A flap perfusion state dynamic evaluation and survival prediction method and device for DFU repair

Abstract

This invention discloses a method and apparatus for dynamic assessment of flap perfusion status and survival prediction for DFU (disease-induced fibrosis) repair, belonging to the field of medical data processing technology. The method simultaneously acquires continuous blood flow perfusion and surface temperature video streams of the flap, performs spatiotemporal alignment based on cardiac cycle characteristics to generate a multimodal registration dataset; decomposes the blood flow sequence into time-frequency components to reconstruct low-frequency vasoconstriction signals to generate an absolute perfusion matrix, and calculates the conduction phase difference to construct a phase delay matrix; calculates the cross-correlation extreme values ​​of the thermal radiation sequence and vasoconstriction signals to generate a heat flow cross-correlation matrix; aggregates the above matrices to construct a microcirculation assessment tensor, locates and filters out spurious perfusion zones to extract the effective perfusion boundary; extracts the topological evolution vector of boundary contraction, fuses the tensor decay gradient to calculate the ischemic failure tendency matrix, performs spatiotemporal state extrapolation calculations, and calibrates the flap necrosis spread trajectory and crisis prediction time nodes. The apparatus is used to implement the above method.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, specifically to a method and apparatus for dynamic assessment of flap perfusion status and survival prediction for DFU repair. Background Technology

[0002] Diabetic foot ulcers are chronic, non-healing wounds caused by long-term local tissue damage. Flap transplantation is a routine surgical intervention for wound coverage and repair, preserving limb function. However, due to the prevalence of peripheral microvascular sclerosis and microcirculatory abnormalities in diabetic patients, microvascular network reconstruction after flap transplantation is extremely challenging. The evolution of postoperative perfusion status directly affects the survival rate of the transplanted tissue. Microvascular crisis leading to ischemia or congestion can easily cause partial or full-thickness necrosis of the flap, increasing the risk of secondary amputation. Therefore, continuous monitoring and evaluation of the postoperative blood supply reconstruction status in the flap area is crucial for improving the success rate of repair surgery.

[0003] Existing flap perfusion monitoring techniques typically utilize optical angiography or thermal imaging equipment to acquire static absolute blood flow images or instantaneous temperature cross-sections of the flap surface. In the pathological microenvironment of diabetic foot, the smooth muscle contraction function of microvessels is impaired, often resulting in abnormal opening of arteriovenous shunts. Large amounts of blood flow directly back to the skin without passing through the capillary network for metabolic exchange. Current methods, relying directly on the magnitude of static blood perfusion to assess flap viability, may erroneously identify high blood flow signals generated by arteriovenous shunts as good tissue blood supply, leading to significant judgment bias. Furthermore, existing technologies often use independent time-point data comparisons, exhibiting low sensitivity to spatiotemporal changes in internal tissue heat flow and vascular impedance conduction. This results in a lag effect in predicting the evolution of flap necrosis, significantly compressing the clinical window for vascular exploration or spasmolytic resuscitation. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a method and apparatus for dynamic evaluation of flap perfusion status and survival prediction for DFU repair to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for dynamic assessment of flap perfusion status and survival prediction for DFU repair, comprising the following steps: S1, simultaneously acquiring continuous blood flow perfusion video stream and surface temperature video stream of the target flap, extracting physiological cardiac cycle features to perform spatiotemporal alignment mapping on the dual-band video streams, generating a multimodal registration dataset; S2, performing time-frequency decomposition on the blood flow sequence in the multimodal registration dataset, reconstructing the low-frequency vasoconstriction signal and extracting the time-averaged blood flow to generate an absolute perfusion matrix; using the flap vascular pedicle as the zero-phase origin, calculating the phase difference of low-frequency vasoconstriction signal transmission pixel by pixel, constructing a phase delay matrix; S3, extracting the thermal radiation fluctuation sequence in the multimodal registration dataset, and calculating the mutual correlation between the thermal radiation fluctuation sequence and the low-frequency vasoconstriction signal at the same coordinate through a sliding window. The extreme values ​​are used to generate a heat flow cross-correlation matrix; the channel aggregation absolute perfusion matrix, phase delay matrix and heat flow cross-correlation matrix are used to construct a microcirculation assessment tensor; the feature clusters in the microcirculation assessment tensor where absolute perfusion exceeds the baseline threshold, cross-correlation extreme values ​​are lower than the coordination threshold and conduction phase difference exceeds the impedance threshold are located as pseudo-perfusion regions, a spatial mask is constructed to filter out pseudo-perfusion regions, and the topological outer edge of the remaining connected domain is extracted as the effective perfusion boundary; S4, the topological evolution vector of the effective perfusion boundary shrinking towards the zero phase origin is extracted along the time axis, the temporal decay gradient of the microcirculation assessment tensor and the topological evolution vector are fused, and the ischemic failure tendency matrix of the flap mesh is calculated; based on the spatial distribution gradient and temporal evolution rate of the ischemic failure tendency matrix, spatiotemporal state extrapolation calculation is performed to calibrate the spatial spread trajectory of flap necrosis and the time node of crisis prediction.

[0006] In a preferred embodiment, the specific process of simultaneously acquiring continuous blood flow perfusion video stream and surface temperature video stream of the target flap, extracting physiological cardiac cycle features, and performing spatiotemporal alignment mapping on the dual-band video streams to generate a multimodal registration dataset is as follows: extracting common anatomical edge contour feature points from the continuous blood flow perfusion video stream and surface temperature video stream, and performing pixel-level spatial registration by solving the spatial affine transformation matrix; extracting physiological cardiac cycle features to strip the pulse wave AC component from the continuous blood flow perfusion video stream, locating the extreme coordinates of the systolic peak of the pulse wave AC component, and setting the extreme coordinates of the systolic peak as the time axis anchor point; performing interpolation resampling time alignment on the surface temperature video stream based on the time axis anchor point; and cascading the multidimensional feature channels to spatially register and temporally align the continuous blood flow perfusion video stream and surface temperature video stream to generate a multimodal registration dataset.

[0007] In a preferred embodiment, the specific process of performing time-frequency decomposition on the blood flow sequence in the multimodal registration dataset, reconstructing the low-frequency vasomotor signal, and extracting the time-mean value of blood flow to generate an absolute perfusion matrix is ​​as follows: A continuous wavelet transform is performed on the blood flow sequence in the multimodal registration dataset along the time axis to map the time-frequency energy distribution spectrum; cardiogenic and respiratory frequency bands are configured as stopband masks to filter out physiological cardiopulmonary noise components; wavelet coefficients within the endothelial-myogenic frequency band are extracted, and an inverse wavelet transform is performed to reconstruct the low-frequency vasomotor signal; a physiological monitoring sliding window is set, and the time-domain integral mean of the blood flow sequence within the physiological monitoring sliding window is calculated as the time-mean value of blood flow, and an absolute perfusion matrix is ​​generated by topological mapping according to a unified spatial coordinate system.

[0008] In a preferred embodiment, the specific process of constructing a phase delay matrix by calculating the conduction phase difference of the low-frequency contraction and relaxation signal pixel by pixel, with the flap vascular pedicle as the zero-phase origin, is as follows: extract the low-frequency contraction and relaxation signal corresponding to the zero-phase origin as a reference sequence; perform Hilbert transform on the low-frequency contraction and relaxation signals in the reference reference sequence and the remaining spatial coordinates in the multimodal registration dataset to solve the analytical signals; separate the instantaneous phase components of the analytical signals, perform phase dewinding processing on the instantaneous phase components to map the continuous phase flow; calculate the orthogonal projection difference between the continuous phase flow in the remaining spatial coordinates and the continuous phase flow in the reference reference sequence as the conduction phase difference, and spatially map the conduction phase difference pixel by pixel to generate the phase delay matrix.

[0009] In a preferred embodiment, the specific process of extracting the thermal radiation fluctuation sequence from the multimodal registration dataset and calculating the cross-correlation extreme values ​​between the thermal radiation fluctuation sequence and the low-frequency contraction signal at the same coordinate using a sliding window to generate the heat flow cross-correlation matrix is ​​as follows: The low-frequency baseline drift trend term of the thermal radiation fluctuation sequence is filtered out; the thermal radiation fluctuation subsequence and the low-frequency contraction subsequence at the same coordinate are simultaneously extracted within the sliding time window along the time axis; the discrete cross-correlation function of the thermal radiation fluctuation subsequence and the low-frequency contraction subsequence is calculated; the cross-correlation peak points in the hysteresis domain of the discrete cross-correlation function are extracted as cross-correlation extreme values; and the cross-correlation extreme values ​​are mapped according to the spatial topology to construct the heat flow cross-correlation matrix by traversing the unified spatial coordinate system.

