Anesthesia preoperative risk assessment method and system based on multi-source data fusion

Abstract

The invention discloses an anesthesia preoperative risk assessment method and system based on multi-source data fusion, and relates to the technical field of medical informationization, and the method comprises the following steps: S001, building a cross-time-scale illumination and skin state combined observation baseline, injecting a micro-amplitude polarization mark in an image collection chain, and obtaining an image collection chain; extracting a hot spot track and a boundary definition curve through frame-by-frame image analysis, and constructing a time reference point set of a reflection behavior; and S002, based on the time reference point set, executing optical reflection decomposition on the observation image sequence, separating mirror reflection and diffuse reflection components under a unified light field model, generating a double-domain characteristic spectrum of a reflection domain and a blood flow response domain, and constructing an artifact risk map. According to the method, the time reference point and the polarization mark are constructed, illumination and reflection behavior modeling and separation are achieved, anti-fact playback, causal checking and dynamic regulation are combined, artifacts are accurately removed, real blood flow features are extracted, and the accuracy and stability of preoperative risk assessment are improved.

Description

Technical Field

[0001] This invention relates to the field of medical information technology, specifically to a method and system for pre-anesthesia risk assessment based on multi-source data fusion. Background Technology

[0002] Preoperative anesthesia risk assessment based on multi-source data fusion refers to the simultaneous collection, correlation analysis, and fusion modeling of physiological, imaging, laboratory, and behavioral data from different sources before a patient undergoes anesthesia or surgery, using intelligent sensing systems, medical information systems, and multimodal data analysis technologies. This enables the scientific quantification and intelligent prediction of anesthesia-related risks. Specifically, this method relies on wearable intelligent sensing systems to monitor key physiological parameters such as heart rate, blood pressure, blood oxygen saturation, respiratory rate, and electrocardiogram signals in real time. It also combines this with data stored in the hospital information system, including past medical history, drug allergy records, laboratory test indicators, and preoperative questionnaire data. Through data fusion algorithms, feature extraction and risk factor correlation analysis are performed on the multi-source heterogeneous information to construct an individualized risk assessment model. This model can identify potential high-risk conditions, such as abnormal anesthetic drug metabolism, circulatory instability, or airway management difficulties, providing anesthesiologists with precise risk grading and intervention recommendations. This transforms preoperative assessment from experience-based judgment to data-driven intelligent decision-making, improving anesthesia safety and clinical management efficiency.

[0003] The existing technology has the following shortcomings:

[0004] In existing technologies, preoperative risk assessment methods based on microcirculation imaging features typically rely on optical microscopy to extract the flow velocity, density, and morphological characteristics of subcutaneous capillaries to determine the patient's circulatory stability and tissue perfusion status. However, because existing technologies do not adequately suppress interference from tissue surface reflections, high-brightness artifacts and shadow distortions easily occur during imaging when the illumination angle, surface humidity, or skin gloss changes. These artifacts are extremely similar to the characteristics of real vasospasm in morphology and grayscale distribution, causing the model to misclassify optical reflections as vascular abnormalities during feature extraction and classification, thus generating false high-risk markers. This misidentification can lead to the assessment system outputting abnormally high preoperative risk levels, inducing physicians to implement unnecessary drug interventions or anesthesia prophylaxis measures. This not only increases the burden on patients but may also disrupt the accurate formulation of preoperative anesthesia strategies, seriously affecting the reliability of overall risk assessment and the safety of clinical decisions.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for pre-anesthesia risk assessment based on multi-source data fusion, so as to solve the problems in the background art mentioned above.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a pre-anesthesia risk assessment method based on multi-source data fusion, comprising the following steps:

[0008] S001, establish a joint observation baseline for illumination and skin condition across time scales, inject micro-polarization markers into the image acquisition chain, extract bright spot trajectories and boundary sharpness curves through frame-by-frame image analysis, and construct a set of time reference points for reflection behavior;

[0009] S002, based on a set of time reference points, performs optical reflectance decomposition on the observed image sequence, separates specular reflection and diffuse reflection components under a unified light field model, generates a dual-domain feature spectrum of the reflection domain and the blood flow response domain, and constructs an artifact risk map;

[0010] S003 utilizes artifact risk maps to perform temporal counterfactual replay, simulating illumination angle and surface humidity disturbances, eliminating transient textures, and extracting a set of consistent real blood flow evidence across multiple time series.

[0011] S004, based on a set of real blood flow evidence, constructs a consistency criterion for morphology, grayscale and temporal sequence, performs causal verification, locates high-risk suspicious points through fine-grained thresholds, and limits the response boundary of risk output;

[0012] S005, based on the risk response boundary, performs dynamic image control. Through phase conjugate polarization rotation scanning linked with reversible time grid rearrangement and energy return path, it adjusts the optical incident and polarization states online to achieve artifact suppression and blood flow signal enhancement, and outputs closed-loop controlled preoperative risk assessment results.

[0013] Preferably, step S001 includes:

[0014] Before image acquisition, the lighting conditions are adjusted and the incident light angle change sequence is set to form a two-dimensional dynamic acquisition baseline with the principal axis of illumination and the auxiliary axis of skin condition change.

[0015] During image acquisition, micro-polarization markers are injected, and the polarization direction is switched at a fixed frequency using an adjustable polarizer, thus establishing a time correspondence between image frames and polarization directions.

[0016] Based on dual-baseline image data, the image sequence is analyzed frame by frame to extract bright spot trajectories, grayscale fluctuations and boundary sharpness features, and dynamic annotation is performed in combination with polarization response differences;

[0017] Based on the changing trends of image features and polarization differences, key frame positions are selected, and a set of time reference points for illumination changes and reflection behavior is constructed for time alignment and reflection interference calibration in the image processing process.

