Necrotizing fasciitis surgery timing aided decision method and system based on lesion expansion trend modeling

Abstract

The application relates to the technical field of medical auxiliary decision and medical information processing, and discloses a necrotizing fasciitis operation opportunity auxiliary decision method and system based on lesion expansion trend modeling, wherein the method comprises the following steps: acquiring asynchronous multi-source observation data; performing a first-visit multi-modal observation reconstruction task; performing a deep expansion proxy quantity construction task; performing an expansion trend preprocessing task; performing an operation trigger judgment task; performing hierarchical trigger control; and outputting an operation auxiliary decision result. Compared with the method for judging the operation opportunity of necrotizing fasciitis by relying on single-time-point images in the prior art, especially under the condition that the surface swelling range of a patient is not significant in the emergency first-visit stage, the technical problem that the continuous depiction of the deep expansion trend of a lesion cannot be realized. Since the application establishes an asynchronous multi-source observation unified reconstruction link, establishes time-weighted trend modeling and hierarchical trigger control, and realizes auxiliary evaluation of the operation opportunity of a necrotizing fasciitis lesion.

Description

Technical Field

[0001] This invention relates to the field of medical decision support and medical information processing technology, and in particular to a method and system for assisting decision-making on the timing of surgery for necrotizing fasciitis based on lesion expansion trend modeling. Background Technology

[0002] While the clinical management of necrotizing fasciitis is generally recognized as prioritizing early surgical exploration or debridement, the methods for determining "when to perform immediate surgery" remain significantly inadequate in actual emergency department settings. Currently, physicians typically rely on a single laboratory test result, single-time-point imaging findings, the extent of skin redness and swelling, the intensity of pain complaints, and clinical experience for a comprehensive assessment. This approach can provide some guidance in typical cases, but for early, insidiously progressing cases, especially when superficial skin manifestations are not yet prominent while deep fascial infection has rapidly spread along the interstitial spaces, it is often difficult to identify the true extent of disease progression in a timely and accurate manner.

[0003] For example, in the initial emergency room visit, data from different modalities often exhibit significant asynchrony and incompleteness: vital signs can be acquired continuously, laboratory indicators are typically sampled intermittently, and while bedside ultrasound can be completed relatively quickly, CT scans and their reports often involve time delays. Simultaneously, some patients at the initial visit may only present with mild localized redness, swelling, or pain, without exhibiting typical signs such as extensive necrosis or subcutaneous pneumothorax. In such cases, if static scoring models, single-modal assessment methods, or fixed-time-rhythm review methods are still used, two types of problems are likely to occur: one is underestimating the rate of deep disease progression, leading to delays in surgical intervention; the other is over-reliance on a single abnormal result when evidence is insufficient, resulting in unnecessary high-risk triggering of judgments.

[0004] Therefore, there is an urgent need for an auxiliary decision-making method that can still jointly assess and grade the extent and trend of deep lesion expansion and the timing of surgical intervention even when multi-source asynchronous acquisition, delayed appearance of local surface signs, missing parts of key modalities, or differences in observation quality. This would improve the real-time performance, accuracy, and clinical reliability of determining the timing of surgery for necrotizing fasciitis. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose an auxiliary decision-making method for surgical timing in necrotizing fasciitis based on lesion expansion trend modeling. This method aims to solve the technical problem that existing technologies rely on single-time-point image judgment to determine the timing of surgery for necrotizing fasciitis, especially in the initial emergency stage when the patient's superficial redness and swelling are not yet significant, and cannot continuously depict the deep expansion trend of the lesion.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for assisting decision-making on the timing of surgery for necrotizing fasciitis based on lesion expansion trend modeling.

[0007] The method for assisting in decision-making regarding surgical timing in necrotizing fasciitis based on lesion expansion trend modeling includes:

[0008] Step S10: Obtain asynchronous multi-source observation data within the preset initial diagnosis observation window, and perform the initial diagnosis multimodal observation reconstruction task based on the asynchronous multi-source observation data using a unified time axis alignment and observation quality coding mechanism, and output the aligned observation feature tensor and the observation quality state set;

[0009] Step S20: Based on the aligned observation feature tensor and the observation quality state set, a modal confidence weighted fusion mechanism is used to perform the deep extended proxy quantity construction task, and output the fused extended level sequence, modal divergence quantity and global confidence.

[0010] Step S30: Based on the fusion of extended horizontal sequence, modal divergence and global confidence, the extended trend preprocessing task is performed using a time-weighted extended trend modeling mechanism to output the extended trend state vector;

[0011] Step S40: Based on the extended trend state vector, a dynamic threshold adaptive update mechanism is used to perform the surgical trigger determination task, and the surgical trigger risk value and dynamic determination threshold are output;

[0012] Step S50: Perform graded trigger control based on surgical trigger risk value and dynamic judgment threshold, and output auxiliary decision results for surgery.

[0013] Preferably, step S10, which involves acquiring asynchronous multi-source observation data within a preset initial observation window, performing initial multi-modal observation reconstruction based on the asynchronous multi-source observation data using a unified time axis alignment and observation quality coding mechanism, and outputting aligned observation feature tensors and observation quality state sets, specifically includes:

[0014] Step S101: Obtain the triage time Pre-set observation window for initial consultation Vital signs sequence within Laboratory indicator series Bedside ultrasound observation sequence Surface manifestation sequence ,in, This indicates the preset initial observation window length, and further acquisition of CT observation sequences. This forms the original asynchronous observation set;

[0015] Step S102: Construct a unified discrete-time grid based on the original asynchronous observation set And for any mode in a unified discrete-time grid The original observation sequence is time-aligned using a time-decaying kernel alignment method to generate modal alignment features corresponding to a unified discrete time grid;

[0016] Step S103: Construct the aligned observation feature tensor based on the modal alignment features using a modal feature splicing and fusion method. Based on modal alignment features, an observation quality state set is constructed using a joint coding method based on missing mask and observation time interval quantization. .

[0017] Preferably, in step S20, the step of constructing the deep extended proxy quantity based on the aligned observation feature tensor and the observation quality state set, using a modal confidence weighted fusion mechanism, and outputting the fused extended level sequence, modal divergence quantity, and global confidence score, specifically includes:

[0018] Step S201: Based on the aligned observation feature tensor Bedside ultrasound extended proxy scoring was constructed respectively. Laboratory extended agent rating Unstable vital signs assessment Physical appearance proxy score And for CT observation sequences Building CT Extended Agent Scoring ;

[0019] Step S202: Based on the observation quality state set A modal credibility weight construction mechanism is used to generate credibility weights for each effective mode. and utilize trusted weights The scores of each modality's extended agent are weighted and fused to obtain a fused extended level sequence. ;

[0020] Step S203: Based on the extended agent scores of each modality and the fusion extended level sequence Calculate the modal divergence based on the deviation between them. And based on the credibility weight The global confidence score is obtained by using a modal divergence exponential decay method to aggregate the credible weights. ;

[0021] Step S204: Final output: fused extended horizontal sequence Modal divergence and global confidence .