[0010] In a preferred embodiment, the construction logic of constructing the microcirculation evaluation tensor by channel aggregation of the absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix is ​​as follows: extract the spatial resolution features of the absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix, perform bilinear interpolation resampling and scale alignment; perform orthogonal stacking of the scale-aligned absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix according to the preset feature dimension axis; solve the covariance distribution of the orthogonal stacked features and perform whitening transformation to eliminate feature dimension differences, and output the microcirculation evaluation tensor.

[0011] In a preferred embodiment, the specific process of identifying pseudo-perfusion regions in the microcirculation evaluation tensor as feature clusters whose absolute perfusion exceeds the baseline threshold, whose cross-correlation extrema are lower than the coordination threshold, and whose conduction phase difference exceeds the impedance threshold, and constructing a spatial mask to filter out pseudo-perfusion regions, and extracting the outer edge of the remaining connected domain topology as the effective perfusion boundary is as follows: The microcirculation evaluation tensor is sliced ​​along the feature dimension axis to extract spatial pixel multidimensional feature vectors; each dimension component of the multidimensional feature vector is compared with the baseline threshold, coordination threshold, and impedance threshold respectively, and the comparison results are used to perform Boolean logic intersection operations to solve the spatial mapping coordinates of the feature clusters to locate the pseudo-perfusion regions; a spatial mask is constructed based on the spatial mapping coordinates of the feature clusters using binarized state elements, and the Hadamard product operation is performed between the spatial mask and the microcirculation evaluation tensor to zero out the pseudo-perfusion region signal; discrete noise points are stripped and zeroed out through morphological opening operations, and gradient operator convolution is performed on the outer contour of the remaining connected domain to locate ridge lines and extract the effective perfusion boundary.

[0012] In a preferred embodiment, the specific process of extracting the topological evolution vector of the effective perfusion boundary shrinking towards the zero-phase origin along the time axis, fusing the temporal decay gradient of the microcirculation assessment tensor with the topological evolution vector, and solving the ischemic failure tendency matrix of the flap mesh is as follows: solving the spatial deformation difference of the effective perfusion boundary in adjacent sampling periods to extract the normal displacement of the boundary nodes and generate the topological evolution vector; calculating the partial derivative with respect to the time dimension of the microcirculation assessment tensor to solve the temporal decay gradient distribution matrix, calculating the vector projection of the topological evolution vector towards the zero-phase origin to quantize the shrinkage characteristics; extracting the vector projection and the temporal decay gradient distribution matrix and performing element-level inner product operations on a grid-by-grid basis to map the ischemic failure tendency matrix.

[0013] In a preferred embodiment, the specific process of performing spatiotemporal state extrapolation calculations based on the spatial distribution gradient and temporal evolution rate of the ischemic failure tendency matrix to calibrate the spatial spread trajectory of flap necrosis and the time node of crisis prediction is as follows: The spatial distribution gradient field is calculated by taking the directional derivative of the ischemic failure tendency matrix in the spatial coordinate system, and the temporal evolution rate is calculated by taking the partial derivative of the ischemic failure tendency matrix in the time axis; a first-order partial differential state evolution equation is constructed by simultaneously solving the spatial distribution gradient field and the temporal evolution rate; Runge-Kutta numerical integration is performed on the first-order partial differential state evolution equation according to a preset extrapolation step size to update the grid pathological state value; the contour line offset path of the grid pathological state value in the numerical integration iteration domain that touches the critical state of necrosis is tracked to generate the spatial spread trajectory of flap necrosis; the crisis prediction time node is calculated based on the cumulative number of iteration steps that the contour line offset path crosses the specified anatomical key points, combined with the extrapolation step size.

[0014] A device for dynamic assessment and survival prediction of flap perfusion status for DFU repair, used to execute the aforementioned method for dynamic assessment and survival prediction of flap perfusion status for DFU repair, includes: a registration unit for simultaneously acquiring continuous blood flow perfusion video stream and surface temperature video stream of the target flap, extracting physiological cardiac cycle features to perform spatiotemporal alignment mapping on the dual-band video streams, and generating a multimodal registration dataset; a phase calculation unit for performing time-frequency decomposition on the blood flow sequence in the multimodal registration dataset, reconstructing low-frequency vasoconstriction signals and extracting the time-mean value of blood flow to generate an absolute perfusion matrix; calculating the phase difference of low-frequency vasoconstriction signal transmission pixel by pixel with the flap vascular pedicle as the zero phase origin, and constructing a phase delay matrix; and a tensor fusion unit for extracting the thermal radiation fluctuation sequence in the multimodal registration dataset, and calculating the thermal radiation fluctuation sequence and its corresponding phase using a sliding window. The cross-correlation extremes of low-frequency contraction and relaxation signals are used to generate a heat flow cross-correlation matrix. The channel aggregation absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix are used to construct a microcirculation assessment tensor. The feature clusters in the microcirculation assessment tensor where absolute perfusion exceeds the baseline threshold, cross-correlation extremes are below the coordination threshold, and conduction phase difference exceeds the impedance threshold are identified as pseudo-perfusion regions. A spatial mask is constructed to filter out pseudo-perfusion regions, and the topological outer edge of the remaining connected domains is extracted as the effective perfusion boundary. A state extrapolation unit is used to extract the topological evolution vector of the effective perfusion boundary shrinking towards the zero-phase origin along the time axis. The temporal decay gradient of the microcirculation assessment tensor and the topological evolution vector are fused to solve the ischemic failure tendency matrix of the flap mesh. Based on the spatial distribution gradient and temporal evolution rate of the ischemic failure tendency matrix, spatiotemporal state extrapolation calculation is performed to calibrate the spatial spread trajectory of flap necrosis and the time node of crisis prediction.

[0015] The technical effects and advantages of this invention are as follows:

[0016] (1) A method for dynamic assessment of flap perfusion status and survival prediction for DFU repair involves acquiring continuous blood flow perfusion video streams and surface temperature video streams of the target flap, extracting physiological cardiac cycle features, and performing spatiotemporal alignment mapping on the dual-band video streams to generate a multimodal registration dataset. The blood flow sequences in the multimodal registration dataset are decomposed in time and frequency to reconstruct low-frequency vasoconstriction signals, and the time-averaged blood flow is extracted to generate an absolute perfusion matrix. A phase delay matrix is ​​constructed by calculating the phase difference of low-frequency vasoconstriction signal transmission pixel by pixel, with the flap vascular pedicle as the zero-phase origin. This method eliminates physiological interference from individual patient heart rhythm and respiratory fluctuations, transforming discrete multi-source vital signs into a microvascular impedance and hemodynamic feature matrix under a unified spatiotemporal reference. This provides deep physical data support for subsequent identification of microcirculatory abnormalities caused by local vascular lesions in diabetic patients, overcoming the problem of single feature dimension caused by relying solely on instantaneous single-modal imaging in existing technologies.

[0017] (2) A device for dynamic assessment and survival prediction of flap perfusion status for DFU repair. The thermal radiation fluctuation sequence is extracted and the cross-correlation extreme value of the low-frequency contraction and relaxation signal in the same coordinate is calculated by sliding window to generate a thermal flow cross-correlation matrix. The channel aggregates the absolute perfusion matrix, phase delay matrix and thermal flow cross-correlation matrix to construct a microcirculation assessment tensor. The feature clusters with absolute perfusion exceeding the baseline threshold, cross-correlation extreme value below the coordination threshold and conduction phase difference exceeding the impedance threshold are located as pseudo-perfusion areas. A spatial mask is constructed to filter out pseudo-perfusion areas and extract the topological outer edge of the remaining connected domain as the effective perfusion boundary. The topological evolution vector of the effective perfusion boundary shrinking towards the zero phase origin is extracted along the time axis. The ischemic failure tendency matrix is ​​calculated by fusing the temporal decay gradient of the microcirculation assessment tensor and the topological evolution vector. The spatiotemporal state extrapolation calculation is performed based on the spatial distribution gradient and the temporal evolution rate to calibrate the spatial spread trajectory of flap necrosis and the time node of crisis prediction. Therefore, it can accurately separate and filter out the spurious high-perfusion zero-metabolism microcirculation short-circuit unique to diabetes, lock the real capillary material exchange boundary, and perform physical state extrapolation by relying on attenuation gradient and topological evolution rate. This solves the problem that the prediction of existing technologies lags behind the appearance of tissue blackening and necrosis, and improves the advance warning of clinical crisis and the spatial accuracy of lesion trajectory prediction.

[0018] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0019] Figure 1 This is a flowchart of a method for dynamic assessment of flap perfusion status and survival prediction for DFU repair according to the present invention.