[0018] Preferably, step S002 includes:

[0019] Based on the set of time reference points, local illumination modeling is performed on the image frames at the corresponding time nodes, the image is divided into multiple analysis sub-regions, and the regions are divided into specular reflection-dominated regions and diffuse reflection-dominated regions according to the characteristics of bright spot trajectory changes.

[0020] By combining a unified light field modeling framework, physical decoupling is performed on different reflection components in the image, and pixel-level spatial separation of specular reflection components and diffuse reflection components is completed based on the reflection intensity distribution law.

[0021] After completing the spatial separation of the reflection components, the reflection behavior feature spectrum and the blood flow response feature spectrum are constructed respectively, and a two-dimensional matrix set is established in pixels to identify the intersection region of reflection and blood flow.

[0022] Based on the results of dual-domain feature spectrum analysis, an artifact risk map is constructed, and the interference risk level of image regions is marked for reference in subsequent artifact processing.

[0023] Preferably, the spatial separation of the specular reflection component is determined based on the reflection intensity gradient, edge sharpness, and consistency with the illumination direction, while the spatial separation of the diffuse reflection component is determined based on the uniformity of grayscale distribution and structural stability.

[0024] Preferably, step S003 includes:

[0025] Based on the artifact risk map, sensitive areas of interference were selected, and the image acquisition sequence was reconstructed under different lighting angles and skin humidity conditions to form a complete perturbation combination group.

[0026] Based on the reconstructed image sequence, the stability of blood vessel morphology, gray-level distribution and boundary contour features is calibrated, and regions that remain consistent under most perturbation conditions are identified as high-stability regions.

[0027] Transient texture recognition is performed on regions with poor stability to remove bright spots, dark spots, or abnormal edge structures that lack coherence and reproducibility in perturbed frames;

[0028] We extract real blood flow features from highly stable regions and construct a set of real blood flow evidence with cross-perturbation consistency and temporal coherence for subsequent causal verification.

[0029] Preferably, the image features in the set of true evidence of blood flow include the consistency curve of blood vessel orientation, the standard deviation sequence of grayscale distribution, the edge sharpness score, and the response trend curves under the corresponding changes in illumination angle and skin humidity.

[0030] Preferably, step S004 includes:

[0031] Based on a set of real blood flow evidence, a joint consistency judgment criterion of image morphological distribution, gray intensity and temporal evolution path is constructed, and a comprehensive consistency score is assigned to each image region.

[0032] Based on the overall consistency score, causal verification is carried out to identify interference-induced image regions that show abnormal changes after the disturbance and return to their original state after the disturbance ends as high-risk suspicious areas;

[0033] For high-risk suspicious areas, a fine-grained threshold sequence is introduced, and the changes in morphology, grayscale and evolution path are compared according to the time segment interval to achieve high-precision positioning;

[0034] After locating high-risk areas, the anomaly response time window is compressed, and the response boundary for risk output is defined in conjunction with the spatial location range, for reference in subsequent assessment and intervention.

[0035] Preferably, the defined response boundary is determined by continuously reviewing the time period between the first abnormal frame and the termination frame of the high-risk suspicious area, eliminating areas that do not meet the consistency performance, and retaining only the image areas that continuously show high consistency change characteristics throughout the entire disturbance process for risk assessment.

[0036] Preferably, step S005 includes:

[0037] Based on the defined high-risk response boundary, a phase conjugate polarization rotation scanning mechanism is initiated to dynamically adjust the polarization state of the incident light in order to counteract the interference of specular reflection energy.

[0038] A reversible temporal raster rearrangement structure is constructed by linking image frame sequences. High-interference frames are reversed and low-interference compensation frames are inserted to achieve temporal structure buffering and light field input stabilization.

[0039] Based on time rearrangement, an energy return path adjustment process based on the Hamiltonian variational principle is initiated to guide the energy of the bright area to return to the low interference area in order to achieve image energy balance.

[0040] Blood flow-related indicators are re-extracted from the modulated image sequence, and adaptive interference suppression assessment results with spatial accuracy and temporal coherence are output for preoperative risk level determination.

[0041] The preoperative anesthesia risk assessment system based on multi-source data fusion includes a baseline illumination observation module, a reflectance decomposition modeling module, a counterfactual playback verification module, a causal consistency determination module, and a dynamic regulation assessment module.

[0042] The illumination observation baseline module establishes a joint observation baseline for illumination and skin condition across time scales, injects micro-polarization markers into the image acquisition chain, and extracts bright spot trajectories and boundary sharpness curves through frame-by-frame image analysis to construct a set of time reference points for reflection behavior.

[0043] The reflection decomposition modeling module performs optical reflection decomposition on the observed image sequence based on a set of time reference points. Under a unified light field model, it separates specular reflection and diffuse reflection components, generates a dual-domain feature spectrum of the reflection domain and the blood flow response domain, and constructs an artifact risk map.

[0044] The counterfactual playback verification module uses the artifact risk map to perform temporal counterfactual playback, simulates the light angle and surface humidity disturbance, removes transient textures, and extracts a set of consistent blood flow evidence across multiple time series.

[0045] The causal consistency determination module, based on the real blood flow evidence set, constructs morphological, grayscale and temporal consistency criteria, performs causal verification, locates high-risk suspicious points through fine-grained thresholds, and limits the response boundary of risk output;

[0046] The dynamic control and assessment module, based on the risk response boundary, performs dynamic image control. Through phase conjugate polarization rotation scanning linked with reversible time grid rearrangement and energy return path, it adjusts the optical incident and polarization state online to achieve artifact suppression and blood flow signal enhancement, and outputs closed-loop controlled preoperative risk assessment results.

