interval and the original data during subsequent result export and interactive verification, ensuring that the analysis results can be traced back to the specific location of the original data.

[0057] The preprocessing also includes MVMD enhancement.

[0058] To improve the ability to distinguish abnormal response structures under noisy conditions, this invention performs multivariable variational mode decomposition (MVMD) on the original waveform matrix. MVMD is an algorithm that can simultaneously decompose multi-channel signals into multiple band-limited mode components. By applying bandwidth constraints and modal orthogonality constraints, structural components with different center frequencies and bandwidths can be separated from the original waveform.

[0059] Specifically, the system is based on a preset number of modes. Penalty parameters (Controlling the tightness of modal bandwidth), starting mode index Parameters such as whether to invert the modes and whether to discard leading sample points are used to perform joint decomposition on the entire waveform matrix to obtain... Let the modal component be . The first waveform Each modal component is The enhanced waveform is composed of modal components within the target modal range. ;

[0060] in, This is the initial combined mode index. The total number of modes. The enhanced waveform can highlight the abnormal response structure (such as the characteristic components of the AMR waveform) to a certain extent, and suppress some background fluctuations, high-frequency noise, or signal components unrelated to anomaly interpretation. The enhanced waveform can be used directly as the analysis signal source for subsequent template construction and waveform-by-waveform matching, or it can participate in the fusion evidence calculation together with the original waveform.

[0061] If the original waveform is of high quality and has a sufficient signal-to-noise ratio, the enhancement process can be skipped. In this case, this step can be omitted, and subsequent steps can be performed based solely on the original waveform.

[0062] In step S2, after preparing the original and enhanced waveforms, the system determines the analysis signal source for anomaly template construction and waveform-by-waveform matching. In a preferred embodiment, the system uses both the original waveform and the MVMD enhanced waveform simultaneously, forming fusion evidence in the subsequent matching stage. This approach balances the intuitive interpretability of the original waveform with the structural resolution of the enhanced waveform.

[0063] First, the original waveform and the enhanced waveform are standardized row by row. For the first row... The first waveform There are 1 sampling point, and the original value is set to 1. The mean of this waveform is The standard deviation is Then, row-based standardization can be expressed as:

[0064] ;

[0065] Among them, when At that time, it can be made To prevent division by zero. Let the original waveform after standardization be... The standardized enhanced waveform is Then the fused signal can be expressed as:

[0066] ;

[0067] Alternatively, depending on the configuration, you can select the standardized original waveform, the standardized enhanced waveform, or the above-mentioned fused signal as the analysis signal source for subsequent template construction and waveform-by-waveform matching.

[0068] In step S3, the template analysis signal source is used to filter high-confidence anomaly candidate waveforms, which includes the following steps:

[0069] Candidate responses are searched for for each waveform within a preset latency period. The latency period can be defined by the start latency period and the end latency period to constrain the time position of the response search, avoid erroneous extraction of candidate responses in time periods unrelated to clinical interest intervals, and thus improve the targeting of template construction.

[0070] For each candidate response, the local signal-to-noise ratio, area correlation index, and other statistics used to characterize the candidate quality are calculated comprehensively, and the candidate responses are sorted based on their scores to select high-confidence anomalous candidate waveforms.

[0071] Specifically, let the baseline segment of a waveform before the response start point be... Then the absolute deviation of the baseline median (MAD) is:

[0072] ;

[0073] Baseline Scale Defined as the maximum value of MAD, standard deviation, and numerical lower limit constraint:

[0074] ;

[0075] in, The scaling factor for estimating the standard deviation from MAD (for normal distribution). The standard deviation of the baseline segment. To prevent excessively small lower bound values, the following criteria are set for the candidate fragments to be analyzed. Its local signal-to-noise ratio is defined as the ratio of peak amplitude to baseline scale:

[0076] ;

[0077] The absolute area index of a candidate fragment is defined as the sum of the absolute values ​​of the sampled points divided by the sampling rate:

[0078] ;

[0079] After obtaining the aforementioned quality indicators for each candidate, the system can select candidates according to their ranking ratio (e.g., the top candidates). The set of high-confidence abnormal candidate waveforms is selected by considering factors such as the number of candidates, the minimum number of candidates, the minimum number of template waveforms, and the maximum allowable interval between candidates.