[0020] Figure 2 This is a schematic diagram of the time-series distribution of physiological signal time-frequency decomposition and extraction provided in an embodiment of the present invention;

[0021] Figure 3 A schematic diagram of a three-dimensional mapping between a microcirculation evaluation tensor and the spatial localization of feature clusters in a pseudo-perfusion zone, provided as an embodiment of the present invention;

[0022] Figure 4 This is a schematic diagram illustrating the boundary topological evolution and necrosis propagation trajectory based on the extrapolation of the spatiotemporal state of ischemic failure, as provided in an embodiment of the present invention.

[0023] Figure 5 This is a flowchart of a flap perfusion status dynamic assessment and survival prediction device for DFU repair according to the present invention. Detailed Implementation

[0024] This application provides a method and device for dynamic assessment of flap perfusion status and survival prediction for DFU repair, which solves the problems of existing flap monitoring technologies, such as difficulty in identifying false perfusion artifacts caused by arteriovenous shunting and severe lag in predicting flap ischemia and necrosis.

[0025] The overall approach of the scheme in this application embodiment is as follows: Simultaneously acquire video streams of blood flow and surface temperature in the flap repair area for physiological spatiotemporal alignment mapping; perform time-frequency decomposition on the blood flow sequence to reconstruct low-frequency contraction and relaxation signals; and combine this with spatial coordinates to calculate the conduction phase difference of the low-frequency contraction and relaxation signals to generate a phase delay matrix; extract the cross-correlation extreme values ​​of the surface thermal radiation fluctuation sequence and the low-frequency contraction and relaxation signals, and construct a microcirculation assessment tensor by combining absolute perfusion volume and conduction phase difference; locate and filter out pseudo-perfusion areas with high perfusion volume, low cross-correlation, and high conduction delay characteristics, and reconstruct the effective perfusion boundary of the remaining connected regions; extract the topological evolution vector of the effective perfusion boundary contracting over time and the temporal decay gradient of the tensor to generate a flap mesh ischemia-failure tendency matrix; and perform spatiotemporal state extrapolation calculations based on the spatial distribution gradient and temporal evolution rate of the ischemia-failure tendency matrix to determine the spatial spread trajectory of flap necrosis and the time node for crisis prediction.

[0026] Example 1; please refer to Figure 1 This invention provides a technical solution: a method for dynamic assessment of flap perfusion status and survival prediction for DFU repair, comprising the following steps: S1, simultaneously acquiring continuous blood flow perfusion video stream and surface temperature video stream of the target flap, extracting physiological cardiac cycle features to perform spatiotemporal alignment mapping on the dual-band video streams, generating a multimodal registration dataset; S2, performing time-frequency decomposition on the blood flow sequence in the multimodal registration dataset, reconstructing low-frequency vasoconstriction signals and extracting the time-mean value of blood flow to generate an absolute perfusion matrix; using the flap vascular pedicle as the zero-phase origin, calculating the phase difference of low-frequency vasoconstriction signal transmission pixel by pixel, constructing a phase delay matrix; S3, extracting the thermal radiation fluctuation sequence in the multimodal registration dataset, and calculating the cross-correlation between the thermal radiation fluctuation sequence and the low-frequency vasoconstriction signal at the same coordinate through a sliding window. The values ​​are used to generate a heat flow cross-correlation matrix; the channel aggregation absolute perfusion matrix, phase delay matrix and heat flow cross-correlation matrix are used to construct a microcirculation assessment tensor; the feature clusters in the microcirculation assessment tensor where absolute perfusion exceeds the baseline threshold, cross-correlation extreme values ​​are lower than the coordination threshold and conduction phase difference exceeds the impedance threshold are located as pseudo-perfusion regions, a spatial mask is constructed to filter out pseudo-perfusion regions, and the topological outer edge of the remaining connected domain is extracted as the effective perfusion boundary; S4, the topological evolution vector of the effective perfusion boundary shrinking towards the zero phase origin is extracted along the time axis, the temporal decay gradient of the microcirculation assessment tensor and the topological evolution vector are fused, and the flap mesh ischemia failure tendency matrix is ​​calculated; based on the spatial distribution gradient and temporal evolution rate of the ischemia failure tendency matrix, spatiotemporal state extrapolation calculation is performed to calibrate the spatial spread trajectory of flap necrosis and the crisis prediction time node.

[0027] In this implementation scheme, step S1 mainly performs the acquisition of basic physiological signals and low-level registration operations. The multimodal registration dataset refers to a structured data volume that integrates optical signals from blood flow monitoring equipment and temperature signals from infrared thermal imaging equipment into the same digital space. The dual-band video streams involved in this stage correspond to the optical band reflecting vascular physical perfusion and the infrared band reflecting tissue metabolic heat production, respectively. The spatiotemporal alignment mapping operation uses the extracted physiological cardiac cycle features as physical time anchors to resample and synchronize the two independently acquired video streams on the time axis sequence. Simultaneously, pixel-level affine calibration is performed on the spatial coordinate system to ensure that the blood flow and temperature states acquired in subsequent analyses originate entirely from the same microcirculatory anatomical space and are at the same physiological transient cross-section.

[0028] Step S2 is responsible for separating the dynamic characteristic parameters characterizing the true function of the flap's microvessels. Time-frequency decomposition filters out high-frequency noise of cardiac and respiratory origin from the original blood flow sequence. The low-frequency systolic and diastolic signals represent the spontaneous low-frequency rhythmic contraction and relaxation activities of microvascular smooth muscle and endothelial cells, which often exhibit conduction abnormalities in the sclerotic microvessels of diabetic foot patients. The zero-phase origin is set at the anatomical starting point where the feeding aorta enters the flap. The phase delay matrix is ​​used to quantify the time lag in the reception of this rhythmic signal at each spatial coordinate of the flap. The conduction phase difference calculated pixel by pixel directly reflects the pathological impedance encountered by blood as it travels from the vascular pedicle to the distal microvascular network. An abnormal increase in the conduction phase difference parameter indicates that the local vascular channel is in a state of stiffness or severe obstruction.

[0029] Step S3 primarily involves removing the microcirculatory short-circuit artifacts unique to diabetic foot pathology to map the true viable boundary of the flap's material exchange. The heat-fluid cross-correlation matrix is ​​used to quantify the synergistic relationship between local microcirculatory blood flow and tissue metabolic heat production; the lower the extreme value of the internal cross-correlation, the more likely blood is flowing through the area without effective oxygen and heat exchange. The microcirculation assessment tensor is a multidimensional data set formed by orthogonally concatenating the absolute perfusion matrix, phase delay matrix, and heat-fluid cross-correlation matrix along its characteristic dimensions. The pseudo-perfusion zone corresponds to a non-functional intracirculatory region in diabetic foot characterized by high blood flow, high impedance, and low metabolism due to abnormal opening of arteriovenous short circuits. The spatial mask acts as a digital filter, completely zeroing out the pseudo-perfusion zone signal; the remaining connected regions represent viable tissue blocks with genuine capillary material exchange capabilities, and their outermost physical contour constitutes the current effective perfusion boundary of the flap.

[0030] Step S4 performs a spatial and temporal extrapolation from the known microcirculation deterioration trend to the future ischemic necrosis state. The topological evolution vector records the geometric displacement distance and directional characteristics of the effective perfusion boundary gradually receding towards the vascular pedicle reference direction over time. The temporal decay gradient represents the partial derivative values ​​of the multidimensional feature parameters within the microcirculation assessment tensor as the monitoring period progresses. The ischemic failure tendency matrix integrates the macroscopic boundary contraction dimension and the microscopic feature deterioration dimension, mapping the rate at which each spatial grid of the flap approaches the physiological necrosis threshold. The spatiotemporal state extrapolation calculation relies on the spatial distribution derivative within the ischemic failure tendency matrix and the temporal evolution rate to construct an evolution equation. By performing numerical integration operations on the unoccurred monitoring domain, the expected physical region of microcirculation collapse and the crisis outbreak time node are output as a visualized spatial spread trajectory.

[0031] Specifically, the process of simultaneously acquiring continuous blood flow perfusion video streams and surface temperature video streams of the target flap, extracting physiological cardiac cycle features, and performing spatiotemporal alignment mapping on the dual-band video streams to generate a multimodal registration dataset is as follows: Common anatomical edge contour feature points are extracted from the continuous blood flow perfusion video stream and the surface temperature video stream; the spatial affine transformation matrix is ​​calculated to perform pixel-level spatial registration; physiological cardiac cycle features are extracted to remove the pulse wave AC component from the continuous blood flow perfusion video stream; the extreme coordinates of the systolic peak of the pulse wave AC component are located, and the extreme coordinates of the systolic peak are set as the time axis anchor point; interpolation resampling is performed on the surface temperature video stream based on the time axis anchor point; the multi-dimensional feature channels are cascaded, and the continuous blood flow perfusion video stream and the surface temperature video stream after spatial registration and time alignment are generated to produce a multimodal registration dataset.