[0047] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0048] This invention achieves refined modeling and temporal alignment of reflection behavior under varying illumination and skin conditions by introducing a time reference point and micro-polarization markers. Then, through optical decomposition and feature spectrum construction, specular reflection and blood flow signals are effectively separated. Based on this, a counterfactual replay mechanism simulates the impact of different perturbation conditions on image features, accurately identifying and eliminating unstable transient artifact structures and extracting highly consistent real blood flow features. Furthermore, through causal verification and fine-grained thresholding strategies, high-risk suspicious points are precisely located and their response ranges are limited. Finally, by combining phase-conjugate polarization rotation scanning and energy return path modulation, real-time dynamic suppression of reflection artifacts and enhanced extraction of blood flow response signals are achieved, forming a robust evaluation result with closed-loop control. The overall solution not only improves the adaptability to optical interference during image processing but also enhances the generalization ability of the risk identification mechanism to complex clinical conditions, providing more reliable and intelligent data support for preoperative decision-making. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0050] Figure 1 This is a flowchart of the preoperative risk assessment method for anesthesia based on multi-source data fusion according to the present invention.

[0051] Figure 2 This is a schematic diagram of the modules of the pre-anesthesia risk assessment system based on multi-source data fusion of the present invention. Detailed Implementation

[0052] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0053] This invention provides, for example Figure 1 The preoperative anesthesia risk assessment method based on multi-source data fusion, as shown, includes the following steps:

[0054] S001, establish a joint observation baseline for illumination conditions and skin surface conditions across time scales, inject micro-polarization markers into the image acquisition link, extract bright spot trajectories and boundary sharpness change curves through frame-by-frame image analysis, and construct a set of time reference points for illumination changes and surface reflection behavior, which are used for temporal alignment and reflection behavior calibration in subsequent image processing.

[0055] This step addresses the image reflection artifact interference caused by unstable lighting and changes in skin surface condition during preoperative assessment. It proposes an image acquisition method combining time-scale observation and micro-amplitude polarization injection, aiming to establish a temporal reference for lighting and reflection behavior for interference calibration and temporal correction in subsequent image decomposition and risk identification. The specific implementation steps are as follows:

[0056] During the acquisition process, a baseline for observing illumination changes and skin surface conditions across multiple time scales needs to be established. Before image acquisition begins, the external lighting conditions of the predetermined acquisition area are controlled, and a sequence of incident light angle changes of a certain amplitude is set using a controllable light source. This sequence should include multiple levels of change, from low-angle side illumination to direct vertical illumination. Each level of illumination angle should maintain stable exposure for at least two time cycles to observe the reflective behavior of the skin surface under different illumination conditions. Simultaneously, natural changes in skin condition, such as epidermal moisture, oil distribution, and fine line dynamics, are recorded. At least 200 frames of image data are continuously acquired within a preset time range to capture the evolution trajectory of highlight areas caused by minute changes in skin surface condition. Through the above settings, a two-dimensional dynamic acquisition baseline is formed with illumination change as the main axis and skin condition change as the secondary axis, serving as the initial reference architecture for subsequent reflection behavior modeling.

[0057] During image acquisition, a micro-polarization marker is introduced into each frame. This polarization information is injected by placing an adjustable polarizer in front of the acquisition lens and switching the polarization angle at a certain frequency, achieving minimal perturbation tracking of the image's light vector direction. To ensure the physical recognizability of the polarization injection, the polarization angle should transition slowly within a specific range, for example, switching the polarization direction every 5 frames, to avoid feature confusion caused by frequent changes in image content. During this process, the image acquisition frame rate is kept stable at over 60 frames per second to ensure sufficient temporal resolution between polarization changes and the actual acquired content. During image acquisition, each frame is accompanied by a corresponding polarization direction record, and a one-to-one correspondence is established between the time stamp and the image frame number. This allows subsequent analysis to capture the polarization sensitivity of the reflection response in the image from a microscopic perspective. This polarization marking mechanism provides a fundamental support for identifying the correspondence between bright reflection areas and their polarization response characteristics in the next step of analysis.

[0058] Based on dual-baseline image data, frame-by-frame image content analysis is performed to extract dynamic trajectory features related to reflection artifacts. The analysis first traverses the continuous image sequence in chronological order, and performs pixel-level grayscale difference processing on bright spot regions in each frame to determine the center position, shape boundaries, and grayscale distribution changes of the bright spots. The spatial movement trajectory, grayscale fluctuation amplitude, and edge sharpness variation trend of each bright spot in consecutive frames are recorded as core features. Based on this, for each bright spot region, the response differences under different polarization injection states are further analyzed to quantify whether phenomena such as grayscale enhancement, edge sharpening, or deformation instability exist under specific polarization directions. Through this frame-by-frame cumulative analysis, a set of bright spot trajectories spanning the entire sequence is formed, and the significance of reflection interference is comprehensively evaluated based on the continuity of the trajectory, the frequency of boundary changes, and grayscale stability. This process not only extracts the dynamic feature sequence of reflection behavior but also provides raw data support for constructing time anchors.

[0059] Based on the bright spot trajectories and boundary sharpness variation trends extracted from frame-by-frame images, a set of temporal reference points for illumination changes and reflection behavior is constructed. Specifically, key frame locations where each bright spot exhibits enhanced continuity, intensified boundary blurring, or abrupt grayscale fluctuations in the temporal dimension are marked as temporal reference points. Simultaneously, frame numbers with significant differences in polarization response are combined to select key nodes that are representative and valuable for decision-making regarding image reflection behavior. These temporal reference points will serve as a unified alignment benchmark for subsequent image processing steps such as reflection decomposition, blood flow feature extraction, and artifact correction. The constructed set of temporal reference points possesses three key characteristics: temporal continuity, bright spot response independence, and polarization sensitivity, providing stable support for the regularity identification of artifact behavior in subsequent image sequences, causal determination of interference sources, and temporal fusion of multidimensional data.

[0060] S002, based on the set of time reference points, performs optical reflection decomposition on the previous observation image sequence, separates the specular reflection component and diffuse reflection component under the unified light field modeling framework, constructs a dual-domain feature spectrum of reflection domain features and blood flow response domain features, and generates an artifact risk map containing high reflection interference regions based on the dual-domain feature spectrum.