[0022] Preferably, in step S202, based on the observation quality state set A modal credibility weight construction mechanism is used to generate credibility weights for each effective mode. and utilize trusted weights The scores of each modality's extended agent are weighted and fused to obtain a fused extended level sequence. The steps are as follows: First, based on the modal quality score in the observation quality state set... Modal reliability weights are constructed according to the following formula. : ;in, To be compatible with the unified discrete-time grid The corresponding set of effective modes, For the effective mode set The index identifier of any valid mode in the middle, For the first The mass sensitivity coefficient corresponding to each effective mode For the first Each effective mode is in a unified discrete-time grid. Modal quality score at the location, For the first Each effective mode is in a unified discrete-time grid. Modal quality score at the location, For the first The mass sensitivity coefficient corresponding to each effective mode It is a natural exponential function;

[0023] Then add modal confidence weights Corresponding bedside ultrasound extended agent score Laboratory extended agent rating Unstable vital signs assessment Physical appearance proxy score and CT extended agent score Multiply and sum to obtain the fused extended level sequence. .

[0024] Preferably, step S30, which involves performing the extended trend preprocessing task based on the fused extended level sequence, modal divergence, and global confidence using a time-weighted extended trend modeling mechanism, and outputting the extended trend state vector, specifically includes:

[0025] Step S301: Construct time-weighted coefficients based on the fused extended level sequence and global confidence using an exponential time decay weighting method. The expansion rate was constructed using time-weighted local regression based on the fused extended level sequence and modal divergence. ;

[0026] Step S302: Based on the spread rate of adjacent time moments Calculate the difference spread acceleration between adjacent time steps. Simultaneously acquire the body surface manifestation sequence. Based on body surface manifestation sequence The surface spread rate was calculated using the difference in redness and swelling area between adjacent time points. ; and according to the expansion rate Extended acceleration and surface spread rate Constructing hidden representation deviation using a differential coupling method between deep extension and surface extension ;

[0027] Step S303: Obtain information on the area of ​​redness and swelling on the body surface, and combine it with the deviation of the hidden characteristics. Expansion rate Extended acceleration and surface spread rate Combined into an extended trend state vector .

[0028] Preferably, step S40, which involves performing the surgical trigger determination task based on the extended trend state vector using a dynamic threshold adaptive update mechanism, and outputting the surgical trigger risk value and the dynamic determination threshold, specifically includes:

[0029] Step S401: Construct a surgical trigger risk value based on the extended trend state vector using a short-time-window surgical demand risk calculation method. Surgical trigger risk value This value is used to characterize the risk level of a patient needing surgical exploration or debridement within a pre-defined short time window at the current moment; surgical trigger risk value. Calculate according to the following formula: ;

[0030] in, For the Sigmoid function, To be compatible with the unified discrete-time grid Corresponding fusion extended level sequence The value at the current moment, For expansion rate, To extend acceleration, To conceal the deviation in representation, This is the normalized value for lactic acid. This is the inversely normalized value of mean arterial pressure. The bias coefficient, to This represents the corresponding risk contribution coefficient; among which, the surgical trigger risk value... During the calculation process, dimensionless normalization is performed on each input variable;

[0031] Step S402: Construct a dynamic judgment threshold using a dynamic threshold adaptive update method. Dynamically determine threshold Update according to the following formula:

[0032] ;

[0033] in, This represents the next moment in the unified discrete-time grid. Represents the current time in a unified discrete-time grid. The corresponding dynamic judgment threshold, Represents the next time step in a unified discrete-time grid. The corresponding updated dynamic threshold is determined. This represents a threshold truncation function used to restrict the updated dynamic threshold to a preset minimum boundary. With the preset maximum boundary between; This is the expansion rate adjustment coefficient. To integrate the extended level adjustment coefficient, This is the modal bifurcation adjustment coefficient. This is the global confidence level compensation adjustment coefficient; For modal divergence, For global confidence;

[0034] Step S403: Set the surgical trigger risk value With dynamic judgment threshold When conducting a joint comparison, When the condition is met, a high-risk trigger result is output; when When the condition is not met, the system will determine that the surgical triggering conditions have not been met and output a non-high-risk triggering result, and finally output the surgical triggering risk value and dynamic judgment threshold.

[0035] Preferably, step S50, which involves performing graded trigger control based on the surgical trigger risk value and dynamic judgment threshold, and outputting the auxiliary decision-making result for the surgery, specifically includes:

[0036] Step S501: Obtain the modal divergence value and global confidence level, and perform hierarchical triggering judgment in combination with the surgical trigger risk value and dynamic judgment threshold; when the surgical trigger risk value is greater than or equal to the dynamic judgment threshold, and the global confidence level is not lower than the preset lower confidence limit and the modal divergence value is not higher than the preset upper divergence limit, output the immediate surgical triggering status.

[0037] Step S502: When the surgical trigger risk value is greater than or equal to the dynamic judgment threshold, and there is a situation where the global confidence is lower than the preset lower confidence limit or the modal divergence is higher than the preset upper divergence limit, output the high-risk parallel supplementary evidence status; when the surgical trigger risk value is lower than the dynamic judgment threshold but not lower than the preset intermediate risk threshold, output the short window re-evaluation status.

[0038] Step S503: Based on the current decision state after the execution of the hierarchical trigger judgment, the surgical trigger risk value, the dynamic judgment threshold, the modal divergence value, and the global confidence, generate and output the auxiliary decision result of the surgery.

[0039] This invention also provides a decision-making system for surgical timing in necrotizing fasciitis based on lesion expansion trend modeling, comprising:

[0040] The multi-source observation reconstruction module is used to acquire asynchronous multi-source observation data within the preset initial diagnosis observation window. Based on the asynchronous multi-source observation data, a unified time axis alignment and observation quality coding mechanism is used to perform the initial diagnosis multimodal observation reconstruction task, and output the aligned observation feature tensor and the observation quality state set.

[0041] The extended proxy fusion module is used to perform the deep extended proxy quantity construction task based on the aligned observation feature tensor and the observation quality state set, and adopts the modal confidence weighted fusion mechanism to output the fused extended level sequence, modal divergence quantity and global confidence.

[0042] The trend state generation module is used to perform extended trend preprocessing tasks based on the fused extended horizontal sequence, modal divergence, and global confidence, and to output the extended trend state vector by using a time-weighted extended trend modeling mechanism.

[0043] The risk threshold determination module is used to perform the surgical trigger determination task based on the extended trend state vector and the dynamic threshold adaptive update mechanism, and outputs the surgical trigger risk value and the dynamic determination threshold.

[0044] The decision output module is used to perform hierarchical trigger control based on surgical trigger risk values ​​and dynamic judgment thresholds, and output auxiliary decision results for surgery.

[0045] The present invention also provides a surgical timing assistance device for necrotizing fasciitis based on lesion expansion trend modeling, comprising: a memory, a processor, and a surgical timing assistance program for necrotizing fasciitis based on lesion expansion trend modeling stored in the memory and executable on the processor. When the surgical timing assistance program for necrotizing fasciitis based on lesion expansion trend modeling is executed by the processor, a surgical timing assistance method for necrotizing fasciitis based on lesion expansion trend modeling is implemented.

[0046] The present invention also provides a computer program product, including a decision-making aid program for surgical timing of necrotizing fasciitis based on lesion expansion trend modeling. When the decision-making aid program for surgical timing of necrotizing fasciitis based on lesion expansion trend modeling is executed by a processor, it implements the decision-making aid method for surgical timing of necrotizing fasciitis based on lesion expansion trend modeling.