[0032] In this implementation scheme, the anatomical edge morphology of the images from the two video streams is extracted frame by frame, and a scale-invariant feature transformation algorithm is used to filter out the common edge contour feature point set for matching. The process of solving the spatial affine transformation matrix is ​​based on the weighted least squares method, by minimizing the spatial registration error function. Achieve; where, The value represents the spatial registration error function; K represents the total number of successfully matched common anatomical edge feature point pairs. This represents the spatial alignment weight coefficient; This represents the horizontal pixel coordinates of the k-th feature point in the surface temperature video stream. This represents the vertical pixel coordinates of the k-th feature point in the surface temperature video stream; Let denote the third-order affine transformation matrix in space to be solved; This represents the horizontal pixel coordinates of the k-th feature point in a continuous blood perfusion video stream; This represents the vertical pixel coordinates of the k-th feature point in a continuous blood flow perfusion video stream. The spatial alignment weight coefficient is also included. The logic of its establishment is In the formula, This represents the local gradient magnitude of the neighborhood of the pixel corresponding to the k-th feature point; This represents the maximum local gradient magnitude among all common anatomical edge feature points extracted within the entire image. Based on the obtained... Linear algebraic spatial mapping is performed on the pixel coordinates within the surface temperature video stream to achieve pixel-level rigid registration of the dual-band images under a unified anatomical spatial reference. In the temporal dimension, moving average filtering is applied to the temporal fluctuations of perfusion intensity at each pixel location within the spatially registered continuous blood flow perfusion video stream to extract the pulse wave AC component reflecting the cardiac ejection cycle. In the formula, This indicates the amplitude of the separated AC component of the pulse wave; Indicates the pixel location at time point The intensity of the raw continuous blood flow perfusion signal collected at the site; Represents a continuous-time variable; The value represents the length of the baseline extraction filter sliding window; d represents the discrete-time window offset parameter. The establishment logic is as follows: for a pre-recorded initial segment The sequence segment is subjected to a Fast Fourier Transform (FFT) to extract the first dominant frequency peak within its spectral distribution, and the reciprocal of this first dominant frequency peak is calculated as... The possible values ​​of . For The sequence performs zero-crossing point determination of the first derivative and negative extreme value determination of the second derivative, locking the extreme coordinates of the systolic peak where arterial blood filling is greatest as the time axis anchor point. Using this time axis anchor point as a reference, a cubic spline interpolation resampling model is performed on the surface temperature video stream. In the formula, This represents the output value of the surface temperature signal after interpolation and resampling; M represents the total number of original sampled frames of the surface temperature video stream that participated in the interpolation. This represents the sampled value of the original discrete surface temperature in the m-th frame; m represents the index number of the original surface temperature sampling frame. Represents the physical timestamp of the original surface temperature sample in the m-th frame; Denotes the basis functions for cubic spline interpolation; This represents the alignment time difference scalar between the continuous blood perfusion video stream and the surface temperature video stream at the time axis anchor point; This represents the preset target resampling interval step size for the multimodal registration dataset. Subsequently, perfusion intensity values ​​and temperature values ​​with the same spatial cell coordinates and equal-step physical timestamps are directly concatenated and cascaded along the depth channel dimension of the feature matrix to encapsulate and generate a multidimensional registration dataset.

[0033] Specifically, the process of performing time-frequency decomposition on the blood flow sequence in the multimodal registration dataset, reconstructing the low-frequency vasomotor signal, and extracting the time-mean value of blood flow to generate the absolute perfusion matrix is ​​as follows: A continuous wavelet transform is performed on the blood flow sequence in the multimodal registration dataset along the time axis to map the time-frequency energy distribution spectrum; cardiogenic and respiratory frequency bands are configured as stopband masks to filter out physiological cardiopulmonary noise components; wavelet coefficients within the endothelial and myogenic frequency bands are extracted, and inverse wavelet transform is performed to reconstruct the low-frequency vasomotor signal; a physiological monitoring sliding window is set, and the time-domain integral mean of the blood flow sequence within the physiological monitoring sliding window is calculated as the time-mean value of blood flow, and the absolute perfusion matrix is ​​generated by topological mapping according to a unified spatial coordinate system.

[0034] In this implementation scheme, continuous wavelet transform is performed on the blood flow sequence of each spatial pixel in the multimodal registration dataset along the time axis. The one-dimensional time-domain signal is mapped to a two-dimensional time-frequency energy distribution spectrum using complex Morlet wavelet basis functions. The calculation process is described as follows: In the formula, represents the two-dimensional time-frequency wavelet coefficient matrix after continuous wavelet transform; s represents the scaling parameter of the wavelet transform; This represents the time shift parameter of the wavelet transform; A continuous representation function representing the discrete blood flow sequence of the original input; Represents the independent variable in continuous time; This represents the complex conjugate form of the complex Morlet wavelet basis function. Subsequently, a frequency stopband mask is constructed to filter out high-frequency fluctuating blood flow noise, and the stopband mask function is defined. Its establishment method is to determine the scaling parameter. The inverse mapping frequency band of the signal is defined such that the stopband mask function takes a value of one when it falls within the boundary of the endothelial-muscular frequency band, and a value of zero when it is in interfering frequency bands such as the cardiac and respiratory frequency bands. The stopband mask function is directly multiplied by the time-frequency wavelet coefficient matrix to extract the wavelet coefficients within the desired frequency band, and an inverse wavelet transform is performed to reconstruct the low-frequency systolic and diastolic signal. In the formula, This represents the reconstructed low-frequency systolic and diastolic signal after filtering out physiological cardiopulmonary noise components. This represents the admissibility constant of the corresponding wavelet basis function; This represents the complex Morlet wavelet basis function. Furthermore, a physiological monitoring sliding window is set to extract the mean value of blood flow, and the integral mean is obtained by numerical accumulation. In the formula, Represents the horizontal coordinate in space and the vertical axis The absolute infusion value mapped at the location; This indicates the total sampling span encompassed by the physiological monitoring sliding window; represents the absolute intensity quantization value of the discrete blood flow sampling signal; w represents the discrete sample sequence number within the physiological monitoring sliding window. The absolute perfusion values ​​of each pixel obtained by solving are traversed throughout the unified spatial coordinate system to perform a gridded topological mapping, thereby generating the absolute perfusion matrix.

[0035] Specifically, taking the flap vascular pedicle as the zero-phase origin, the process of calculating the low-frequency contraction-extension signal transmission phase difference pixel by pixel and constructing the phase delay matrix is ​​as follows: extract the low-frequency contraction-extension signal corresponding to the zero-phase origin as the reference sequence; perform Hilbert transform on the low-frequency contraction-extension signals in the reference reference sequence and the other spatial coordinates in the multimodal registration dataset to solve the analytical signals; separate the instantaneous phase components of the analytical signals, perform phase dewinding processing on the instantaneous phase components to map the continuous phase flow; calculate the orthogonal projection difference between the continuous phase flow in the other spatial coordinates and the continuous phase flow in the reference sequence as the transmission phase difference, and spatially map the transmission phase difference pixel by pixel to generate the phase delay matrix.