[0061] This step, based on the established set of time reference points, further decomposes the optical reflection behavior of the previously acquired image sequences. Through a unified light field modeling framework, it separates and analyzes specular and diffuse reflection, extracting feature representations of different reflection behaviors and blood flow responses to generate an artifact risk map with precise annotations of reflection interference locations. The specific steps are as follows:

[0062] Based on the constructed set of time reference points, image frames at corresponding time nodes are preferentially analyzed. The image sequence is read frame by frame, and the local illumination environment and pixel structure are modeled and analyzed with the time reference points as the core. Each image frame is divided into multiple analysis sub-regions, with the boundary range covered by the bright spot trajectory as the core region. By comparing the pixel grayscale changes and boundary sharpness before and after the reference point, the changes in reflection behavior are classified into layers. In this process, by calculating the intensity change rate and contour edge change density of the bright spot region in the temporal neighborhood, regions exhibiting high-frequency grayscale oscillations, edge blur expansion, and strong correlation with polarization direction are initially classified as specular reflection-dominant regions, while regions exhibiting low-frequency fluctuations, clear and stable contours, and insensitive polarization response are classified as diffuse reflection-dominant regions. Each type of region is spatially labeled separately to provide target localization basis for the subsequent physical separation of reflection components.

[0063] Based on the partitioning results, and combined with a unified light field modeling framework, physical decoupling of specular and diffuse reflection components is performed. During light field modeling, the reflection intensity distribution map of the observation area under different incident angles and polarization directions is reconstructed, and the observed light intensity variation pattern is matched with the reflection behavior type. The specular reflection component exhibits strong directionality, steep changes, and high consistency with the light source angle in space, thus representing a high-gradient, highly directional energy concentration region in the model. In contrast, the diffuse reflection component presents as an isotropic, uniformly distributed region significantly affected by skin structure. After light field modeling is completed, image pixels are grouped according to the distribution patterns of each reflection component in the modeling map, and light intensity remapping is performed at the pixel level, thereby achieving spatial separation of specular and diffuse reflection. This processing method not only physically distinguishes the two reflection behaviors at the image level but also lays the structural foundation for feature spectrum construction.

[0064] After spatial decoupling of specular and diffuse reflection components, a dual representation structure of reflection behavior feature spectrum and blood flow response feature spectrum is further constructed. The construction of the reflection feature spectrum relies on the gray-level distribution characteristics, edge variation patterns, and polarization response morphology of the specular reflection region. The extracted features include brightness peaks, edge sharpness indices, reflection symmetry indices, and temporal stability coefficients. The construction of the blood flow response feature spectrum relies on the dynamic change trajectory of blood vessel morphology in non-reflection interference regions, mainly including the consistency of blood vessel orientation, periodic gray-level fluctuation amplitude, flow continuity parameters along the time axis, and cross-sectional gray-level distribution gradient. Both feature spectra are spatially mapped at the pixel level, forming a one-to-one correspondence of two-dimensional matrix sets. This is used to identify the intersection of reflection and blood flow features in any region of the image, thereby determining whether there is a risk of visual confusion in that region.

[0065] Based on the dual-domain feature spectrum, an artifact risk map of high-reflectivity interference areas is generated. During map construction, the image regions are graded for interference sensitivity according to the overlap between specular reflection features and blood flow response features. Regions with highly concentrated specular reflection indicators, drastic edge changes, and simultaneous occurrence of unstable flow signals in the blood flow response spectrum are identified as potentially high-risk artifact regions and assigned high-intensity markings. Regions with blurred boundaries but stable blood flow signals are marked as medium-risk regions. Regions with significant differences between the two and no overlap are classified as low-risk or no-risk regions. Finally, an artifact risk map covering the entire image is formed. This map visualizes the risk level at the pixel level and aligns its coordinates with keyframes in the time reference point set, providing quantitative support and location reference for subsequent artifact source tracing, counterfactual simulation, and interference suppression. Through this series of refined reflection behavior decompositions and dual-domain feature extractions, accurate identification of high-reflectivity interference regions and complete generation of basic data for risk warning are achieved.

[0066] S003, based on the artifact risk map, performs a temporal counterfactual replay process that includes changes in illumination angle and skin surface humidity disturbances, verifies the stability of blood vessel morphology, grayscale and boundary features in the image under various environmental disturbance conditions, removes transient texture features that change significantly under disturbance conditions, and extracts a set of true blood flow evidence that shows consistent performance under multiple time series.

[0067] This step, based on the generated artifact risk map, conducts environmental disturbance simulations and image playback processes across multiple time series. A counterfactual playback strategy is used to verify the stability of vascular image features, further filtering out transient interference features lacking temporal consistency. Finally, a set of image evidence that can be identified as genuine blood flow manifestations is extracted. The specific steps are as follows:

[0068] Based on image regions marked as high-risk and medium-risk in the artifact risk map, key areas containing specular reflection interference, diffuse reflection fluctuations, and boundary blurring expansion phenomena were selected as interference-sensitive areas. Performance samples of these areas under different lighting conditions were extracted from the corresponding image sequences, and the lighting angle variation sequence was reconstructed to perform environmental perturbation simulation. Specifically, by controlling the rotation angle of the adjustable light source array in the acquisition direction, at least ten lighting angle levels were set, gradually transitioning from a low incident angle close to the skin surface to a vertical incident angle. At each angle level, images of the interference-sensitive areas were re-acquired under stable illumination conditions. To further enhance the perturbation effect, the epidermal humidity level of the image acquisition area was adjusted. Different humidification states were formed on the skin surface by controlling a small amount of water vapor spray, including slight humidification, moderate water film, and near-saturation. A complete interference combination was formed under each humidity level combined with different lighting angles. Based on this, the image acquisition sequence was reconstructed to ensure sufficient perturbation sample coverage of reflection interference behavior in the temporal dimension during the acquisition process.