[0080] In this way, the present invention does not rely on pre-given external general templates or historical databases, but uses high-confidence abnormal waveforms within the same file as template construction sources, thereby achieving truly individualized adaptive template construction and improving the adaptability to the waveform characteristics of the current case.

[0081] After obtaining a set of high-confidence anomaly candidate waveforms, the system constructs one or more anomaly templates based on these candidate waveforms. During template construction, a stable template waveform representation can be formed based on the peak position, duration, waveform polarity, and morphological consistency of the candidate waveforms.

[0082] Let the set of selected template contribution fragments be... ,in To contribute to the fragment index set, the template waveform can be given by the median of each sample point, thus enhancing the robustness of the template to outlier candidate fragments:

[0083] ;

[0084] Let the peak position of each contributing segment be . Then the template latency position can be represented as the median of the peak positions of each contributing segment:

[0085] ;

[0086] In some embodiments, the system generates only one master anomaly template for subsequent waveform-by-waveform matching, which is suitable for scenarios where the anomaly waveforms are relatively uniform. In other embodiments, the system can also generate multiple anomaly templates based on cluster analysis of candidate segments, and assign different template identifiers and sorting levels to each template, thereby adapting to scenarios where the anomaly patterns are diverse.

[0087] For each generated anomalous template, the system records template metadata. This metadata preferably includes: template identifier, template source (original waveform / enhanced waveform / fused signal), template confidence level, indicators of template reliability, template start and end sampling points, template latency (expressed in sampling points or milliseconds), template duration, template polarity (positive / negative / bidirectional), list of contributing waveform row numbers, template clustering score, and template ranking in the case of multiple templates. This metadata provides a complete basis for the subsequent interpretation, verification, and traceability of results.

[0088] When the system detects that the number of high-confidence abnormal candidate waveforms is insufficient (e.g., below the minimum template waveform number threshold), or the morphological consistency between candidate waveforms is insufficient to support the formation of a stable template, the system can record the specific reasons for the insufficient template construction (e.g., insufficient number of candidates, excessive morphological dispersion between candidates, latency exceeding the preset range, etc.), and enter backup processing logic when allowed (e.g., relax candidate screening conditions, fall back to simple amplitude judgment mode, or mark as requiring manual judgment), so as to avoid outputting unstable or unreliable judgment results when the template quality is obviously insufficient.

[0089] To improve the stability of distinguishing between anomalous and normal waveforms, this invention further constructs a baseline reference as a negative evidence model. The core idea is to extract characteristic patterns of normal or low-response states from early waveforms in the same file to form a "non-anomalous" reference representation, which then plays a role in suppressing false alarms in subsequent waveform-by-waveform scoring.

[0090] Specifically, the system prioritizes the beginning of the waveform sequence (e.g., the front part). The early waveform subset was selected, and waveforms with a local signal-to-noise ratio lower than a preset low response threshold were selected as reference candidates. The selection strategy is based on the following clinical assumption: in the early stage of MVD surgery, the muscle relaxation effect has not completely disappeared, and the early waveforms include low response waveforms from the preoperative baseline acquisition segment, which can be used as a reference for the normal / low response state.

[0091] When the number of candidate reference waveforms reaches a preset minimum requirement, the system forms a baseline reference based on the candidate waveforms. Let the candidate reference segment be... ,in For the reference candidate index set, the baseline reference can be represented as the median of the candidate segments at each sample point:

[0092] ;

[0093] For the local signal-to-noise ratio set of the selected reference candidates Its confidence level can be optimally quantified as follows:

[0094] ;

[0095] in, The preset maximum permissible signal-to-noise ratio threshold for baseline reference. This indicates that the value will be truncated to a range. A higher confidence level indicates a more reliable representation of the low response / normal state. The system also records statistical information such as the source of the baseline reference (based on the original waveform / enhanced waveform), confidence level, list of contributing waveform row numbers, and response intensity (median local signal-to-noise ratio).