[0036] In this implementation scheme, after determining the two-dimensional coordinates of the flap vascular pedicle end through anatomical topological features, the low-frequency contraction / extension signal generated at the zero-phase origin position is extracted as the benchmark reference sequence for global alignment. The low-frequency contraction / extension signals generated at each of the other spatial coordinates in the multimodal registration dataset are then analyzed using Cauchy principal value integrals and Hilbert transforms to construct the analytical signals, resulting in the following structural formula: In the formula, This represents the complex analytic signal generated by the real-domain transformation; Represents the imaginary unit constant; This represents the local time integral element independent variable introduced when calculating the Cauchy principal value integral. Subsequently, the ratio of the real and imaginary parts of the analytical signal is used to separate the instantaneous phase components. In the formula, This represents the instantaneous phase component that is folded into the principal value interval by the inverse trigonometric function; This indicates the quantization operator for extracting the imaginary part of a complex number; This represents the quantization operator for extracting the real part of a complex number. Phase dewinding is performed on the instantaneous phase component to map a continuous phase flow. Specifically, the absolute amplitude of phase jumps at adjacent discrete time steps is detected. When the jump amplitude exceeds the jump discrimination threshold, a periodic compensation calculation is performed. The calculation equation is as follows: In the formula, This represents the continuous phase flow after unwinding and unfolding. The term represents the order of the coiling and shaping correction, which is accumulated successively in the positive and negative directions of the phase jump. The aforementioned jump discrimination threshold is established by extracting the maximum natural fluctuation amplitude of the spontaneous low-frequency oscillation phase of a healthy skin flap sample without pathological features in a resting state as the quantitative basis for this threshold. Finally, the cosine distance between the continuous phase flow and the continuous phase flow of the reference sequence in other spatial coordinates is calculated and converted into an orthogonal projection difference as the propagated phase difference. The calculation process is as follows: In the formula, The scalar value of the propagated phase difference, representing the calculated spatial coordinate position of the target; This represents the total discrete sequence length of the continuous phase-flow signal sequence within the analysis window; z represents the discrete-time count index for the step progression of the phase-flow sequence. This represents the continuous phase flow sequence samples generated by reconstruction and expansion in the remaining two-dimensional spatial coordinates; This represents a continuous phase flow sequence sample corresponding to the origin position reference sequence. The spatial phase difference obtained from traversal calculations is spatially filled and mapped, and then combined and encapsulated to generate a phase delay matrix characterizing the microcirculatory conduction resistance of blood pulsation in the flap tissue. Combined with... Figure 2 As shown, in the temporal distribution of physiological signal time-frequency decomposition and extraction, the original acquired blood perfusion signal contains multi-frequency fluctuations. The pulse wave AC component, after baseline extraction and filtering isolation, exhibits a curve with periodic maxima, and its systolic peak is extracted and established as a unified time axis anchor point. Simultaneously, the low-frequency systolic and diastolic signal reconstructed by continuous wavelet transform and inverse wavelet transform filters out cardiac and respiratory interference bands, characterizing the low-frequency autonomous rhythm of microvascular smooth muscle. This low-frequency systolic and diastolic signal is used as a reference benchmark for subsequent extraction of conduction phase delay and construction of the phase delay matrix.

[0037] Specifically, the process of extracting the thermal radiation fluctuation sequence from the multimodal registration dataset, calculating the cross-correlation extreme values ​​between the thermal radiation fluctuation sequence and the low-frequency contraction signal in the same coordinate using a sliding window, and generating the heat flow cross-correlation matrix is ​​as follows: The low-frequency baseline drift trend term of the thermal radiation fluctuation sequence is filtered out; the thermal radiation fluctuation subsequence and the low-frequency contraction subsequence in the same coordinate are simultaneously extracted within the sliding time window along the time axis; the discrete cross-correlation function of the thermal radiation fluctuation subsequence and the low-frequency contraction subsequence is calculated; the cross-correlation peak points in the hysteresis domain of the discrete cross-correlation function are extracted as cross-correlation extreme values; and the cross-correlation extreme values ​​are mapped according to the spatial topology to construct the heat flow cross-correlation matrix by traversing the unified spatial coordinate system.

[0038] In this implementation scheme, a low-order polynomial fitting is performed on the thermal radiation fluctuation sequence to filter out the low-frequency baseline drift trend term reflecting the interference of gradual changes in ambient temperature. The calculation logic is as follows: In the formula, This represents the discrete sequence of thermal radiation fluctuations after filtering out baseline drift. Indicates the index of discrete-time sampling points; This represents the original collected discrete signal value of thermal radiation fluctuations; This indicates the highest order of the polynomial fit. Let represent the coefficients of the polynomials solved using the least squares method. A fixed-length sliding time window is set on the time axis, and the thermal radiation fluctuation subsequence and the low-frequency contraction subsequence under the same coordinates are simultaneously extracted within this window and normalized to zero mean. The normalized subsequences are denoted as […]. and Solve for the discrete cross-correlation function of the two. In the formula, This represents the discrete cross-correlation function value under specific spatial coordinates and a specific number of lag steps. Represents the horizontal grid coordinates of the target pixel; Represents the vertical grid coordinates of the target pixel; This represents the number of discrete-time lag steps in the cross-correlation calculation process; This indicates the total number of sampling points included in the sliding time window; Indicates the sequence number of the local sampling point within the time window; This represents the normalized thermal radiation wave subsequence; This represents the normalized low-frequency condenser sequence. Extreme values ​​are extracted from the calculation results within the defined effective lag domain. In the formula, The extracted peak values ​​of cross-correlation are used as the extreme values ​​of cross-correlation. This represents the maximum physiological delay time step limit. Iterate through all pixels in the spatial coordinate system, filling in the obtained cross-correlation extrema according to their physical topological positions to generate a thermal cross-correlation matrix that quantifies the degree of coupling between blood flow and metabolism.

[0039] Specifically, the construction logic of the microcirculation evaluation tensor is as follows: extract the spatial resolution features of the absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix, perform bilinear interpolation resampling and scale alignment; orthogonally stack the scale-aligned absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix according to the preset feature dimension axis; solve the covariance distribution of the orthogonally stacked features and perform whitening transformation to eliminate feature dimension differences, and output the microcirculation evaluation tensor.

[0040] In this implementation scheme, the spatial size parameters of the absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix are extracted. Using the matrix with the highest resolution as a benchmark, a bilinear interpolation algorithm is used to resample the other two matrices, ensuring complete scale alignment of the three in the spatial resolution dimension. The scale-aligned matrices are then orthogonally stacked along the feature channel dimension to form an initial spatial feature tensor. To eliminate the dimensional differences and numerical magnitude interference caused by the three different physical characteristics—blood perfusion intensity, transmission phase time, and cross-correlation coefficient—a ZCA whitening transformation is performed on this tensor. First, the covariance distribution of the orthogonally stacked features is calculated. In the formula, The covariance matrix representing the orthogonal stacking features; This represents the total number of valid pixels within a unified spatial coordinate system. This represents the one-dimensional line index of the flattened pixel. Indicates the first The three-dimensional initial column vector features corresponding to each pixel; This represents the global statistical mean vector of the initial column vector features of all pixels. Subsequently, eigenvalue decomposition is performed on the covariance matrix, followed by whitening mapping. In the formula, This represents the dimensionless microcycle evaluation tensor vector output after whitening conversion; The matrix representing the orthogonal eigenvectors of the covariance matrix; This represents a diagonal matrix composed of eigenvalues. The reconstructed dimensionless dataset is the output microcyclic evaluation tensor.

[0041] Specifically, the process of identifying pseudo-perfusion regions in the microcirculation evaluation tensor where absolute perfusion exceeds the baseline threshold, cross-correlation extrema are below the coordination threshold, and conduction phase difference exceeds the impedance threshold, and constructing a spatial mask to filter out pseudo-perfusion regions, and extracting the outer edge of the remaining connected domain topology as the effective perfusion boundary, is as follows: The microcirculation evaluation tensor is sliced ​​along the feature dimension axis to extract multi-dimensional feature vectors of spatial pixels; each dimension component of the multi-dimensional feature vector is compared with the baseline threshold, coordination threshold, and impedance threshold respectively; the comparison results are used to perform Boolean logic intersection operations to solve the spatial mapping coordinates of the feature clusters to locate the pseudo-perfusion regions; a spatial mask is constructed based on the spatial mapping coordinates of the feature clusters using binarized state elements; the spatial mask and the microcirculation evaluation tensor are subjected to Hadamard product operations to zero out the pseudo-perfusion region signals; discrete noise points are stripped away and zeroed out through morphological opening operations; gradient operator convolution is performed on the outer contour of the remaining connected domains to locate ridges and extract the effective perfusion boundary.