[0069] For the reconstructed multi-temporal image sequence, the vascular-related features in each frame were individually labeled and their changing trends analyzed, mainly focusing on three dimensions: first, the geometric stability of the vascular morphology, including the direction of the main vascular trunk, the connectivity of branch structures, and the magnitude of changes in the curvature of the centerline; second, the stability of gray-level distribution, i.e., the changing trends of parameters such as the average gray-level value, gray-level gradient, and gray-level contrast of the vascular region across multiple frames; and third, the clarity index of the boundary contour, focusing on analyzing whether there are obvious blurring, breakage, or deformation phenomena of the vascular edges under multiple illumination and humidity disturbances. In this process, image frames from different perturbation combinations were compared one-to-one, recording the performance differences of the same region under changes in illumination and humidity parameters. Using representative vascular regions as templates, their feature change trajectories throughout the entire temporal perturbation sequence were summarized to construct a stability analysis table, and regions that maintained basic consistency in structure and gray-level features under most perturbation conditions were marked as highly stable regions, while those that did not were considered perturbation-sensitive regions.

[0070] Based on stability calibration, transient texture features were removed from the identified perturbation-sensitive regions. By tracking the grayscale fluctuation curves and edge structure evolution of these regions in the image sequence, bright spots, dark spots, or abnormal patches that suddenly appear under certain angles or humidity conditions were identified. These features are usually accompanied by drastic changes in local grayscale values ​​and instantaneous morphological expansion, lacking stability in temporal continuity. Therefore, in multi-frame comparative analysis, any image feature that appears and disappears irregularly in adjacent perturbation frames, lacks structural continuity, and cannot be reproduced under other conditions was identified as transient texture. These transient texture features were labeled as temporary interference in the image space and removed through image content masking and pixel replacement. Simultaneously, their corresponding spatial location and time point were recorded in the original image to avoid affecting the subsequent identification and verification of blood flow features.

[0071] After removing transient interference, vascular feature regions exhibiting high consistency and temporal coherence under various perturbation conditions were extracted from the stability analysis results. The image feature sets of these regions were then compiled as a set of true blood flow evidence. The extracted content included the temporal consistency curve of the main vascular trunk direction, the standard deviation sequence of grayscale distribution, edge stability scores, and response curves to changes in illumination direction and surface humidity. During the construction process, it was ensured that each region included in the true evidence set possessed structural reproducibility and image feature similarity across perturbation conditions, while regions exhibiting severe deformation or grayscale deviations from the average range under any combination of perturbations were excluded. The final set of true blood flow evidence not only possesses traceability and verifiability under multiple environmental perturbations but also lays a physically and logically supported data foundation for subsequent causal relationship verification and high-risk area identification, achieving accurate extraction of stable features and effective isolation of temporary interferences in vascular images.

[0072] S004, based on a set of real blood flow evidence, constructs a joint consistency judgment criterion of morphological distribution, gray intensity and temporal evolution, performs a causal verification process, achieves precise location of high-risk suspicious areas through fine-grained threshold sequences, compresses the time window triggered by erroneous identification, and limits the response boundary of high-risk output;

[0073] This step, building upon the extraction of the true blood flow evidence set, further conducts a precise causality verification process by constructing a joint consistency judgment criterion for image features. It also combines temporal scale and spatial structural changes to perform fine-grained localization of high-risk suspicious areas, thereby limiting the response boundary of risk output and improving the accuracy and stability of the assessment. The specific implementation steps are as follows:

[0074] Based on the established set of real blood flow evidence, key feature parameters are normalized and integrated to construct a joint consistency judgment criterion based on image morphological distribution, grayscale intensity, and temporal evolution path. Regarding morphological distribution, a regional geometric distribution framework is constructed with the main vascular trunk direction, branch point topology, and connectivity map as the core. Regarding grayscale intensity, a regional contrast mapping model is formed based on the average grayscale value, grayscale gradient, and grayscale fluctuation range of different blood flow regions. Regarding temporal evolution, an evolutionary analysis structure including continuity scoring, trend assessment, and stability window identification is constructed using the trajectory of vascular morphological changes and grayscale stability in continuous image sequences as input. The feature matrices of these three dimensions are uniformly bound by spatial location and temporal frame index to form a joint judgment structure oriented towards multi-dimensional features. In this judgment structure, each image region is assigned a comprehensive consistency score, which reflects whether the region exhibits inherent coherence and physiological rationality under multiple perturbation conditions and time scales.

[0075] Based on the joint consistency judgment criteria, a causal verification process for image regions is conducted. This process uses time sequence as a clue to trace the entire trajectory of the appearance and evolution of each characteristic region in the real blood flow evidence set in the perturbation sequence, and performs frame-by-frame comparative analysis with high-risk regions in the artifact risk map. By establishing a causal chain between vascular morphological stability and reflection interference response, it identifies which regions' abnormal changes appear immediately after specific lighting or humidity perturbations and do not possess cross-temporal structural persistence. These regions are initially marked as potential artifact points, and then further confirmed item by item whether there are non-physiological jumps in grayscale intensity changes, and whether the edge structure exhibits optically abnormal expansion or breakage. Finally, regions that meet the following two conditions are selected: first, no obvious abnormality before perturbation, but significant non-physiological changes during perturbation; second, rapid recovery to the original state after perturbation. These regions are interference-induced image change regions with a high degree of causal indication and are key suspects for subsequent high-risk judgment.

[0076] For identified suspicious areas, a fine-grained threshold sequence is further introduced to perform high-precision localization processing. Specifically, the suspicious areas are divided into multi-scale windows within the image frame sequence. Each suspicious area is further subdivided into smaller frame segments along the time dimension, and the change curves of its morphology, grayscale, and evolution path are calculated in each frame segment. Simultaneously, multiple judgment thresholds based on statistically derived measured data are set, such as edge change rate threshold, grayscale jump amplitude threshold, and connectivity decrease threshold. The performance of each suspicious area within each sub-time window is compared item by item. When an area continuously exceeds two or more judgment thresholds in multiple frame segments, and its change trend shows a clear temporal synchronization relationship with illumination or humidity disturbance events, the area is considered to have high-risk characteristics. In terms of spatial localization, pixel-level boundary delineation is performed for each image area judged as high-risk, and the marking is maintained consistently throughout the entire image frame sequence to ensure accurate tracking of the risk area across multiple image frames.