[0096] When there are insufficient reference candidate waveforms (e.g., all early waveforms have obvious abnormal responses), the system can mark the baseline reference as insufficient or low confidence, but still allow subsequent steps to continue scoring and interpretation based solely on positive evidence from the abnormal template, thereby ensuring that the overall process is not interrupted due to the failure of baseline reference construction.

[0097] In step S4, the latency constraint matching involves searching for the waveform segment position most similar to the anomaly template within a preset allowable offset range to obtain the best matching result.

[0098] Let the segment to be matched be The exception template is Both have the same length (the number of sampling points corresponding to the duration of the template) and their mean values ​​are respectively and The normalized correlation is defined as the Pearson correlation coefficient:

[0099] ;

[0100] When the denominator is less than or equal to zero, the correlation is denoted as This constraint matching mechanism can take into account the actual situation that there is slight drift in the waveform latency during the operation (such as due to factors such as fine adjustment of the stimulation electrode position and changes in tissue impedance), while avoiding the risk of mismatch caused by unconstrained full-range search (such as mismatching irrelevant noise fluctuations or distal artifacts as abnormal templates).

[0101] For each waveform to be judged, the system calculates one or more of the following evidence indicators at the optimal matching position:

[0102] (1) Template correlation evidence: truncating normalized correlation to intervals The relevance score is obtained:

[0103] ;

[0104] (2) Evidence of local signal-to-noise ratio: Normalize the local signal-to-noise ratio to the interval Use reference values Scaling:

[0105] ;

[0106] (3) Evidence of latency bias: Let the latency of the best match be the deviation from the latency of the template. The maximum allowable offset is Then the incubation period score is:

[0107] ;

[0108] The score is in perfect match ( (When the score is full) It decreases linearly as the deviation increases, and becomes [a certain value] after exceeding the allowable range. .

[0109] (4) Morphological score It can be calculated based on the similarity between the shape features of the waveform segment (such as the ratio of peak to valley, the ratio of rise time to fall time, half width at half maximum, etc.) and the corresponding features of the template.

[0110] (5) Evidence from the area under the absolute curve (AUC): Let the median of the absolute area of ​​the template contribution waveform set be... The absolute deviation of the median of the absolute area is The minimum scaling parameter is (To prevent the scale term from being too small), the area scale term is defined as follows:

[0111] ;

[0112] The absolute area of ​​the segment to be analyzed The AUC similarity is defined as:

[0113]

[0114] The similarity is equal Take the maximum value at time It decreases smoothly as the deviation increases.

[0115] Based on the above indicators, the system first constructs a basic evidence score. The relevance score is then used as a weighted combination of the dominant evidence score, signal-to-noise ratio score, and latency score.

[0116]

[0117] in, The score is for evidence of template relevance. The score represents the evidence for the local signal-to-noise ratio. The score for evidence of latency bias. , , The weights are non-negative and satisfy the following conditions: and order The larger value is chosen to reflect the dominance of relevant evidence; let the AUC evidence weight be... The comprehensive evidence score for the abnormal template is defined as a weighted fusion of the basic evidence score and the AUC similarity:

[0118] ;

[0119] Among them, when If AUC information is unavailable, it can degenerate into .

[0120] (6) Baseline reference negative evidence: Calculate the non-negative correlation between the segment to be analyzed and the baseline reference. and low energy score The matching score is then defined as:

[0121] ;

[0122] This score comprehensively reflects the degree to which the analyzed fragment is morphologically close to a normal low-response mode and the degree to which its own energy / response intensity is low.