[0047] The beneficial effects of this invention are as follows: By introducing a unified time axis alignment, observation quality coding, modal reliability weighted fusion, and time-weighted expansion trend modeling mechanism for asynchronous multi-source observation data, this invention no longer relies solely on single test results, single-time-point imaging performance, or static scoring results to determine the timing of surgery for necrotizing fasciitis. Instead, it can continuously characterize the level of deep lesion expansion, expansion rate, expansion acceleration, and the degree of deviation of occult manifestations in the emergency initial consultation scenario. It is particularly suitable for occult cases where the superficial redness and swelling are not yet obvious, but the deep fascial layer has already progressed rapidly, thereby improving the prospectiveness and accuracy of auxiliary determination of surgical timing.

[0048] This invention constructs a linkage update mechanism between surgical trigger risk value and dynamic judgment threshold, and combines modal divergence, global confidence, and adaptive update strategy for review window. It can output graded decision results such as immediate surgery, high-risk parallel supplementary evidence, or short-window review based on the current lesion progression status and the reliability of observation evidence. This avoids the judgment lag, misjudgment, and omission problems caused by waiting for supplementary examinations, relying on single modal evidence, or fixed review rhythm in the prior art, thereby improving the real-time performance, stability, and clinical application reliability of surgical auxiliary decision-making for necrotizing fasciitis. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating the first embodiment of a method for assisting decision-making on surgical timing in necrotizing fasciitis based on lesion expansion trend modeling, according to the present invention.

[0051] Figure 2 This is a schematic diagram of a multimodal deep extension proxy score heatmap of the first embodiment of the necrotizing fasciitis surgery timing auxiliary decision-making method based on lesion extension trend modeling of the present invention.

[0052] Figure 3This is a schematic diagram of the modal reliability weight dynamic allocation heatmap of the first embodiment of the necrotizing fasciitis surgery timing auxiliary decision-making method based on lesion expansion trend modeling of the present invention.

[0053] Figure 4 This is a schematic diagram of the joint output of modal divergence and global confidence of the first embodiment of the auxiliary decision-making method for surgical timing of necrotizing fasciitis based on lesion expansion trend modeling of the present invention.

[0054] Figure 5 This is a schematic diagram of the trend state response surface of the first embodiment of the necrotizing fasciitis surgery timing auxiliary decision-making method based on lesion expansion trend modeling of the present invention.

[0055] Figure 6 This is a two-dimensional contour diagram of the trend state of the first embodiment of the necrotizing fasciitis surgery timing auxiliary decision-making method based on lesion expansion trend modeling of the present invention.

[0056] Figure 7 This is a schematic diagram of the device for an auxiliary decision-making method for surgical timing in necrotizing fasciitis based on lesion expansion trend modeling, according to the present invention. Detailed Implementation

[0057] 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.

[0058] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the necrotizing fasciitis surgery timing auxiliary decision-making method based on lesion expansion trend modeling of the present invention. The first embodiment of the necrotizing fasciitis surgery timing auxiliary decision-making method based on lesion expansion trend modeling of the present invention is presented.

[0059] In the first embodiment, the necrotizing fasciitis surgical timing-assisted decision-making method based on lesion expansion trend modeling includes:

[0060] Step S10: Obtain asynchronous multi-source observation data within the preset initial diagnosis observation window, and perform the initial diagnosis multimodal observation reconstruction task based on the asynchronous multi-source observation data using a unified time axis alignment and observation quality coding mechanism, and output the aligned observation feature tensor and the observation quality state set;

[0061] It should be noted that "asynchronous multi-source observation data" refers to multiple types of patient observation information generated at different times within the same initial observation window due to differences in the type of acquisition equipment, acquisition frequency, acquisition process, and output delay. It does not require all observation data to be generated synchronously at the same moment. It includes vital sign sequences, laboratory indicator sequences, bedside ultrasound observation sequences, body surface manifestation sequences, and CT observation sequences further incorporated when examination conditions are available and data has been actually acquired. Among them, the vital signs sequence mainly reflects the patient's current systemic circulation and infection stress status, including information such as heart rate, mean arterial pressure, body temperature, and respiratory rate; the laboratory indicator sequence mainly reflects the intensity of the inflammatory response, tissue perfusion status, and degree of metabolic disorder, including information such as white blood cell count, C-reactive protein, lactate, serum sodium, and procalcitonin; the bedside ultrasound observation sequence mainly reflects deep involvement signs such as fluid-filled dark areas around the deep fascia layer, fascial thickening, subcutaneous gas accumulation, and abnormal soft tissue echoes; the surface manifestation sequence mainly reflects local changes that are visible to the naked eye or perceptible to palpation, including the area of ​​redness and swelling, pain score, local skin temperature difference, and boundary progression; and the CT observation sequence is used to supplement the description of the length of deep fascial involvement, gas distribution, fluid accumulation, and necrosis-related imaging signs when imaging results have been formed. The "unified time axis alignment and observation quality coding mechanism" does not simply refer to aggregating and arranging the data from the different modalities mentioned above. Instead, it involves first constructing a unified discrete time grid around a preset initial observation window, then mapping the data of each modality to this unified time reference, and simultaneously adding its missing state, the time interval between the most recent valid observation and the current time grid, and the observation quality level to each modality. Here, the "aligned observation feature tensor" refers to the comprehensive modal feature expression reconstructed under the unified time reference, which can be directly used subsequently; the "observation quality state set" refers to the quality description information that corresponds one-to-one with the comprehensive modal features, used to characterize whether a certain modality is complete, missing, delayed, or of low quality at a certain moment.

[0062] Understandably, necrotizing fasciitis is characterized by rapid progression, atypical presentation, and inconsistent data acquisition during the initial emergency visit. Without a unified reconstruction of multi-source data, subsequent calculations regarding the extent, speed, or surgical risk of deep spread may be based on time-disjointed data combinations, leading to biased results. This step maps data from different modalities—originally independent, with varying sampling rhythms and completeness—to a common discrete-time benchmark. This allows vital signs, laboratory abnormalities, bedside ultrasound results, and surface manifestations within the same time grid to be jointly understood under the same initial visit context. Simultaneously, this step avoids treating missing data as meaningless null values, preserving the fact of the missing data and its duration. This enables subsequent steps to differentiate between scenarios such as "not yet collected," "collected a long time ago but data is outdated," and "collected but image quality is poor" when assigning modal confidence weights, thus improving adaptability to the complex environment of the initial visit. Furthermore, the technical effect of this step lies not in improving the judgment ability of a single modality, but in avoiding structural errors at the input level in subsequent steps, ensuring continuity and interpretability of the entire decision-making chain from the starting point. For example, when an emergency patient with significant lower limb pain but not yet widespread surface redness and swelling is admitted, suspected necrotizing fasciitis, vital signs such as slightly rapid heart rate and slightly low mean arterial pressure can be continuously obtained within 10 minutes of triage; a bedside ultrasound is completed 20 minutes after triage, revealing a significant hypoechoic area and fascial thickening near the fascia layer; laboratory results return 45 minutes after triage, indicating elevated white blood cell and lactate levels; the nurse records the extent of local redness and swelling and pain score 70 minutes after triage; and preliminary structured CT results are only available 100 minutes after triage. If the traditional approach were used, these data might simply be written into the medical record in the order they were acquired, and then the doctor would make a comprehensive judgment based on experience, or roughly compare them after being aggregated by the hour. The result of this approach is that while bedside ultrasound may indicate a tendency for deeper involvement within the first 40 minutes, the initial diagnosis might still be conservative due to the limited extent of surface redness and swelling and the lack of CT scan results. In contrast, this invention maps information from different time points onto a unified initial diagnosis discrete time grid, simultaneously identifying the time intervals and quality status of each modality. For example, in earlier time grids, bedside ultrasound and vital signs are identified as valid information closer to the current state, while CT scans are explicitly marked as missing because they haven't been acquired yet. In subsequent time grids, laboratory abnormalities are further incorporated into the aligned feature tensor, and although surface manifestations appear later, their progression is still interpreted within a unified temporal framework. The resulting reconstruction visually presents a scenario where, although the extent of surface redness and swelling is temporarily limited, within the current initial diagnosis observation window, information from deeper modalities has continuously shown abnormalities, and the yet-to-be-acquired CT scan is no longer mistakenly considered "no abnormalities."