[0077] After completing the spatiotemporal localization of high-risk suspicious areas, the scope of high-risk outputs is further limited by compressing the abnormal response window in the time dimension. Specifically, in the image frame sequence, a response time span analysis is performed on all located high-risk areas to identify the frame number of the first abnormal manifestation and the frame number of the abnormal termination. Combined with the actual trigger time and duration of the disturbance event, a high-risk response time window is constructed. Based on this, a stability check is performed on the image regions within each time window to ensure the continuity and consistency of risk manifestations within that time period. Regions that do not meet the requirement of continuous manifestation are further removed from the risk markers to reduce misjudgment interference. The final response boundary includes the joint limitation result of the spatial location range and the time window range. This result will serve as the reference basis for subsequent image processing, assessment output, and intervention suggestion generation, ensuring that the judgment of high-risk areas has clear boundaries, a controllable range, and high confidence throughout the assessment process. Through the above implementation process, based on fully utilizing the real blood flow evidence set, a highly accurate and verifiable risk area identification mechanism is constructed, realizing a complete closed-loop processing from image feature extraction to spatiotemporal causal determination.

[0078] S005, based on the defined risk response boundary, initiates the dynamic image control process. Through the phase conjugate polarization rotation scanning mechanism, it links the Hamiltonian variational model of the reversible time grid rearrangement structure and energy return path to adjust the optical incident angle and polarization state online, thereby achieving real-time suppression of reflection artifacts and steady-state enhancement of blood flow response signals. Finally, it generates a closed-loop preoperative anesthesia risk assessment result with adaptive interference suppression capability.

[0079] This step, based on the defined high-risk response boundary, further initiates a dynamic image modulation process. Combining optical control mechanisms and temporal structure adjustment methods, it suppresses reflection artifacts in real time and enhances the blood flow response signal in a steady state, ultimately outputting an adaptive pre-anesthesia risk assessment result. The specific steps are as follows:

[0080] Based on the defined high-risk response boundaries, optical modulation operations targeting image reflection behavior are activated in real time, initiating a polarization rotation scanning mechanism based on the phase conjugation principle. This mechanism adjusts the polarization state of the incident light wavefront by setting a continuously rotatable electronically controlled polarization adjuster, maximizing phase cancellation between the polarization direction of the incident light and the reflection direction of the identified specular reflection area. During execution, the polarization rotation device rotates with high precision along the optical axis, with each angular displacement accompanied by a complete image acquisition cycle, ensuring that the phase relationship between the reflected signal and the real blood flow signal is visualized at the image level. Simultaneously, by combining the previously extracted spatial angle information between the main vascular trunk and the reflection direction, real-time matching of the polarization angle with the target structure in the image is achieved, thereby intervening in the local phase reversal of the reflection path to achieve local reflection energy cancellation. This scanning process does not rely on static polarization setup but dynamically adjusts according to the image content, achieving continuous coverage and real-time adaptation, providing the basic optical field input conditions for downstream temporal structure adjustment.

[0081] Building upon the phase-conjugate polarization scanning process, a reversible temporal raster rearrangement structure is constructed, linked to the image acquisition structure in the temporal dimension, to meet the dynamic reconstruction requirements of image content. This structure uses the high-risk response time periods identified in the acquired frame sequence as its backbone. High-interference frames from the original image sequence are arranged in reverse order according to the perturbation time axis, and compensation frames acquired under low-interference conditions are inserted as structural adjustment nodes. This ensures image temporal continuity while temporally buffering key frames that may contain reflection artifacts. During execution, all compensation frames undergo timestamp recoding to create a consistent temporal index with preceding and following image frames, preventing temporal misalignment caused by rearrangement from interfering with blood flow continuity. The rearranged temporal raster can synchronize with the dynamic evolution of polarization states, thereby offsetting image structural shifts caused by reflection fluctuations in the temporal dimension, improving the structural stability of the true blood flow signal, and providing highly consistent input data for subsequent energy reconstruction.

[0082] After the temporal raster structure is rearranged, an energy return path adjustment process based on Hamiltonian variational theory is initiated. This process aims to guide the return of abnormally concentrated energy regions caused by artifacts in the image, reducing their impact on evaluation metrics. Based on the light field intensity distribution, this process extracts pixel blocks in image frames whose brightness exceeds twice the standard deviation of the global average, and identifies their spatial overlap with marked vascular paths in the set of real blood flow evidence. By establishing a path energy propagation model, the shortest path propagation direction of energy in image space is analyzed, and the optimal return point is constructed using Hamiltonian variational method, guiding high-energy pixels back to low-interference areas with the lowest offset loss path. Physically, this return path is a combination of multiple curve segments and nodes, and its directional distribution is constrained by the local structure of the image. In image rendering, this energy transfer manifests as the gradual convergence of bright reflective areas to boundary areas, while the core vascular area recovers its proper grayscale and edge representation, achieving an adaptive balance of light field energy. This processing can be applied continuously to multiple image sequences, achieving self-stabilizing repair of bright areas across time sequences.