[0123] Finally, by combining positive evidence (abnormal template score) and negative evidence (clean matching score), the final evidence score after introducing baseline reference penalty is obtained. Let the penalty weight be... ,but:

[0124] ;

[0125] in, To calculate the evidence score for each waveform, The score is a negative evidence matching score. The score is the comprehensive evidence score for the abnormal template.

[0126] When the waveform to be analyzed has both high anomaly template matching degree and high baseline similarity (this contradictory situation may occur in noisy waveforms), the baseline reference penalty will effectively reduce its final score, thereby suppressing false alarms.

[0127] In step S5, the global state decoding adopts dynamic programming decoding.

[0128] After obtaining the waveform-by-waveform comprehensive evidence score for each waveform, this invention does not directly make independent threshold judgments on each waveform. Instead, it uses the comprehensive evidence score sequence of the entire waveform stack as input to perform global state decoding. The goal of the state decoding is to divide the waveform sequence into two states: "abnormal" and "normal," and to suppress unreasonable frequent jumps by introducing state transition rules. This step is based on the understanding that the evolution of intraoperative AMR waveforms usually follows a macroscopic pattern of "persistent abnormality → transition zone → persistent normality," and that evidence fluctuations caused by instantaneous noise in a single waveform should not directly lead to state reversal.

[0129] A sequence decoding scheme based on dynamic programming is adopted. Let the... The final evidence score for the waveform is The anomaly detection threshold is (For example Then, its transmission costs for abnormal and normal states are defined as follows:

[0130] ;

[0131] ;

[0132] The intuitive meaning of launch cost is: when a high-resolution waveform of evidence is marked as normal, it will produce a "high-resolution deviation". "A proportional cost; similarly, when a piece of evidence with a low score is marked as abnormal, a corresponding cost will also be incurred." Used to adjust the relative weights between launch cost and transfer cost.

[0133] set up Indicates the first Waveform decoding to state The minimum cumulative cost, Indicates from state Transition to state The transition penalty is then applied, and the dynamic programming recursive formula for state decoding is:

[0134] ;

[0135] in, and All are taken from the state set The transfer penalty can be configured as follows: when The penalty is when the state remains unchanged. ;when The penalty during (state transition) is a positive value (e.g.) or Furthermore, different switching costs can be set for "abnormal → normal" and "normal → abnormal" (for example, the latter can be set higher to reflect the clinical prior that "once an abnormality disappears, it should not easily reappear").

[0136] Through the dynamic programming decoding described above, the final output state sequence represents the optimal path in the sense of minimizing the global cumulative cost, effectively utilizing the context of the entire waveform stack for smoothing. After state decoding, the system outputs a waveform-by-waveform state sequence, with each waveform corresponding to a decoded state, which can be further converted into binary labels (e.g., ...). Indicates abnormal. (Indicates normal) output.

[0137] The step of determining the disappearance point of the abnormal wave based on the number of consecutive normal states at the end and a preset threshold includes the following steps:

[0138] First, check if any abnormal state has occurred in the entire state sequence. If there is no abnormal state in the entire sequence, the vanishing point determination is usually not performed, and "vanishing premise not met" or an equivalent flag is directly output.

[0139] If there is at least one abnormal state in the state sequence, the system continues to count the number of consecutive normal states at the end of the sequence. Let the state sequence be... ( (where the total number of waveforms is 1), then the number of consecutive normal states at the end is defined as the maximum consecutive normal length counted backwards from the end:

[0140] ;

[0141] Let the preset threshold for consecutive normal values ​​at the end be... If both of the following conditions are met simultaneously: first, there is at least one previous anomalous state in the sequence; second... If the abnormal wave disappears, the system determines that the condition for the disappearance of the abnormal wave is met. The system records the position of the first normal state that appears after the last abnormal state as the disappearance point, and outputs the position of the disappearance point, the number of consecutive normal states at the end, and whether the disappearance condition is met.

[0142] If the number of consecutive normal states at the end does not reach the required level Even if a normal state exists in the sequence, the system determines that the disappearance condition has not been met, indicating that the abnormal wave may not have disappeared stably or still needs to be observed.