[0042] In this implementation scheme, the microcirculation assessment tensor is sliced ​​along the feature dimension axis, and the dimensionless multidimensional feature vector components of absolute perfusion, cross-correlation extrema, and conduction phase difference contained in each spatial pixel are extracted. Baseline threshold, coordination threshold, and impedance threshold are set. The method for determining these three key thresholds is as follows: a calibration domain is defined in the mirror image region of the contralateral healthy foot of the non-flap repair area of ​​the homologous patient; the statistical mean of each component of the multidimensional feature vector within the calibration domain is extracted; and the mean of each component is superimposed and subtracted from the standard deviation of the corresponding component in the calibration domain as the comparison threshold parameters. Boolean logic intersection operation is then performed. In the formula, This represents the spatial mask value constructed by the binary state element; This represents the logical AND conditional operator; This represents the absolute perfusion component within the microcirculation assessment tensor; This indicates the set baseline threshold. This represents the cross-correlation extreme components within the microcirculation evaluation tensor; This indicates the set collaboration threshold; This represents the conduction phase difference component within the microcirculation evaluation tensor; This represents the set impedance threshold. When both logical conditions are met, the pixel is marked as a pseudo-injection region and the mask is assigned a value of one; otherwise, it is assigned a value of zero. The Hadamard product operation is then performed after subtracting the spatial mask from the matrix containing all one elements. In the formula, This represents the tensor matrix after the pseudo-infusion zone signal is set to zero; This represents the element-wise multiplication operator for the Hadamard product; Represents a one-dimensional matrix with the same dimension as the evaluation tensor; This represents the overall spatial mask matrix. Using morphological structuring elements, an opening operation (i.e., erosion followed by dilation) is performed on the binarized contours of the pure tensor matrix to remove independent discrete noise points and isolated small dead zones. For the remaining large-area connected region outer contours, a two-dimensional convolution is performed using the Sobel operator. In the formula, This represents the extracted contour gradient magnitude; This represents the convolution kernel of the horizontal gradient operator; Represents the remaining connected binary graph after morphological opening operation; The vertical gradient operator convolution kernel is represented by `*`; the spatial two-dimensional convolution operator is represented by `*`. The coordinates of the continuous closed pixels connected by the outermost ridge of the gradient magnitude are used as the effective perfusion boundary defining the survival and ischemia states. Combined with... Figure 3As shown, the microcirculation assessment tensor forms a specific data mapping distribution in three-dimensional space. This three-dimensional coordinate space is orthogonally constructed by absolute perfusion volume, cross-correlation extrema, and conduction phase difference. The feature points of the effective perfusion grid are distributed within the conventional area defined by the baseline threshold and the co-occurrence threshold. When an abnormal opening of the arteriovenous shunting occurs locally on the flap, the corresponding grid pixels aggregate in space to form a feature cluster with high absolute perfusion volume, low cross-correlation extrema, and high conduction phase difference. The system constructs a spatial mask through Boolean logic intersection operations to isolate and zero out the pseudo-perfusion area feature clusters within this specific spatial quadrant, thereby removing the microcirculation artifacts caused by non-nutritive blood flow.

[0043] Specifically, the process of extracting the topological evolution vector of the effective perfusion boundary shrinking towards the zero-phase origin along the time axis, fusing the temporal decay gradient of the microcirculation assessment tensor with the topological evolution vector, and solving the ischemic failure tendency matrix of the flap mesh is as follows: solving the spatial deformation difference of the effective perfusion boundary in adjacent sampling periods to extract the normal displacement of the boundary nodes and generate the topological evolution vector; calculating the partial derivative with respect to the time dimension of the microcirculation assessment tensor to solve the temporal decay gradient distribution matrix, calculating the vector projection of the topological evolution vector towards the zero-phase origin to quantize the shrinkage characteristics; extracting the vector projection and the temporal decay gradient distribution matrix and performing element-wise inner product operations on a grid-by-grid basis to map the ischemic failure tendency matrix.

[0044] In this implementation scheme, the effective infusion boundary coordinate set generated within two adjacent physical sampling periods is extracted, the nearest point iteration algorithm is used to spatially register the boundary contour nodes, and the normal displacement of the corresponding nodes is calculated to generate the topology evolution vector. In the formula, Indicates the first The sampling period of the first The topological evolution vectors of the boundary nodes; Indicates the sequence number of discrete nodes on the effective infusion boundary; Indicates the discrete-time index of the current sampling period; This represents the discrete-time index of the previous sampling period; Represents the exact geometric coordinate vector of the boundary node in two-dimensional space; This indicates that the node points to the unit normal vector within the tissue. Subsequently, the first-order forward difference approximation partial derivative is calculated along the time dimension axis of the microcirculation evaluation tensor to solve for the temporal decay gradient distribution matrix. In the formula, Indicates the coordinates of the spatial grid points at the horizontal index. With vertical index The numerical value of the temporal decay gradient distribution matrix at the location; This represents the extracted microcirculation assessment tensor multidimensional feature set. To quantify the physical characteristics of the boundary receding towards the vascular pedicle, the vector projection of the topological evolution vector toward the zero-phase origin is calculated. In the formula, A vector projection scalar representing the quantized shrinkage characteristics; This represents the coordinate vector of the zero-phase origin at the pre-established vascular pedicle insertion point; This represents the Euclidean norm solving operator. The projected scalar, existing only at boundary nodes, is generalized to the entire flap mesh using inverse distance weighted interpolation, and then element-wise inner product operations are performed with the temporally decaying gradient distribution matrix on a mesh-by-mesh basis. In the formula, This represents the ischemic failure tendency matrix output by the solution; This represents the tensor inner product operation; This represents the shrinkage feature projection scalar value interpolated to the target mesh through spatial generalization.

[0045] Specifically, the process of extrapolating the spatiotemporal state based on the spatial distribution gradient and temporal evolution rate of the ischemic failure tendency matrix to calibrate the spatial spread trajectory of flap necrosis and the predicted time node of crisis is as follows: The spatial distribution gradient field is calculated by taking the directional derivative of the ischemic failure tendency matrix in the spatial coordinate system, and the temporal evolution rate is calculated by taking the partial derivative of the ischemic failure tendency matrix in the time axis; a first-order partial differential state evolution equation is constructed by simultaneously solving the spatial distribution gradient field and the temporal evolution rate; Runge-Kutta numerical integration is performed on the first-order partial differential state evolution equation according to the preset extrapolation step size to update the grid pathological state value; the contour line offset path of the grid pathological state value in the numerical integration iteration domain that touches the critical state of necrosis is traced to generate the spatial spread trajectory of flap necrosis; the predicted time node of crisis is calculated based on the cumulative number of iteration steps that the contour line offset path crosses the specified anatomical key points, combined with the extrapolation step size.

[0046] In this implementation scheme, the directional derivative is obtained by applying the central difference method to each grid node within the ischemic failure tendency matrix, and the spatial distribution gradient field is calculated. In the formula, The gradient field vector representing the spatial distribution of the ischemic failure tendency matrix; and These represent the partial differential operators along the horizontal and vertical orthogonal directions, respectively. The ischemic failure tendency matrix within the historical sampling period is extracted synchronously, and its time evolution rate is calculated by polynomial fitting and differentiation along the time axis. In the formula, The time evolution rate parameter represents the acceleration of pathological deterioration. A first-order partial differential state evolution equation simulating pathological diffusion is constructed by simultaneously resolving the spatially distributed gradient field and the time evolution rate. In the formula, Indicates the virtual extrapolation evolution time. The grid pathological status value below; Indicates the spatial diffusion coefficient; Represents the time evolution gain coefficient. Spatial diffusion coefficient. With time evolution gain coefficient The method for establishing this baseline is as follows: Baseline calibration is performed by retrieving the natural variation patterns of blood flow in the patient's own healthy skin area, and the reciprocal of the blood flow fluctuation rate in the healthy area is used as... The normalized proportion of the conduction phase difference in the microcirculation assessment tensor exceeding the calibration threshold was used as... Set a fixed extrapolation time step. The fourth-order Runge-Kutta numerical analysis method is applied to perform discrete numerical integration on the partial differential state evolution equation, updating the mesh pathological state values ​​round by round. In the numerical integration iteration loop, the updated mesh pathological state values ​​are compared. When they reach the limiting scalar value corresponding to the critical necrosis state, the geometric mesh coordinates are recorded and connected to form a contour line offset path. The spatial envelope of this path, expanding with virtual time, represents the spatial spread trajectory of flap necrosis. The inward expansion process of the contour line offset path is detected. The integration loop terminates when the contour line physically penetrates pre-marked key anatomical points. Physical time is calculated based on the cumulative number of steps. In the formula, This indicates the final calculated criticality forecast timeline; This represents the total number of numerical integration iterations performed when the contour lines cross key points of the dissection; This indicates the preset discrete extrapolation time step; This represents the scaling constant that maps virtual integral time to the real physical clock. Combined with... Figure 4 As shown, the extrapolation of the spatiotemporal state of ischemic failure manifests as a dynamic evolution of the boundary topology. The initial effective boundary on the outer side represents the baseline contour established after the pseudo-perfusion zone is stripped away. Driven by Runge-Kutta numerical integration, the mesh pathological state values ​​are updated progressively with the virtual evolution time step, driving the extrapolated boundary to continuously contract towards the zero-phase origin of the vascular pedicle. Based on the mesh normal displacement and attenuation gradient calculated by the partial differential state evolution equation, the necrosis spread trajectory vector that runs through each level of contour lines is calibrated. When the contour line offset path of this spread trajectory touches the set anatomical key point, the associated extrapolated time step is output, which is converted into the crisis prediction time node.