[0083] After completing the triple processes of optical modulation, temporal structure rearrangement, and energy path adjustment, a risk assessment is performed on the processed image sequence to generate a closed-loop pre-anesthesia risk assessment result with adaptive interference suppression capabilities. This assessment is based on the previously constructed joint consistency judgment criteria, re-extracting indicators such as vascular morphology, grayscale stability, and flow coherence from the current image structure and comparing them item by item with the image parameters in the initial unmodified state. For image regions exhibiting high consistency, high connectivity, and low reflection interference intensity after polarization modulation and temporal rearrangement, their risk level is downgraded according to a threshold standard; while for regions still exhibiting signal discontinuities, structural jumps, or abnormal energy residues, the original risk level label is retained or enhanced. The final output is presented as a multi-region risk level identification map, possessing spatial distribution accuracy, temporal response coherence, and optical interference adaptive capabilities. It can be directly used as a reference for anesthesiologists' pre-operative intervention decisions, achieving the goal of data-driven dynamic and refined risk management.

[0084] This invention achieves refined modeling and temporal alignment of reflection behavior under varying illumination and skin conditions by introducing a time reference point and micro-polarization markers. Then, through optical decomposition and feature spectrum construction, specular reflection and blood flow signals are effectively separated. Based on this, a counterfactual replay mechanism simulates the impact of different perturbation conditions on image features, accurately identifying and eliminating unstable transient artifact structures and extracting highly consistent real blood flow features. Furthermore, through causal verification and fine-grained thresholding strategies, high-risk suspicious points are precisely located and their response ranges are limited. Finally, by combining phase-conjugate polarization rotation scanning and energy return path modulation, real-time dynamic suppression of reflection artifacts and enhanced extraction of blood flow response signals are achieved, forming a robust evaluation result with closed-loop control. The overall solution not only improves the adaptability to optical interference during image processing but also enhances the generalization ability of the risk identification mechanism to complex clinical conditions, providing more reliable and intelligent data support for preoperative decision-making.

[0085] This invention provides, for example Figure 2 The anesthesia preoperative risk assessment system shown includes a light observation baseline module, a reflectance decomposition modeling module, a counterfactual playback verification module, a causal consistency determination module, and a dynamic regulation assessment module.

[0086] The illumination observation baseline module establishes a joint observation baseline for illumination and skin condition across time scales, injects micro-polarization markers into the image acquisition chain, and extracts bright spot trajectories and boundary sharpness curves through frame-by-frame image analysis to construct a set of time reference points for reflection behavior.

[0087] The reflection decomposition modeling module performs optical reflection decomposition on the observed image sequence based on a set of time reference points. Under a unified light field model, it separates specular reflection and diffuse reflection components, generates a dual-domain feature spectrum of the reflection domain and the blood flow response domain, and constructs an artifact risk map.

[0088] The counterfactual playback verification module uses the artifact risk map to perform temporal counterfactual playback, simulates the light angle and surface humidity disturbance, removes transient textures, and extracts a set of consistent blood flow evidence across multiple time series.

[0089] The causal consistency determination module, based on the real blood flow evidence set, constructs morphological, grayscale and temporal consistency criteria, performs causal verification, locates high-risk suspicious points through fine-grained thresholds, and limits the response boundary of risk output;

[0090] The dynamic control and assessment module, based on the risk response boundary, performs dynamic image control. Through phase conjugate polarization rotation scanning linked with reversible time grid rearrangement and energy return path, it adjusts the optical incident and polarization state online to achieve artifact suppression and blood flow signal enhancement, and outputs closed-loop controlled preoperative risk assessment results.

[0091] The preoperative anesthesia risk assessment method based on multi-source data fusion provided in this embodiment of the invention is implemented through the above-mentioned preoperative anesthesia risk assessment system based on multi-source data fusion. For details of the specific methods and procedures of the preoperative anesthesia risk assessment system based on multi-source data fusion, please refer to the embodiments of the preoperative anesthesia risk assessment method based on multi-source data fusion described above, which will not be repeated here.

[0092] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for preoperative anesthesia risk assessment based on multi-source data fusion, characterized in that, Includes the following steps: S001, establish a joint observation baseline for illumination and skin condition across time scales, inject micro-polarization markers into the image acquisition chain, extract bright spot trajectories and boundary sharpness curves through frame-by-frame image analysis, and construct a set of time reference points for reflection behavior; S002, based on a set of time reference points, performs optical reflectance decomposition on the observed image sequence, separates specular reflection and diffuse reflection components under a unified light field model, generates a dual-domain feature spectrum of the reflection domain and the blood flow response domain, and constructs an artifact risk map; S003 utilizes artifact risk maps to perform temporal counterfactual replay, simulating illumination angle and surface humidity disturbances, eliminating transient textures, and extracting a set of consistent real blood flow evidence across multiple time series. S004, based on a set of real blood flow evidence, constructs a consistency criterion for morphology, grayscale and temporal sequence, performs causal verification, locates high-risk suspicious points through fine-grained thresholds, and limits the response boundary of risk output; S005, based on the risk response boundary, performs dynamic image control. Through phase conjugate polarization rotation scanning linked with reversible time grid rearrangement and energy return path, it adjusts the optical incident and polarization states online to achieve artifact suppression and blood flow signal enhancement, and outputs closed-loop controlled preoperative risk assessment results.

2. The method for preoperative anesthesia risk assessment based on multi-source data fusion according to claim 1, characterized in that, Step S001 includes: Before image acquisition, the lighting conditions are adjusted and the incident light angle change sequence is set to form a two-dimensional dynamic acquisition baseline with the principal axis of illumination and the auxiliary axis of skin condition change. During image acquisition, micro-polarization markers are injected, and the polarization direction is switched at a fixed frequency using an adjustable polarizer, thus establishing a time correspondence between image frames and polarization directions. Based on dual-baseline image data, the image sequence is analyzed frame by frame to extract bright spot trajectories, grayscale fluctuations and boundary sharpness features, and dynamic annotation is performed in combination with polarization response differences; Based on the changing trends of image features and polarization differences, key frame positions are selected, and a set of time reference points for illumination changes and reflection behavior is constructed for time alignment and reflection interference calibration in the image processing process.