[0143] This step transforms the judgment of whether the abnormal wave has "stablely disappeared" from traditional subjective experience into an automated quantitative judgment process with clear mathematical constraints. The position of the disappearance point output can serve as an objective reference indicator for intraoperative decision-making, which helps to reduce the differences in the judgment of the disappearance time point among different operators and improves the consistency of intraoperative review and postoperative review.

[0144] In this embodiment, step S6 is also included: outputting waveform status, template metadata, baseline reference information and vanishing point determination results, and providing an interactive review interface.

[0145] After completing the abnormal wave identification and vanishing point determination, the system exports complete waveform-by-waveform interpretation results. The exported results preferably include the following information categories:

[0146] (1) Waveform interpretation result table, including waveform row number, decoding status label (abnormal / normal), template identifier, best matching latency, latency deviation, correlation, local signal-to-noise ratio, morphological score, absolute area, AUC similarity, clean score and comprehensive evidence score, etc.

[0147] (2) Abnormal template summary, including template identifier, template source, template confidence level, template latency, template duration, template polarity, list of contributing waveform row numbers and template metadata;

[0148] (3) Baseline reference summary, including baseline reference confidence, list of contributing waveform row numbers, median reference response intensity and sufficiency indicator;

[0149] (4) Summary of vanishing point determination, including the vanishing point location, the number of consecutive normal states at the end, whether the vanishing condition is met, the threshold for consecutive normal states at the end, and whether there are previous abnormal states, etc.

[0150] The system also provides an interactive review interface, which can simultaneously display the original waveform stack (arranged in rows, with abnormal rows highlighted or marked with different colors), enhanced waveforms (MVMD results or fused signals), waveform-by-waveform status label bars (displaying the state sequence with color blocks or labels), evidence score curves (showing the changing trend of the comprehensive evidence score of each waveform), as well as the currently used template status parameters (such as latency range, maximum allowable offset, anomaly judgment threshold, state transition penalty, end continuous normal threshold, etc.) and MVMD parameters (such as modality number, penalty parameter, starting modality index, etc.).

[0151] Users can adjust relevant parameters in the interface and observe the impact of parameter changes on the abnormal wave identification results and vanishing point determination results by triggering the "Recalculate" button or equivalent operations. For example, adjusting the threshold for consecutive normal waves at the end. The stringency of the vanishing point determination can be changed; the anomaly detection threshold can be adjusted. The sensitivity of abnormal / normal classification can be adjusted. Through this interactive review mechanism, the present invention can not only provide automated judgment conclusions, but also retain complete judgment criteria and parameter tracking paths, supporting clinicians to manually review the results and conduct scenario analysis during and after surgery, thereby improving the interpretability, clinical credibility, and flexibility of the overall solution.

[0152] Example 2:

[0153] The present invention also provides a device / equipment / system for identifying abnormal waves and determining vanishing points in intraoperative nerve monitoring based on template state decoding and MVMD enhancement, comprising a memory, a processor, and a computer program stored in the memory, characterized in that the processor executes the computer program to implement the steps of the above method.

[0154] The present invention also provides a computer-readable storage medium having a computer program / instructions stored thereon, which, when executed by a processor, implement the steps of the above-described method.

[0155] The present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the above-described method.

[0156] Example 3:

[0157] This embodiment processes the intraoperative neuromonitoring waveform stack of a single hemifacial spasm MVD patient to illustrate the processing procedure and output results of the present invention for a single case.

[0158] like Figure 2 As shown, the processing results of the patient's waveform stack are as follows: the left side is the imported original waveform stack, and the right side is the waveform stack enhanced by MVMD; the bands marked with different colors in the figure are the abnormal waves identified by the template algorithm of this invention. A total of 4 abnormal templates (T1 to T4) were constructed, and each template is distinguished by a different color; it can be seen that the abnormal waves are concentrated in the middle of the sequence and gradually weaken until they disappear stably at the end of the sequence, which is consistent with the evolution pattern of AMR waveform after successful decompression in clinical practice.