[0063] Step S20: Based on the aligned observation feature tensor and the observation quality state set, a modal confidence weighted fusion mechanism is used to perform the deep extended proxy quantity construction task, and output the fused extended level sequence, modal divergence quantity and global confidence.

[0064] It is important to clarify that "deep expansion proxy quantity" refers to a proxy quantity that does not directly use the actual lesion boundary or actual necrosis volume of necrotizing fasciitis as the sole observable object. Instead, it extracts alternative proxy quantities from multi-source observational data available at the initial consultation stage, which can indirectly characterize the extent, continuity, active expansion, and tendency of insidious progression of deep fascia involvement. The core reason for this is that the actual expansion state of the lesion on the deep fascial plane is often not directly obtainable at the same time and with the same precision during the initial emergency consultation stage, especially before surgical exploration, pathological verification, or complete image post-processing. The so-called "actual expansion degree" itself is not directly visible. Therefore, this step does not attempt to directly reconstruct the complete lesion geometry under insufficient input conditions. Instead, it establishes a deep expansion proxy system for the initial consultation scenario by combining bedside ultrasound expansion proxy score, laboratory expansion proxy score, vital sign instability proxy score, surface manifestation proxy score, and CT expansion proxy score constructed when CT observation sequences are available. Among them, the bedside ultrasound extended proxy score mainly reflects near-deep structural abnormalities such as fluid-filled dark areas in the fascia layer, fascial thickening, and subcutaneous gas accumulation; the laboratory extended proxy score mainly reflects the degree of inflammatory burden, decreased tissue perfusion, and metabolic disorders; the vital sign instability proxy score mainly reflects the state of infection stress and changes in circulatory load; the surface manifestation proxy score mainly reflects superficial signs; and the CT extended proxy score, when present, is used to supplement structural evidence of the extent of deep involvement, gas distribution, and fluid accumulation. Furthermore, the "modal credibility weighted fusion mechanism" refers to not simply averaging the proxy scores of each modality, but rather distinguishing the credibility of different modalities under the current unified discrete time grid based on the observation quality state set output in step S10, and generating credibility weights accordingly. This allows modalities with higher quality, stronger timeliness, and closer to the current judgment time to contribute a larger proportion to the fusion result. "Fusion Expansion Level Sequence" refers to the comprehensive characterization result that continuously expresses the degree of deep expansion on a unified discrete-time grid; "Modal Disagreement" refers to the degree of consistency deviation between different modal surrogate scores and fusion expansion level; "Global Confidence" refers to the comprehensive evaluation of whether the fusion expansion level has sufficient credible support at the current moment.

[0065] It should be understood that "deep expansion proxy quantity" refers to a proxy quantity that does not directly take the actual lesion boundary or actual necrosis volume of necrotizing fasciitis as the only observable object. Instead, it extracts alternative proxy quantities from multi-source observation data available at the initial diagnosis stage, which can indirectly characterize the extent of deep fascial involvement, continuity of involvement, active expansion, and tendency of insidious progression. The core reason is that at the emergency initial diagnosis stage, it is often impossible to directly obtain the actual expansion state of the lesion on the deep fascial plane at the same time and with the same precision, especially before surgical exploration, pathological verification, or complete image post-processing. The so-called "actual expansion degree" itself is not directly visible. Therefore, this step does not attempt to directly reconstruct the complete lesion geometry under insufficient input conditions. Instead, it establishes a deep expansion proxy system for the initial diagnosis scenario through bedside ultrasound expansion proxy score, laboratory expansion proxy score, vital sign instability proxy score, surface manifestation proxy score, and CT expansion proxy score constructed when CT observation sequences are available. Among them, the bedside ultrasound extended proxy score mainly reflects near-deep structural abnormalities such as fluid-filled dark areas in the fascia layer, fascial thickening, and subcutaneous gas accumulation; the laboratory extended proxy score mainly reflects the degree of inflammatory burden, decreased tissue perfusion, and metabolic disorders; the vital sign instability proxy score mainly reflects the state of infection stress and changes in circulatory load; the surface manifestation proxy score mainly reflects superficial signs; and the CT extended proxy score, when present, is used to supplement structural evidence of the extent of deep involvement, gas distribution, and fluid accumulation. Furthermore, the "modal credibility weighted fusion mechanism" refers to not simply averaging the proxy scores of each modality, but rather distinguishing the credibility of different modalities under the current unified discrete time grid based on the observation quality state set output in step S10, and generating credibility weights accordingly. This allows modalities with higher quality, stronger timeliness, and closer to the current judgment time to contribute a larger proportion to the fusion result. "Fusion Expansion Level Sequence" refers to the comprehensive characterization result that continuously expresses the degree of deep expansion on a unified discrete-time grid; "Modal Disagreement" refers to the degree of consistency deviation between different modal surrogate scores and fusion expansion level; "Global Confidence" refers to the comprehensive evaluation of whether the fusion expansion level has sufficient credible support at the current moment.

[0066] For example, such as Figure 2As shown in the figure, this heatmap illustrates the changes over time in bedside ultrasound proxy scores, laboratory proxy scores, vital signs proxy scores, surface manifestation proxy scores, and CT proxy scores. The figure clearly shows that in the early stages of emergency department visits, bedside ultrasound and laboratory proxy scores significantly increase before surface manifestations, indicating that deep fascial involvement, inflammatory burden, and perfusion abnormalities have begun to form continuous abnormal signals, while the surface manifestation proxy score remains at a low level, exhibiting the typical characteristic of surface signs lagging behind deep progression. Simultaneously, the CT proxy score is initially null or low, gradually entering the effective range later, reflecting the reality that different modalities are sampled and formed at different times during the initial visit. Figure 3 The figure illustrates the influence of each modality on the fusion result under the current unified discrete-time grid. As can be seen, in the early stages of the initial diagnosis, the weights corresponding to bedside ultrasound and laboratory modalities are significantly higher than those corresponding to surface manifestation modalities. The weight of the CT modality, which has not yet been acquired or has just entered a valid state, is close to zero in the early stages and gradually increases later. This indicates that the present invention does not simply average the proxy scores of each modality, but dynamically distinguishes the freshness, completeness, and reliability of modalities based on the observation quality state set, giving higher weights to modalities that are closer to the current pathological state and can directly reflect deep abnormalities. Even if the surface manifestations are mild at a certain moment, the influence of deep abnormality modalities will not be diluted due to weak surface signs; similarly, even if CT is included later, it will not unconditionally suppress other earlier and higher-quality initial diagnostic evidence at the initial inclusion, thus achieving a data-driven weight allocation more in line with emergency scenarios. Figure 4 As shown, modal discrepancy reflects the degree of consistency deviation between the modal surrogate scores and the fusion expansion level, while global confidence characterizes whether the current fusion result has sufficient reliable evidence. The figure shows that in the mid-stage of the initial diagnosis, as the abnormalities in deep-related modalities gradually increase while surface manifestations remain relatively lagging, the modal discrepancy increases to some extent, identifying an evidence conflict that "deep progression and superficial manifestation are not synchronized." However, at the same time, the global confidence does not significantly decrease to an unusable level, indicating that high-quality bedside ultrasound and laboratory observations still provide stable support for the current fusion conclusion. This invention does not treat all modal differences as noise, but rather explicitly outputs these differences as intermediate states, enabling subsequent steps to distinguish between "the current conclusion is high-risk and the evidence is consistent" and "the current conclusion is high-risk but still has some modal conflict."