[0047] Example 2; please refer to Figure 5A device for dynamic assessment and survival prediction of flap perfusion status for DFU repair is provided, used to execute the method for dynamic assessment and survival prediction of flap perfusion status for DFU repair described in the embodiments. The device includes: a registration unit for simultaneously acquiring continuous blood flow perfusion video stream and surface temperature video stream of the target flap, extracting physiological cardiac cycle features to perform spatiotemporal alignment mapping on the dual-band video streams, and generating a multimodal registration dataset; a phase calculation unit for performing time-frequency decomposition on the blood flow sequence in the multimodal registration dataset, reconstructing low-frequency vasoconstriction signals, and extracting the time-averaged blood flow to generate an absolute perfusion matrix; calculating the low-frequency vasoconstriction signal transmission phase difference pixel-by-pixel with the flap vascular pedicle as the zero-phase origin, and constructing a phase delay matrix; and a tensor fusion unit for extracting the thermal radiation fluctuation sequence in the multimodal registration dataset and calculating the thermal radiation fluctuation sequence through a sliding window. The heat flow cross-correlation matrix is ​​generated by cross-correlation extreme values ​​of low-frequency contraction and relaxation signals with the same coordinates. The microcirculation assessment tensor is constructed by channel aggregation of the absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix. The feature clusters in the microcirculation assessment tensor where absolute perfusion exceeds the baseline threshold, cross-correlation extreme values ​​are lower than the coordination threshold, and conduction phase difference exceeds the impedance threshold are identified as pseudo-perfusion regions. A spatial mask is constructed to filter out pseudo-perfusion regions, and the topological outer edge of the remaining connected domain is extracted as the effective perfusion boundary. A state extrapolation unit is used to extract the topological evolution vector of the effective perfusion boundary shrinking towards the zero-phase origin along the time axis. The temporal decay gradient of the microcirculation assessment tensor and the topological evolution vector are fused to solve the ischemic failure tendency matrix of the flap mesh. Spatiotemporal state extrapolation calculation is performed based on the spatial distribution gradient and temporal evolution rate of the ischemic failure tendency matrix to determine the spatial spread trajectory of flap necrosis and the time node of crisis prediction.

[0048] In this implementation scheme, the registration unit serves as the front-end data access and preprocessing module of the system, responsible for performing the spatiotemporal reference unification operation of multi-source signals in step S1. This unit synchronously receives dual-band raw video streams of optical blood flow and infrared temperature via a hardware interface, and uses its built-in digital signal processing logic to identify the systolic peak characteristics of the pulse wave as a physical time anchor. Based on this, the unit drives the processor to perform affine transformation mapping and interpolation resampling alignment operations on temperature and blood flow pixels, eliminating the inherent deviations in spatial perspective and sampling frequency between the two sensing devices. Finally, it encapsulates and generates a coordinate-aligned and time-synchronized multimodal registration dataset in memory, providing a unified physical spatial reference for subsequent calculations.

[0049] The phase resolution unit is responsible for performing hemodynamic frequency decomposition and impedance conduction quantization operations in step S2. After receiving the registration dataset, this unit uses continuous wavelet transform instructions to perform pixel-by-pixel frequency domain separation of the blood flow sequence, extracts specific frequency bands to reconstruct low-frequency vasomotor signals, and simultaneously calculates the mean value through a time-domain integrator accumulator to generate an absolute perfusion matrix. Subsequently, this unit extracts the two-dimensional coordinates of the vascular pedicle insertion end and solidifies them as a zero-phase reference node. It then calls the arithmetic unit to perform Hilbert transform and analytical signal phase extraction on the global data pixel by pixel, calculates the conduction phase delay of the low-frequency vasomotor signals of each grid relative to the reference node, and spatially stores the results to generate a phase delay matrix.

[0050] The tensor fusion unit is used to perform the pathological short-circuit artifact removal and real capillary exchange boundary reconstruction operations in step S3. This unit divides the storage space into sliding time windows, simultaneously reads thermal radiation fluctuations and low-frequency contraction / reduction sequences, and calls the coprocessor to perform discrete cross-correlation function operations to extract extreme values ​​and generate a thermal flow cross-correlation matrix. Next, the unit performs channel splicing and whitening transformation on the three matrices generated above to construct a multidimensional microcirculation assessment tensor. Further, the unit's built-in logic comparator performs out-of-limit comparisons between each dimension component of the tensor and the corresponding baseline, coordination, and impedance thresholds, outputting a binary spatial mask to remove pseudo-perfusion area pixels with high blood flow and low metabolism, and calls a morphological convolution operator to locate and extract the effective perfusion boundary in the remaining connected domain.

[0051] The state extrapolation unit is responsible for simulating the ischemic deterioration trend and calculating the crisis nodes in step S4. This unit reads boundary register data from adjacent sampling periods, calculates the normal displacement of the boundary coordinates to generate a topological evolution vector, and simultaneously solves the temporal decay gradient output by the tensor differentiation command. Through grid inner product operations, it maps the ischemic failure tendency matrix. Based on this, the unit initiates partial differential equation solving logic, extracts the spatial directional derivative and temporal axial partial derivative of the tendency matrix to construct a first-order state evolution equation, and performs Runge-Kutta numerical integration at a preset step size. This unit continuously tracks the contour coordinates of the state values ​​crossing the necrosis critical point within the integration iteration domain, maps the offset trajectory to a spatial spread map, and calculates and outputs the physical crisis prediction time node based on the accumulated steps of the iterative pedometer touching the anatomical key points.

[0052] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0053] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0054] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0055] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0056] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0057] 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 dynamic evaluation of flap perfusion status and survival prediction for DFU repair, characterized in that, Includes the following steps: S1. Simultaneously acquire continuous blood flow perfusion video stream and surface temperature video stream of the target flap, extract physiological cardiac cycle features to perform spatiotemporal alignment mapping on the dual-band video stream, and generate a multimodal registration dataset; S2. Perform time-frequency decomposition on the blood flow sequence in the multimodal registration dataset, reconstruct the low-frequency vasoconstriction signal, and extract the time-mean value of blood flow to generate an absolute perfusion matrix; take the flap vascular pedicle as the zero phase origin, solve the low-frequency vasoconstriction signal transmission phase difference pixel by pixel, and construct a phase delay matrix. S3. Extract the thermal radiation fluctuation sequence from the multimodal registration dataset, calculate the cross-correlation extreme values ​​between the thermal radiation fluctuation sequence and the low-frequency contraction signal at the same coordinate through a sliding window, and generate a heat flow cross-correlation matrix; construct a microcirculation evaluation tensor by channel aggregation of the absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix; locate the feature clusters in the microcirculation evaluation tensor where the absolute perfusion exceeds the baseline threshold, the cross-correlation extreme values ​​are lower than the coordination threshold, and the conduction phase difference exceeds the impedance threshold as pseudo-perfusion regions, construct a spatial mask to filter out pseudo-perfusion regions, and extract the topological outer edge of the remaining connected domain as the effective perfusion boundary; S4. Extract the topological evolution vector of the effective perfusion boundary shrinking towards the zero phase origin along the time axis, fuse the microcirculation assessment tensor temporal decay gradient and the topological evolution vector, and solve the flap mesh ischemia failure tendency matrix; perform spatiotemporal state extrapolation calculation based on the spatial distribution gradient and temporal evolution rate of the ischemia failure tendency matrix, and calibrate the spatial spread trajectory of flap necrosis and the crisis prediction time node. The specific process of performing time-frequency decomposition on the blood flow sequences in the multimodal registration dataset, reconstructing the low-frequency vasomotor signal, and extracting the time-mean value of blood flow to generate the absolute perfusion matrix is ​​as follows: Continuous wavelet transform is performed on the blood flow sequences in the multimodal registration dataset along the time axis to map the time-frequency energy distribution spectrum; Configure the cardiac and respiratory frequency bands as a band-blocking mask to filter out physiological cardiopulmonary noise components; Wavelet coefficients within the endothelial-myogenic frequency band are extracted, and inverse wavelet transform is performed to reconstruct the low-frequency vasomotor signal. A physiological monitoring sliding window is set up, and the mean of the time-domain integral of the blood flow sequence within the physiological monitoring sliding window is calculated as the blood flow time mean. An absolute perfusion matrix is ​​generated by topological mapping according to a unified spatial coordinate system. The specific process of constructing the phase delay matrix by calculating the low-frequency contraction and relaxation signal transmission phase difference pixel by pixel, with the flap vascular pedicle as the zero-phase origin, is as follows: Extract the low-frequency contraction / expansion signal corresponding to the zero-phase origin as a reference sequence; Hilbert transform was performed on the low-frequency scaling signals in the reference sequence and the other spatial coordinates in the multimodal registration dataset to solve the analytical signals. Separate the instantaneous phase components of the analytical signal, perform phase dewinding on the instantaneous phase components, and map a continuous phase flow; The orthogonal projection difference between the continuous phase flow and the continuous phase flow of the reference sequence in each of the remaining spatial coordinates is calculated as the conducted phase difference. The conducted phase difference is then spatially mapped pixel by pixel to generate a phase delay matrix.