[0159] Table 1 shows the specific parameters obtained by the algorithm of this invention after processing each waveform of the patient, including waveform index, manual label, predicted label, template ID, template score, abnormal score, correlation, template SNR, best matching latency (ms), latency deviation (ms), absolute area (Abs AUC), and AUC similarity.

[0160] As shown in Table 1, in waveforms 1-51, both the algorithm-predicted label and the manually labeled label were 0 (normal), and their overall abnormal score was generally low (mostly in the range of 0). Between; in waveforms 52 and 86, both the algorithm-predicted label and the manually labeled label were 1 (anomaly), and their comprehensive evidence scores increased significantly (mostly in The correlation index increased significantly between the 87th and 88th waveforms; from the 87th waveform onwards, the label dropped back to 0 (normal) and continued until the end of the sequence. In this embodiment, the algorithm-predicted label was completely consistent with the manually labeled label, verifying the accuracy of the present invention in interpreting waveforms in a single case, and the continuous normal state at the end met the preset threshold, thus confirming the number of successful disappearance point determinations.

[0161] Table 1. Specific parameters obtained by the algorithm of this invention after processing each waveform of the patient.

[0162]

[0163]

[0164] Example 4:

[0165] This embodiment performs batch processing on 186 waveform stacks, totaling 9515 waveforms, from 93 patients with hemifacial spasm MVD, to evaluate the overall performance of the method of the present invention from both waveform-level and surgical outcome-level perspectives.

[0166] like Figure 3 As shown, the batch processing results have two types of confusion matrices: the left figure is the abnormal waveform discrimination matrix for each waveform, reflecting the consistency between the algorithm's predicted labels and manually labeled labels at the waveform level (abnormal / normal binary classification); the right figure is the surgical outcome discrimination matrix based on whether the abnormal waveform disappears at the end of the waveform stack, reflecting the algorithm's predictive ability for surgical outcome at the stack level (disappearance true / false). Each cell in the confusion matrix corresponds to the count of true positives, false positives, true negatives, and false negatives, respectively.

[0167] like Figure 4 As shown, the macro and micro average evaluation metrics of the anomaly wave determination (waveform level) results are compared. These evaluation metrics may include accuracy, precision, recall, and F1 score. The macro average is the arithmetic mean of each category metric, reflecting the overall performance of the method under category balance; the micro average is a weighted statistical calculation based on the total sample size, reflecting the overall performance of the method under the overall sample size.

[0168] like Figure 5 As shown, a comparison of the macro and micro average evaluation indicators for surgical outcome determination (stack level) results is presented, along with the indicator definitions and... Figure 3 The difference lies in that the statistical unit is a waveform stack (i.e., the final disappearance determination result of each surgery), which is used to evaluate the overall performance of this invention at the level of surgical endpoint determination.

[0169] The above batch experimental results show that the method of the present invention can achieve stable evaluation index performance at both the waveform-level abnormal wave identification and stack-level surgical outcome determination levels, verifying the robustness and applicability of the method based on template state decoding and MVMD enhancement under cross-case and cross-acquisition conditions.

[0170] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0171] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. 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 this application.

[0172] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0173] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0174] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying abnormal waves and determining their disappearance points during intraoperative nerve monitoring, characterized in that: Includes the following steps: Acquire intraoperative nerve monitoring waveform stack data, perform preprocessing, and obtain preprocessed waveform data; Select intraoperative neuromonitoring waveform stack data, preprocessed waveform data, or a fusion of the two signals as the template analysis signal source; Within a preset latency period, high-confidence candidate waveforms of anomalies are selected from the template analysis signal sources and one or more anomaly templates are constructed. Meanwhile, features of normal or low response states are extracted from the early low response waveforms of the template analysis signal source to construct a baseline reference as a negative evidence model. Latency constraint matching is performed on each waveform in the template analysis signal source to perform evidence scoring and calculate the comprehensive evidence score for each waveform. Based on the waveform-by-wavelength comprehensive evidence score, the template analysis signal source is globally decoded to obtain the waveform-by-wavelength state sequence. The disappearance point of the abnormal wave is determined according to the number of consecutive normal states at the end and a preset threshold.