3. The method for preoperative anesthesia risk assessment based on multi-source data fusion according to claim 1, characterized in that, Step S002 includes: Based on the set of time reference points, local illumination modeling is performed on the image frames at the corresponding time nodes, the image is divided into multiple analysis sub-regions, and the regions are divided into specular reflection-dominated regions and diffuse reflection-dominated regions according to the characteristics of bright spot trajectory changes. By combining a unified light field modeling framework, physical decoupling is performed on different reflection components in the image, and pixel-level spatial separation of specular reflection components and diffuse reflection components is completed based on the reflection intensity distribution law. After completing the spatial separation of the reflection components, the reflection behavior feature spectrum and the blood flow response feature spectrum are constructed respectively, and a two-dimensional matrix set is established in pixels to identify the intersection region of reflection and blood flow. Based on the results of dual-domain feature spectrum analysis, an artifact risk map is constructed, and the interference risk level of image regions is marked for reference in subsequent artifact processing.

4. The preoperative anesthesia risk assessment method based on multi-source data fusion according to claim 3, characterized in that, The spatial separation of the specular reflection component is determined based on the reflection intensity gradient, edge sharpness, and consistency with the illumination direction, while the spatial separation of the diffuse reflection component is determined based on the uniformity of grayscale distribution and structural stability.

5. The method for preoperative anesthesia risk assessment based on multi-source data fusion according to claim 1, characterized in that, Step S003 includes: Based on the artifact risk map, sensitive areas of interference were selected, and the image acquisition sequence was reconstructed under different lighting angles and skin humidity conditions to form a complete perturbation combination group. Based on the reconstructed image sequence, the stability of blood vessel morphology, gray-level distribution and boundary contour features is calibrated, and regions that remain consistent under most perturbation conditions are identified as high-stability regions. Transient texture recognition is performed on regions with poor stability to remove bright spots, dark spots, or abnormal edge structures that lack coherence and reproducibility in perturbed frames; We extract real blood flow features from highly stable regions and construct a set of real blood flow evidence with cross-perturbation consistency and temporal coherence for subsequent causal verification.

6. The method for preoperative anesthesia risk assessment based on multi-source data fusion according to claim 5, characterized in that, Image features in the set of true evidence of blood flow include the consistency curve of blood vessel orientation, the standard deviation sequence of grayscale distribution, edge sharpness score, and the response trend curves under changes in illumination angle and skin humidity.

7. The method for preoperative anesthesia risk assessment based on multi-source data fusion according to claim 1, characterized in that, Step S004 includes: Based on a set of real blood flow evidence, a joint consistency judgment criterion of image morphological distribution, gray intensity and temporal evolution path is constructed, and a comprehensive consistency score is assigned to each image region. Based on the overall consistency score, causal verification is carried out to identify interference-induced image regions that show abnormal changes after the disturbance and return to their original state after the disturbance ends as high-risk suspicious areas; For high-risk suspicious areas, a fine-grained threshold sequence is introduced to compare changes in morphology, grayscale, and evolution path according to time-divided frame intervals; After locating high-risk areas, the anomaly response time window is compressed, and the response boundary for risk output is defined in conjunction with the spatial location range, for reference in subsequent assessment and intervention.

8. The method for preoperative anesthesia risk assessment based on multi-source data fusion according to claim 7, characterized in that, The defined response boundary is determined by continuously reviewing the time period between the first abnormal frame and the termination frame in the high-risk suspicious area, eliminating areas that do not meet the consistency performance, and retaining only the image areas that continuously show high consistency change characteristics throughout the entire disturbance process for risk assessment.

9. The method for preoperative anesthesia risk assessment based on multi-source data fusion according to claim 1, characterized in that, Step S005 includes: Based on the defined high-risk response boundary, a phase conjugate polarization rotation scanning mechanism is initiated to dynamically adjust the polarization state of the incident light in order to counteract the interference of specular reflection energy. A reversible temporal raster rearrangement structure is constructed by linking image frame sequences, reversing the order of high-interference frames and inserting low-interference compensation frames. Based on time rearrangement, an energy return path adjustment process based on the Hamiltonian variational principle is initiated to guide the energy of the bright area to return to the low interference area in order to achieve image energy balance. Blood flow-related indicators are re-extracted from the modulated image sequence, and adaptive interference suppression assessment results are output for preoperative risk level determination.

10. A pre-anesthesia risk assessment system based on multi-source data fusion, used to implement the pre-anesthesia risk assessment method based on multi-source data fusion as described in any one of claims 1-9, characterized in that, It includes a baseline illumination observation module, a reflectance decomposition modeling module, a counterfactual playback verification module, a causal consistency determination module, and a dynamic regulation evaluation module. The illumination observation baseline module establishes a joint observation baseline for illumination and skin condition across time scales, injects micro-polarization markers into the image acquisition chain, and extracts bright spot trajectories and boundary sharpness curves through frame-by-frame image analysis to construct a set of time reference points for reflection behavior. The reflection decomposition modeling module performs optical reflection decomposition on the observed image sequence based on a set of time reference points. Under a unified light field model, it separates specular reflection and diffuse reflection components, generates a dual-domain feature spectrum of the reflection domain and the blood flow response domain, and constructs an artifact risk map. The counterfactual playback verification module uses the artifact risk map to perform temporal counterfactual playback, simulates the light angle and surface humidity disturbance, removes transient textures, and extracts a set of consistent blood flow evidence across multiple time series. The causal consistency determination module, based on the real blood flow evidence set, constructs morphological, grayscale and temporal consistency criteria, performs causal verification, locates high-risk suspicious points through fine-grained thresholds, and limits the response boundary of risk output; The dynamic control and assessment module, based on the risk response boundary, performs dynamic image control. Through phase conjugate polarization rotation scanning linked with reversible time grid rearrangement and energy return path, it adjusts the optical incident and polarization state online to achieve artifact suppression and blood flow signal enhancement, and outputs closed-loop controlled preoperative risk assessment results.

Images