[0067] Step S30: Based on the fusion of extended horizontal sequence, modal divergence and global confidence, the extended trend preprocessing task is performed using a time-weighted extended trend modeling mechanism to output the extended trend state vector;

[0068] It should be noted that the "time-weighted expansion trend modeling mechanism" addresses the characteristics of necrotizing fasciitis in the initial emergency visit scenario, namely "rapid lesion evolution, irregular observation times, fluctuating modal consistency, and unstable conclusions at a single time point." It further transforms the fused expansion level sequence, modal divergence, and global confidence obtained in step S20 into an expansion trend state vector that describes the direction, speed, acceleration characteristics, and degree of misalignment in lesion depth representation. The "expansion trend preprocessing task" referred to here is not, in essence, denoising or normalization in the conventional sense, but rather elevates the static fusion result of the previous step into a dynamic state representation with temporal evolution semantics. Furthermore, "expansion rate" refers to how quickly the comprehensive characterization value of deep expansion changes over time on a unified discrete-time grid; "expansion acceleration" refers to whether the expansion rate itself is still increasing, in other words, it describes whether the lesion is not simply expanding, but is in an accelerated expansion phase; "surface expansion rate" refers to the rate of change of visible signs such as surface redness and swelling area and boundary advancement reflected by the body surface manifestation sequence in step S10; "caused characterization deviation" refers to the degree of deviation between the deep expansion trend and the surface visible trend, used to characterize the hidden progression state of "significant deep advancement but relatively lagging surface progression". The "expansion trend state vector" output in step S30 is not a single indicator, but a joint expression composed of the fused expansion level, expansion rate, expansion acceleration, hidden characterization deviation, and auxiliary states related to the current trend calculation. Its role is to provide a more forward-looking input basis for surgical trigger risk calculation than a single-point fused value.

[0069] Understandably, the present invention introduces a time-weighted expansion trend modeling mechanism in step S30. Its main technical effect lies in transforming the fusion expansion level sequence, which originally only represented the current overall state, into a trend-based state that reflects the risk of lesion evolution within a short time window. For necrotizing fasciitis, the real determining factor for whether surgical exploration or debridement should be performed as soon as possible is not only whether the current absolute expansion level is high, but also whether the expansion is rapidly progressing, whether it is in a significantly accelerating phase, and whether deep progression has begun to break free from the constraints of the surface manifestation. If we only look at the fusion expansion level at a certain moment output in step S20, although we can know that the lesion expansion degree is high under the current comprehensive evidence, we still cannot distinguish between two clinically completely different situations: "the current level is high but has stabilized" and "the current level is high and is still rising rapidly." This step introduces a time-weighted coefficient, giving greater influence to historical points that are closer to the present moment and have higher credibility of evidence, making the trend modeling results closer to the immediate risk at the initial diagnosis. By constructing expansion rate and expansion acceleration, it can capture the dynamic changes of lesions transitioning from slow expansion to rapid expansion, or from rapid expansion to further accelerated expansion. By constructing surface expansion rate and hidden characteristic deviation, it can identify dangerous states where superficial signs are still not prominent, but deep fascial abnormalities have been continuously enhanced. Therefore, the technical effect of this step is not limited to obtaining more intermediate variables, but lies in elevating "static abnormality intensity" to "dynamic progression trend," so that when the subsequent step S40 makes a surgical trigger determination, it is no longer just based on a threshold comparison of the degree of abnormality at a certain moment, but can perform earlier and more robust risk identification based on the trend status.

[0070] For example, such as Figure 5 As shown, this figure uses the rate of expansion as one response axis, the deviation of occult features as another, and the trend risk response value as the vertical output. It visually presents the relationship between the overall evolution risk of the lesion when the "speed of deep expansion" and the "degree of misalignment of deep and superficial features" change together. The three-dimensional response surface in the figure gradually rises from a low-value region to a high-value region, indicating that when the rate of expansion is low and the deviation of occult features is small, the determined trend risk remains at a low level. However, as the rate of expansion and the deviation of occult features increase, the response surface rises rapidly, indicating that the lesion has transitioned from general expansion to a more dangerous, rapid evolutionary state. Meanwhile, the contour projection at the bottom further unfolds the three-dimensional surface on the plane, making the boundary changes between different risk levels clearer and facilitating the observation of the transition process from low-risk and medium-risk areas to high-risk areas. It is evident that this invention combines the expansion rate with the deviation of the hidden characterization in the form of a response surface. When the progression of deep lesions accelerates while superficial signs lag, the true degree of lesion deterioration is often underestimated by traditional methods, while this invention can identify this state in advance at the trend level. Figure 6 As shown, this figure uses the rate of expansion and the deviation of concealed features as the horizontal and vertical axes, and uses different color shades and contour line densities to represent the changes in trend risk response values. The figure shows a continuous transition between low-trend-risk, medium-trend-risk, and high-trend-risk areas, indicating that in step S30, the invention does not simply and crudely classify the patient's state into "normal" or "dangerous," but rather divides the lesion evolution into different levels of risk areas based on different combinations of trend state vectors. As the rate of expansion increases from low to high and the deviation of concealed features increases from small to large, the contour lines gradually tighten and the color gradually deepens, indicating a continuous increase in trend risk response values. Especially in the upper right region, a concentrated distribution of high-trend-risk areas can be clearly seen, indicating that when both the deep expansion rate and the degree of deep-to-shallow misalignment reach a high level, it is identified as a dangerous trend state that requires close attention. The actual trend trajectory points in the initial observation window of the figure gradually migrate from the lower left area to the upper right area, reflecting that the patient's status continuously crosses different risk levels over time. It can not only calculate the expansion rate and the deviation of the hidden representation at a certain moment, but also use the positional changes of these state quantities in the trend space to reveal the continuous evolution path of the lesion from low risk to medium and high risk.