2. The method according to claim 1, characterized in that: The preprocessing includes time window trimming and sampling parameter determination.

3. The method according to any one of claims 1-2, characterized in that: The preprocessing also includes MVMD enhancement.

4. The method according to claim 1, characterized in that: The process of filtering high-confidence anomaly candidate waveforms from the template analysis signal source includes the following steps: Search for candidate responses for each waveform within a preset latency period; For each candidate response, the local signal-to-noise ratio, area correlation index, and other statistics used to characterize the candidate quality are calculated comprehensively, and the candidate responses are sorted based on their scores to select high-confidence anomalous candidate waveforms.

5. The method according to claim 1, characterized in that: It also includes, For each abnormal template record metadata, the metadata includes: template identifier, template source, template confidence, indicator of whether the template is reliable enough, template start and end sampling points, template latency, template duration, template polarity, list of contributing waveform row numbers, template clustering score, and template ranking in the case of multiple templates.

6. The method according to claim 1, characterized in that: The latency constraint matching involves searching for the waveform segment position most similar to the anomaly template within a preset allowable offset range to obtain the best matching result.

7. The method according to claim 1, characterized in that: The method for calculating the waveform-by-wavelength comprehensive evidence score is as follows: in, To calculate the evidence score for each waveform, The score is a negative evidence matching score. The score is the comprehensive evidence score for the abnormal template.

8. The method according to claim 7, characterized in that: The method for calculating the comprehensive evidence score of the constant template is as follows: ； in, The comprehensive evidence score for the abnormal template, when If AUC information is unavailable, it can degenerate into , in, The score is for evidence of template relevance. The score represents the evidence for the local signal-to-noise ratio. The score for evidence of latency bias. , , The weights are non-negative and satisfy the following conditions: and order The larger value is chosen to reflect the dominance of relevant evidence; let the AUC evidence weight be... ; The negative evidence matching score is calculated as follows: ； in, , This provides evidence for the local signal-to-noise ratio.

9. The method according to claim 1, characterized in that: The calculation of the waveform-by-wavelength comprehensive evidence score includes the following steps: the global state decoding adopts dynamic programming decoding.

10. The method according to claim 1, characterized in that: The step of determining the disappearance point of the abnormal wave based on the number of consecutive normal states at the end and a preset threshold includes the following steps: First, check if any abnormal state has occurred in the entire state sequence. If there is no abnormal state in the entire sequence, the vanishing point determination is usually not performed, and "vanishing premise not met" or an equivalent flag is directly output. If there is at least one abnormal state in the state sequence, the system continues to count the number of consecutive normal states at the end of the sequence. Let the state sequence be... The number of consecutive normal states at the end is defined as the maximum consecutive normal length counted backwards from the end: ； Let the preset threshold for consecutive normal values ​​at the end be... If both of the following conditions are met simultaneously: first, there is at least one previous anomalous state in the sequence; second... If the abnormal wave disappears, the system determines that the condition for the disappearance of the abnormal wave is met. The system records the position of the first normal state that appears after the last abnormal state as the disappearance point, and outputs the position of the disappearance point, the number of consecutive normal states at the end, and whether the disappearance condition is met. If the number of consecutive normal states at the end does not reach the required level Even if a normal state exists in the sequence, the system determines that the disappearance condition has not been met, indicating that the abnormal wave may not have disappeared stably or still needs to be observed.

11. A device / equipment / system for identifying abnormal waves and determining vanishing points in intraoperative neurological monitoring based on template state decoding and MVMD enhancement, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-10.

12. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-10.

13. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any of claims 1-10.

Images