2. The method for dynamic assessment of flap perfusion status and survival prediction for DFU repair according to claim 1, characterized in that: The specific process of simultaneously acquiring continuous blood flow perfusion video streams and surface temperature video streams of the target flap, extracting physiological cardiac cycle features, performing spatiotemporal alignment mapping on the dual-band video streams, and generating a multimodal registration dataset is as follows: Extract common anatomical edge contour feature points from the continuous blood perfusion video stream and the surface temperature video stream, and perform pixel-level spatial registration by solving the spatial affine transformation matrix. The physiological cardiac cycle features were extracted to separate the pulse wave AC component from the continuous blood flow perfusion video stream. The extreme coordinates of the systolic peak of the pulse wave AC component were located and set as the time axis anchor point. Interpolation and resampling time alignment are performed on the surface temperature video stream based on the time axis anchor point; A multimodal registration dataset is generated by cascading multidimensional feature channels, spatially registering and temporally aligning continuous blood perfusion video streams and surface temperature video streams.

3. The method for dynamic assessment of flap perfusion status and survival prediction for DFU repair according to claim 1, characterized in that: The specific process of extracting the thermal radiation fluctuation sequence from the multimodal registration dataset, calculating the cross-correlation extreme values ​​between the thermal radiation fluctuation sequence and the low-frequency contraction / expansion signal at the same coordinate using a sliding window, and generating the heat flow cross-correlation matrix is ​​as follows: Filter out the low-frequency baseline drift trend term of the thermal radiation fluctuation sequence, and simultaneously extract the thermal radiation fluctuation subsequence and the low-frequency contraction subsequence of the same coordinate within the sliding time window along the time axis. Solve the discrete cross-correlation function between the thermal radiation wave subsequence and the low-frequency contraction subsequence, and extract the cross-correlation peak points in the lag domain of the discrete cross-correlation function as the cross-correlation extreme values; By traversing the unified spatial coordinate system, the extreme values ​​of cross-correlation are mapped according to the spatial topology to construct the heat flow cross-correlation matrix.

4. The method for dynamic assessment of flap perfusion status and survival prediction for DFU repair according to claim 1, characterized in that: The construction logic for the microcirculation evaluation tensor, which is constructed from the channel aggregation absolute infusion matrix, phase delay matrix, and heat flux cross-correlation matrix, is as follows: Spatial resolution features of the absolute infusion matrix, phase delay matrix, and heat flux cross-correlation matrix are extracted and bilinear interpolation resampling scale alignment is performed. The scale-aligned absolute infusion matrix, phase delay matrix, and heat flow cross-correlation matrix are orthogonally stacked according to the preset feature dimension axis; Solve the covariance distribution of orthogonal stacked features and perform a whitening transformation operation to eliminate feature dimension differences, outputting a micro-cycle evaluation tensor.

5. A method for dynamic assessment of flap perfusion status and survival prediction for DFU repair according to claim 1 or 4, characterized in that: The specific process of identifying pseudo-perfusion regions in the microcirculation assessment tensor—characteristic clusters where absolute perfusion exceeds the baseline threshold, cross-correlation extrema are below the coordination threshold, and conduction phase difference exceeds the impedance threshold—is as follows: A spatial mask is constructed to filter out pseudo-perfusion regions, and the remaining connected domain topological outer edges are extracted as effective perfusion boundaries. Slicing along the feature dimension axis micro-loop evaluation tensor extracts spatial pixel multidimensional feature vectors; Each component of the multidimensional feature vector is compared with the baseline threshold, the collaborative threshold, and the impedance threshold. Boolean logic intersection is performed on the comparison results to solve the feature cluster space mapping coordinates to locate the pseudo-perfusion area. The configuration of the binary state element is based on the spatial mapping coordinates of the feature cluster to construct a spatial mask. The spatial mask and the microcirculation evaluation tensor are then subjected to the Hadamard product operation to set the pseudo-perfusion zone signal to zero. By stripping away discrete noise points with zeros through morphological opening operations, gradient operator convolution is performed on the outer contour of the remaining connected domain to locate ridges and extract effective perfusion boundaries.

6. The method for dynamic assessment of flap perfusion status and survival prediction for DFU repair according to claim 1, characterized in that: The specific process of extracting the topological evolution vector of the effective perfusion boundary shrinking towards the zero-phase origin along the time axis, fusing the temporal decay gradient of the microcirculation evaluation tensor with the topological evolution vector, and solving the ischemic failure tendency matrix of the flap mesh is as follows: Solve the spatial deformation difference of the effective perfusion boundary between adjacent sampling periods to extract the normal displacement of the boundary nodes and generate the topology evolution vector; The temporal decay gradient distribution matrix is ​​solved by taking the partial derivative with respect to the time dimension of the microcirculation evaluation tensor, and the vector projection quantization shrinkage characteristics of the topological evolution vector toward the zero-phase origin are calculated. Extract the vector projection and time-series decay gradient distribution matrix, perform element-wise inner product operations on a grid-by-grid basis to map the ischemic failure tendency matrix.

7. The method for dynamic assessment of flap perfusion status and survival prediction for DFU repair according to claim 6, characterized in that: Based on the spatial distribution gradient and temporal evolution rate of the ischemic failure tendency matrix, the spatiotemporal state extrapolation calculation is performed to calibrate the spatial spread trajectory of flap necrosis and the time nodes for crisis prediction. The specific process is as follows: The spatial distribution gradient field is calculated by taking the directional derivative of the ischemic failure tendency matrix in the spatial coordinate system, and the time evolution rate is calculated by taking the partial derivative of the ischemic failure tendency matrix in the time axis. A first-order partial differential state evolution equation is constructed by combining the spatially distributed gradient field and the temporal evolution rate. The grid pathological state value is updated by performing Runge-Kutta numerical integration on the first-order partial differential state evolution equation according to the preset extrapolation step size. Track the contour line offset path of the grid pathological state value in the numerical integration iteration domain when it reaches the critical state of necrosis, and generate the spatial spread trajectory of skin flap necrosis; The time node for crisis prediction is calculated by combining the cumulative number of iterations when the contour line offset path crosses the specified key dissection point with the extrapolation step size.

8. A device for dynamic assessment of flap perfusion status and survival prediction for DFU repair, used to execute the method for dynamic assessment of flap perfusion status and survival prediction for DFU repair as described in any one of claims 1-7, characterized in that, include: The registration unit is used to simultaneously acquire the continuous blood flow perfusion video stream and surface temperature video stream of the target flap, extract physiological cardiac cycle features to perform spatiotemporal alignment mapping on the dual-band video streams, and generate a multimodal registration dataset. The phase calculation unit is used to perform time-frequency decomposition on the blood flow sequence in the multimodal registration dataset, reconstruct the low-frequency vasoconstriction signal, and extract the time-mean value of blood flow to generate an absolute perfusion matrix; with the flap vascular pedicle as the zero phase origin, the phase difference of low-frequency vasoconstriction signal is calculated pixel by pixel to construct the phase delay matrix. Tensor fusion units are used to extract the thermal radiation fluctuation sequence from the multimodal registration dataset. The cross-correlation extreme values ​​between the thermal radiation fluctuation sequence and the low-frequency contraction signal at the same coordinate are calculated through a sliding window to generate a heat flow cross-correlation matrix. Channels aggregate the absolute perfusion matrix, phase delay matrix, and heat flow cross-correlation matrix to construct a microcirculation evaluation tensor. The feature clusters in the microcirculation evaluation tensor with absolute perfusion exceeding the baseline threshold, cross-correlation extreme values ​​below the coordination threshold, and conduction phase difference exceeding the impedance threshold are located as pseudo-perfusion regions. A spatial mask is constructed to filter out pseudo-perfusion regions, and the topological outer edge of the remaining connected domain is extracted as the effective perfusion boundary. The state extrapolation unit is used to extract the topological evolution vector of the effective perfusion boundary shrinking towards the zero-phase origin along the time axis, integrate the temporal decay gradient of the microcirculation evaluation tensor with the topological evolution vector, and solve the ischemic failure tendency matrix of the flap mesh. Based on the spatial distribution gradient and temporal evolution rate of the ischemic failure tendency matrix, spatiotemporal state extrapolation calculation is performed to calibrate the spatial spread trajectory of flap necrosis and the time node of crisis prediction.

Images