[0071] Step S40: Based on the extended trend state vector, a dynamic threshold adaptive update mechanism is used to perform the surgical trigger determination task, and the surgical trigger risk value and dynamic determination threshold are output;

[0072] It is important to clarify that the "surgical trigger determination task" is not a direct mechanical threshold comparison of the expansion trend state vector output in step S30. Instead, it addresses the challenges of rapid changes in the condition within a short time window, fluctuating evidence quality, varying tolerance risks among different cases, and the tendency for fixed thresholds to trigger too late or too early in the emergency initial consultation scenario for necrotizing fasciitis. It further constructs a determination process that simultaneously possesses risk expression capabilities and boundary adaptation capabilities. The "surgical trigger risk value" referred to here is a quantitative risk expression based on the current expansion trend state vector, indicating whether the patient has reached the point where surgical exploration or debridement should be performed as soon as possible within a pre-defined short time window. It emphasizes not the general severity of infection, but the probability of triggering the specific clinical goal of "the timing of surgery." The "dynamic determination threshold" referred to here is not a pre-written fixed constant, but a determination boundary that can be dynamically adjusted based on information such as the current lesion expansion level, expansion rate, modal divergence, and global confidence level. Furthermore, the purpose of the dynamic threshold adaptive update mechanism is not just to make the threshold "change", but to make the threshold change have a clear scenario logic: when the expansion trend is faster and the expansion level is higher, the threshold should be appropriately reduced; while when the modal divergence is significantly increased and the global confidence decreases, the threshold should be appropriately increased, so as to avoid generating overly aggressive trigger judgments in the case of insufficient evidence or large internal conflicts.

[0073] Understandably, the present invention introduces a dynamic threshold adaptive update mechanism in step S40. Its main technical effect lies in further transforming the trend state expression already formed in step S30 into a trigger judgment result that can directly serve clinical action selection. For necrotizing fasciitis, the real clinical challenge is not merely knowing that the lesion is developing, but rather determining whether the patient has entered a "no longer suitable waiting" window when information is not yet complete and surface signs may still be atypical. If only an extended trend state vector is available, although it shows that the patient's lesion is developing rapidly, it still lacks an actionable judgment boundary. This step, by constructing a surgical trigger risk value, transforms "trend severity" into "surgical trigger probability or risk intensity." Furthermore, through a dynamic judgment threshold, this risk value can be contextualized based on the consistency and credibility of the current evidence.

[0074] It should be understood that traditional fixed threshold methods typically assume that the same trigger boundary should be used for all cases, all initial consultation scenarios, and all evidence quality conditions. However, this approach is often unreasonable in the initial consultation of necrotizing fasciitis because the disease is characterized by "asynchronous arrival of superficial and deep tissues, inconsistent order of evidence arrival, and rapid deterioration within a short period of time." If the threshold is set too high, it is easy to delay triggering in truly dangerous cases; if the threshold is set too low, it is easy to trigger an over-alarm when the evidence quality is insufficient. Traditional single-indicator limit-breaking methods are more susceptible to fluctuations in individual abnormal values. For example, a short-term increase in lactate or abnormal body temperature does not necessarily mean that the deep fascia has expanded to the point where immediate surgery is required. If only a single laboratory or vital sign indicator is relied upon, "abnormal systemic response" may be misjudged as "deep lesions must be debrided immediately." While traditional static risk scoring can integrate several variables, it typically outputs a fixed value after a single scoring, failing to automatically adjust the judgment boundary as trends change within the initial observation window. Therefore, it struggles to reflect the fact that "the same risk score has different clinical meanings under high-confidence, consistent evidence versus high-disagreement, low-confidence evidence." In contrast, this invention upgrades the judgment task from "single-value judgment" to "risk-boundary coupled judgment" by simultaneously outputting surgical trigger risk values ​​and dynamic judgment thresholds. By allowing expansion rate, fusion expansion level, modal disagreement, and global confidence to participate in threshold updates, it can anticipate accelerating risks and exercise caution when evidence is insufficient, significantly improving the rigidity and lag of traditional fixed judgment logic in initial consultation scenarios. Particularly in diseases like necrotizing fasciitis, which exhibit insidious and rapid progression, the improvements demonstrated by this step directly manifest as earlier identification of truly high-risk patients, less reliance on the complete formation of superficial typical signs, and a more reasonable distinction between the clinical boundaries of "immediate triggering" and "cautious supplementation of evidence."

[0075] Step S50: Perform graded trigger control based on surgical trigger risk value and dynamic judgment threshold, and output auxiliary decision results for surgery.

[0076] It's important to clarify that "tiered trigger control" doesn't mean simply outputting a yes or no conclusion after the risk exceeds a certain threshold. Instead, based on the relationship between the surgical trigger risk value established in step S40 and the dynamic judgment threshold, and combined with information reflecting the consistency and credibility of evidence from previous steps, the current patient status is categorized into different decision-making levels. This results in auxiliary decision-making outcomes that directly support the clinical treatment schedule. The "auxiliary decision-making outcomes for surgery" mentioned here can include high-risk trigger states recommending immediate surgical exploration or debridement assessment; high-risk concurrent evidence supplementation states recommending immediate consultation and simultaneous supplementation of key evidence; and intermediate-risk states recommending shortening the review period and continuing short-window observation. Furthermore, "tiering" is not intended to increase complexity, but rather to address the issue that risk and evidence quality are not always synchronized in the initial consultation scenario. In other words, in some cases, the patient's risk value for triggering surgery is already high, but if the current modal discrepancy is still too large or the global confidence level has not yet reached a sufficiently stable level, directly outputting "immediate surgery" may seem too rigid; while in other cases, although the patient's risk has not completely exceeded the high-risk trigger boundary, the expansion trend has been significantly enhanced. If we still wait for re-evaluation according to the conventional fixed rhythm, it may cause a delay in timing.

[0077] It is understandable that the present invention introduces graded trigger control in step S50, the main technical effect of which is to transform the risk value and dynamic threshold output in step S40 into clinically executable and priority-distinguishing decision results. For the initial management of necrotizing fasciitis, the real key is not just "determining whether it is high-risk," but rather what actions should be taken after a high-risk case, and how to arrange re-evaluation and supplementary evidence for intermediate risks. Through this step, cases with risk values ​​exceeding the dynamic threshold and strong evidence support can be directly output as immediately triggerable for surgery; cases with high risk values ​​but insufficient convergence of evidence discrepancies can be output as high-risk cases with concurrent supplementary evidence, thus prompting clinicians to accelerate consultations, re-scans, and key laboratory re-examinations without letting their guard down; and cases that have not yet reached the immediate trigger level but have significantly deviated from the low-risk range can be output as short-window re-evaluation cases, with a more compact re-evaluation window provided simultaneously. The resulting technical effect is that it no longer simply tells doctors "this case is high-risk," but further tells them "which treatment pace should be adopted for this case under the current evidence conditions." Especially in environments with limited emergency resources, time delays in obtaining examinations, and rapidly changing patient conditions, this mapping from "risk expression" to "action level" allows the auxiliary decision-making results to better align with clinical workflows, reducing the uncertainty of simple numerical output at the application level. For example, at the end of an emergency initial observation window, step S40 has already output a high surgical trigger risk value, exceeding the current dynamic judgment threshold; simultaneously, steps S20 and S30 indicate that the patient's modal dissonance is not high, the global confidence is good, and the expansion rate and expansion acceleration are both in a significantly increased range. In this case, step S50 will no longer remain at the abstract judgment of "high risk," but will directly classify the patient into an immediate surgical trigger state, indicating that it is not advisable to continue waiting according to the routine review schedule. As another example, another patient's surgical trigger risk value also exceeds the dynamic judgment threshold, but their modal dissonance is large because bedside ultrasound strongly suggests deep abnormalities, while surface manifestations are still mild, CT scans are not yet complete, and the global confidence is at a moderate level. For such cases, traditional methods only output a single alarm, which may result in insufficient evidence if an alarm is triggered immediately, or delays in intervention if no alarm is triggered. However, in step S50 of this invention, the case is categorized as a high-risk, concurrent supplementary evidence state, prompting immediate surgical consultation and expedited critical examinations, and continued evaluation at a high level of vigilance, rather than a simple one-size-fits-all approach. For example, if a patient's current surgical trigger risk value does not exceed the dynamic judgment threshold, but the rate of spread is significantly higher than before, the spread acceleration has turned from negative to positive, and the deviation of hidden characteristics continues to increase, the patient will be categorized as a short-window reassessment state, and the next reassessment window will be automatically tightened. In this way, even if the immediate trigger interval has not yet been reached, clinicians will not wait for a longer interval for reassessment as usual, but can reassess the case in a shorter time, preventing rapidly progressing cases from being delayed during observation.As can be seen, step S50, through hierarchical trigger control, truly translates the judgment result into auxiliary decision outputs of different priorities, extending the entire method from data processing, trend modeling, and risk assessment to the organization and arrangement at the clinical action level, thereby significantly enhancing its practical application value in auxiliary decision-making regarding the timing of initial surgery for necrotizing fasciitis.

[0078] Example 2: Furthermore, the present invention provides a decision-making system for the timing of surgery in necrotizing fasciitis based on lesion expansion trend modeling. This system employs a decision-making method for the timing of surgery in necrotizing fasciitis based on lesion expansion trend modeling as described in the above embodiments, and can solve the technical problem of decision-making for the timing of surgery in necrotizing fasciitis based on lesion expansion trend modeling. Compared with the prior art, the beneficial effects of the decision-making system for the timing of surgery in necrotizing fasciitis based on lesion expansion trend modeling provided by the present invention are the same as those of the decision-making method for the timing of surgery in necrotizing fasciitis based on lesion expansion trend modeling provided in the above embodiments. Moreover, other technical features of the decision-making system for the timing of surgery in necrotizing fasciitis based on lesion expansion trend modeling are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0079] Example 3: This invention provides a decision-making aid device for surgical timing in necrotizing fasciitis based on lesion expansion trend modeling. Please refer to... Figure 7A surgical timing assistance device for necrotizing fasciitis based on lesion expansion trend modeling includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the surgical timing assistance method for necrotizing fasciitis based on lesion expansion trend modeling described in Embodiment 1 above. The surgical timing assistance device for necrotizing fasciitis based on lesion expansion trend modeling in this embodiment of the invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. This surgical timing assistance device for necrotizing fasciitis based on lesion expansion trend modeling is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the invention. A surgical timing aid for necrotizing fasciitis based on lesion expansion trend modeling may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the surgical timing aid for necrotizing fasciitis based on lesion expansion trend modeling. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows a necrotizing fasciitis surgery timing aid decision-making device based on lesion expansion trend modeling to wirelessly or wiredly communicate with other devices to exchange data. Although the figure shows a necrotizing fasciitis surgery timing aid decision-making device based on lesion expansion trend modeling with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0080] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for assisting decision-making on the timing of surgery for necrotizing fasciitis based on lesion expansion trend modeling. The computer program product provided by this invention can solve the technical problem of assisting decision-making on the timing of surgery for necrotizing fasciitis based on lesion expansion trend modeling. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the above-described method for assisting decision-making on the timing of surgery for necrotizing fasciitis based on lesion expansion trend modeling, and will not be repeated here.

[0081] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0082] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0083] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for assisting decision-making on surgical timing in necrotizing fasciitis based on lesion expansion trend modeling, characterized in that, The methods include: Step S10: Obtain asynchronous multi-source observation data within the preset initial diagnosis observation window, and perform the initial diagnosis multimodal observation reconstruction task based on the asynchronous multi-source observation data using a unified time axis alignment and observation quality coding mechanism, and output the aligned observation feature tensor and the observation quality state set; Step S20: Based on the aligned observation feature tensor and the observation quality state set, a modal confidence weighted fusion mechanism is used to perform the deep extended surrogate quantity construction task, outputting the fused extended level sequence, modal divergence quantity, and global confidence; Step S20 specifically includes: Step S201: Based on the aligned observation feature tensor Bedside ultrasound extended proxy scoring was constructed respectively. Laboratory extended agent rating Unstable vital signs assessment Physical appearance proxy score And for CT observation sequences Building CT Extended Agent Scoring ; Step S202: Based on the observation quality state set A modal credibility weight construction mechanism is used to generate credibility weights for each effective mode. and utilize trusted weights The scores of each modality's extended agent are weighted and fused to obtain a fused extended level sequence. ; Step S202 specifically involves: first, scoring the modal quality based on the observed quality state set. Modal reliability weights are constructed according to the following formula. : ;in, To be compatible with the unified discrete-time grid The corresponding set of effective modes, For the effective mode set The index identifier of any valid mode in the middle, For the first The mass sensitivity coefficient corresponding to each effective mode For the first Each effective mode is in a unified discrete-time grid. Modal quality score at the location, For the first Each effective mode is in a unified discrete-time grid. Modal quality score at the location, For the first The mass sensitivity coefficient corresponding to each effective mode It is a natural exponential function; Then add modal confidence weights Corresponding bedside ultrasound extended agent score Laboratory extended agent rating Unstable vital signs assessment Physical appearance proxy score and CT extended agent score Multiply and sum to obtain the fused extended level sequence. ; Step S203: Based on the extended agent scores of each modality and the fusion extended level sequence Calculate the modal bifurcation based on the deviation between them. And based on the credibility weight The global confidence score is obtained by using a modal divergence exponential decay method to aggregate credible weights. ; Step S204: Final output: fused extended horizontal sequence Modal divergence and global confidence ; Step S30: Based on the fused extended level sequence, modal divergence, and global confidence, a time-weighted extended trend modeling mechanism is used to perform the extended trend preprocessing task, outputting the extended trend state vector; Step S30 specifically includes: Step S301: Construct time-weighted coefficients based on the fused extended level sequence and global confidence using an exponential time decay weighting method. The expansion rate was constructed using time-weighted local regression based on the fused extended level sequence and modal divergence. ; Step S302: Based on the spread rate of adjacent time moments Calculate the difference spread acceleration between adjacent time steps. Simultaneously acquire the body surface manifestation sequence. Based on body surface manifestation sequence The surface spread rate was calculated using the difference in redness and swelling area between adjacent time points. ; and according to the expansion rate Extended acceleration and surface spread rate Constructing hidden representation deviation using a differential coupling method between deep extension and surface extension ; Step S303: Obtain information on the area of ​​redness and swelling on the body surface, and combine it with the deviation of the hidden characteristics. Expansion rate Extended acceleration and surface spread rate Combined into an extended trend state vector ; Step S40: Based on the extended trend state vector, a dynamic threshold adaptive update mechanism is used to perform the surgical trigger determination task, and the surgical trigger risk value and dynamic determination threshold are output; Step S50: Perform graded trigger control based on surgical trigger risk value and dynamic judgment threshold, and output auxiliary decision results for surgery.

2. The method for assisting in decision-making regarding surgical timing in necrotizing fasciitis based on lesion expansion trend modeling as described in claim 1, characterized in that, Step S10 involves acquiring asynchronous multi-source observation data within a preset initial observation window, performing initial multimodal observation reconstruction based on the asynchronous multi-source observation data using a unified time axis alignment and observation quality coding mechanism, and outputting the aligned observation feature tensor and observation quality state set. Specifically, this includes: Step S101: Obtain the triage time Pre-set observation window for initial consultation Vital signs sequence within Laboratory indicator series Bedside ultrasound observation sequence Surface manifestation sequence ,in, This indicates the preset initial observation window length, and further acquisition of CT observation sequences. This forms the original asynchronous observation set; Step S102: Construct a unified discrete-time grid based on the original asynchronous observation set And for any mode in a unified discrete-time grid The original observation sequence is time-aligned using a time-decaying kernel alignment method to generate modal alignment features corresponding to a unified discrete time grid; Step S103: Construct the aligned observation feature tensor based on the modal alignment features using a modal feature splicing and fusion method. Based on modal alignment features, an observation quality state set is constructed using a joint coding method based on missing mask and observation time interval quantization. .

3. The method for assisting in decision-making regarding surgical timing in necrotizing fasciitis based on lesion expansion trend modeling as described in claim 1, characterized in that, Step S40, which involves performing the surgical trigger determination task based on the extended trend state vector using a dynamic threshold adaptive update mechanism, and outputting the surgical trigger risk value and the dynamic determination threshold, specifically includes: Step S401: Construct a surgical trigger risk value based on the extended trend state vector using a short-time-window surgical demand risk calculation method. Surgical trigger risk value This value is used to characterize the risk level of a patient needing surgical exploration or debridement within a pre-defined short time window at the current moment; surgical trigger risk value. Calculate according to the following formula: ; in, For the Sigmoid function, To be compatible with the unified discrete-time grid Corresponding fusion extended level sequence The value at the current moment, For expansion rate, To extend acceleration, To conceal the deviation in representation, This is the normalized value for lactic acid. This is the inversely normalized value of mean arterial pressure. The bias coefficient, to This represents the corresponding risk contribution coefficient; among which, the surgical trigger risk value... During the calculation process, dimensionless normalization is performed on each input variable; Step S402: Construct a dynamic judgment threshold using a dynamic threshold adaptive update method. Dynamically determine threshold Update according to the following formula: ； in, This represents the next moment in the unified discrete-time grid. Represents the current time in a unified discrete-time grid. The corresponding dynamic judgment threshold, Represents the next time step in a unified discrete-time grid. The corresponding updated dynamic threshold is determined. This represents a threshold truncation function used to restrict the updated dynamic threshold to a preset minimum boundary. With the preset maximum boundary between; This is the expansion rate adjustment coefficient. To integrate the extended level adjustment coefficient, This is the modal bifurcation adjustment coefficient. This is the global confidence level compensation adjustment coefficient; For modal divergence, For global confidence; Step S403: Set the surgical trigger risk value With dynamic judgment threshold When conducting a joint comparison, When the condition is met, a high-risk trigger result is output; when When the condition is not met, the system will determine that the surgical triggering conditions have not been met and output a non-high-risk triggering result, and finally output the surgical triggering risk value and dynamic judgment threshold.

4. The method for assisting in decision-making regarding surgical timing in necrotizing fasciitis based on lesion expansion trend modeling as described in claim 1, characterized in that, Step S50, which involves performing graded trigger control based on the surgical trigger risk value and dynamic judgment threshold, and outputting the auxiliary decision-making results for the surgery, specifically includes: Step S501: Obtain the modal divergence value and global confidence level, and perform hierarchical triggering judgment in combination with the surgical trigger risk value and dynamic judgment threshold; when the surgical trigger risk value is greater than or equal to the dynamic judgment threshold, and the global confidence level is not lower than the preset lower confidence limit and the modal divergence value is not higher than the preset upper divergence limit, output the immediate surgical triggering status. Step S502: When the surgical trigger risk value is greater than or equal to the dynamic judgment threshold, and there is a situation where the global confidence is lower than the preset lower confidence limit or the modal divergence is higher than the preset upper divergence limit, output the high-risk parallel supplementary evidence status; when the surgical trigger risk value is lower than the dynamic judgment threshold but not lower than the preset intermediate risk threshold, output the short window re-evaluation status. Step S503: Based on the current decision state after the execution of the hierarchical trigger judgment, the surgical trigger risk value, the dynamic judgment threshold, the modal divergence value, and the global confidence, generate and output the auxiliary decision result of the surgery.

5. A decision-making system for surgical timing in necrotizing fasciitis based on lesion expansion trend modeling, applied to the decision-making method for surgical timing in necrotizing fasciitis based on lesion expansion trend modeling as described in any one of claims 1 to 4, characterized in that, The necrotizing fasciitis surgical timing auxiliary decision-making system based on lesion expansion trend modeling includes: The multi-source observation reconstruction module is used to acquire asynchronous multi-source observation data within the preset initial diagnosis observation window. Based on the asynchronous multi-source observation data, a unified time axis alignment and observation quality coding mechanism is used to perform the initial diagnosis multimodal observation reconstruction task, and output the aligned observation feature tensor and the observation quality state set. The extended proxy fusion module is used to perform the deep extended proxy quantity construction task based on the aligned observation feature tensor and the observation quality state set, and adopts the modal confidence weighted fusion mechanism to output the fused extended level sequence, modal divergence quantity and global confidence. The trend state generation module is used to perform extended trend preprocessing tasks based on the fused extended horizontal sequence, modal divergence, and global confidence, and to output the extended trend state vector by using a time-weighted extended trend modeling mechanism. The risk threshold determination module is used to perform the surgical trigger determination task based on the extended trend state vector and the dynamic threshold adaptive update mechanism, and outputs the surgical trigger risk value and the dynamic determination threshold. The decision output module is used to perform hierarchical trigger control based on surgical trigger risk values ​​and dynamic judgment thresholds, and output auxiliary decision results for surgery.

6. A decision-making aid device for surgical timing in necrotizing fasciitis based on lesion expansion trend modeling, characterized in that, The necrotizing fasciitis surgery timing assistance device based on lesion expansion trend modeling includes: a memory, a processor, and a necrotizing fasciitis surgery timing assistance program based on lesion expansion trend modeling stored in the memory and executable on the processor. When the necrotizing fasciitis surgery timing assistance program based on lesion expansion trend modeling is executed by the processor, it implements the necrotizing fasciitis surgery timing assistance method based on lesion expansion trend modeling as described in any one of claims 1 to 4.

7. A computer program product, characterized in that, The computer program product includes a decision-making aid program for the timing of surgery for necrotizing fasciitis based on lesion expansion trend modeling. When the decision-making aid program for the timing of surgery for necrotizing fasciitis based on lesion expansion trend modeling is executed by the processor, it implements a decision-making aid method for the timing of surgery for necrotizing fasciitis based on lesion expansion trend modeling as described in any one of claims 1 to 4.

Images