Diabetic foot healing trend prediction method and device based on coupling of multi-phase wound images and plantar pressure distribution

Abstract

The present application relates to the technical field of medical image processing, and discloses a diabetic foot healing trend prediction method and device based on multi-phase wound image and foot pressure distribution coupling, wherein the method comprises the following steps: constructing a standardized wound time series dataset; constructing a cross-modal alignment feature set; generating a fusion prediction feature tensor; inputting a pre-trained double-flow time series prediction model; and performing false improvement identification and grading early warning. Compared with the prior art, which evaluates the healing state according to the area change and color change in a single wound image, especially under the condition that the wound surface is short-term contraction but the local pressure on the wound is not relieved, the technical problem of early identification of the false improvement trend of the apparent shrinkage but the deep load not being relieved cannot be solved. Since the present application extracts features from multi-phase wound image features and foot pressure time series distribution, and combines a machine learning prediction model, the accuracy of diabetic foot healing trend prediction is improved.

Description

Technical Field

[0001] This invention relates to the field of medical image processing technology, and in particular to a method and apparatus for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution. Background Technology

[0002] Currently, the follow-up assessment methods for diabetic foot wounds still have significant shortcomings. For example, during outpatient dressing changes and periodic follow-up visits, clinicians typically assess the wound healing status based on single-taken wound images, changes in wound area, color changes, tissue coverage, or manual observation. While these methods can reflect changes in the surface morphology of the wound to some extent, they lack continuous and quantitative analytical tools to analyze the intrinsic correlation between long-term local pressure on the wound and changes in the wound's appearance. This is especially true for diabetic foot wounds located in high-load areas such as the forefoot metatarsal head region and the plantar surface of the toes. After dressing changes, the wound surface may show short-term shrinkage, increased granulation tissue coverage, or improved edge color. However, because the peak local pressure, duration of pressure, and repetitive pressure trajectory during the patient's gait do not decrease synchronously, the deep tissues of the wound remain under high-load stimulation, making subsequent complications such as enlarged wound edge maceration, deepened undermining, stagnant healing, and even relapse more likely. Existing technologies mainly focus on static image analysis of wounds or extraction of wound surface features, which cannot fully reflect the complex state of wound surface contraction and deep continuous pressure, and are difficult to meet the need for early identification of false improvement in diabetic foot outpatient follow-up.

[0003] Furthermore, while some existing solutions incorporate wound image recognition, tissue segmentation, or area measurement, their analysis primarily focuses on single wound images, emphasizing surface information such as wound boundaries, color, and tissue distribution. They lack continuous modeling of wound changes over multiple follow-up periods. For diabetic foot patients, wound healing status is not solely determined by the appearance of a single image, but is closely related to the history of local pressure over several days, accumulated walking load, shift of the pressure center, and decompression procedures. Judging healing improvement solely based on current wound area reduction or local color improvement easily overlooks the crucial factor of persistent high pressure on the wound, leading to misdiagnosis. Meanwhile, plantar pressure data usually exists in the form of continuous time series, while wound images are mostly collected at intervals of 3 to 7 days. The two types of data have significant differences in collection frequency, spatial location and time span. Existing technologies lack a processing link to perform unified coordinate association, time alignment and historical load coupling analysis between multi-temporal wound images and plantar pressure distribution, which cannot fully meet the need for continuous, stable and precise prediction of the healing trend of diabetic foot.

[0004] Therefore, there is an urgent need for a method that can still achieve joint analysis of multi-temporal wound images and plantar pressure distribution, identification of false improvement status, and prediction of healing trend when the wound surface shrinks in the short term but local pressure remains unrelieved. This would improve the accuracy of wound status assessment, the timeliness of risk warning, and the reliability of subsequent intervention decisions during outpatient follow-up of diabetic foot. Summary of the Invention

[0005] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution. This method aims to solve the technical problem in existing technologies that assess the healing status based on area and color changes in a single wound image, especially when the wound surface contracts in the short term but local pressure remains unrelieved. This method cannot identify the false improvement trend of apparent shrinkage but unrelieved deep load in advance.

[0006] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution.

[0007] The method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution includes:

[0008] Step S10: Obtain multi-temporal wound image sequences and insole-type plantar pressure matrix sequences at at least three follow-up times for the same diabetic foot patient. Combine these with a unified standardization processing method to perform the task of constructing a standardized wound time series dataset and output a standardized wound time series dataset.

[0009] Step S20: Based on the standardized profile time series dataset, perform the time series feature extraction task using the cross-modal alignment feature construction method, and output the cross-modal alignment feature set;

[0010] Step S30: Based on the cross-modal aligned feature set, perform the fusion prediction feature tensor preprocessing task using the historical pressure lag weighting method, and output the fusion prediction feature tensor;

[0011] Step S40: Based on the fused prediction feature tensor, perform the healing trend prediction task using the joint discrimination method of healing trend, and output the healing trend prediction result;

[0012] Step S50: Based on the healing trend prediction results, identify false improvement and output graded early warning, and output a healing trend prediction report.

[0013] Preferably, step S10, which involves acquiring multi-temporal wound image sequences from at least three follow-up times for the same diabetic foot patient, as well as insole-type plantar pressure matrix sequences from adjacent follow-up times, and combining this with a unified standardization processing method to perform a standardized wound time-series dataset construction task, and outputting a standardized wound time-series dataset, specifically includes:

[0014] Step S101: Obtain multi-temporal wound image sequences from at least three follow-up times for the same diabetic foot patient, as well as insole-type plantar pressure matrix sequences from adjacent follow-up times. Obtain the sequence from the multi-temporal wound image sequence... Wound images at each follow-up time. Identify wound images The scale reference in the text is based on the actual side length of the scale reference. With corresponding pixel length Calculate the first Standardized parameters corresponding to each follow-up time point , ;

[0015] Step S102: Obtain the preset reference color block truth vector and the wound image. The observed color patch vectors in the image are used to perform color correction and plantar posture correction on the wound image based on the reference color patch ground truth vector and the observed color patch vectors, resulting in a posture-uniform corrected wound image. and standardize the scale parameters. With corrected wound images Perform associative storage and output a standardized subset of image-associated data.

[0016] Step S103: Perform zero-point correction, gain correction, and image timestamp and pressure timestamp alignment processing on the insole-type plantar pressure matrix sequence to obtain the synchronized corrected pressure matrix sequence, and combine the image standardized associated data subset and the corrected pressure matrix sequence to form a standardized wound time series dataset.

[0017] Preferably, step S20, which involves performing a time-series feature extraction task based on the standardized profile time-series dataset using a cross-modal alignment feature construction method and outputting a cross-modal alignment feature set, specifically includes:

[0018] Step S201: Correct the profile images from the standardized profile time-series dataset. Input a preset profile segmentation model, and the profile segmentation model outputs a profile mask. Based on the profile mask and scale standardization parameters Calculate the cross-sectional area and cross-sectional perimeter ,in, ; and extract the profile mask. The corresponding profile region color feature vector ;

[0019] Step S202: Based on profile mask Construct the outer expansion ring region of the fossa, based on the profile mask. The bleed area mask is extracted using a bleed area extraction method based on HSV color space threshold segmentation and GLCM texture constraints. And calculate the permeation ratio. ,in, ;

[0020] Step S203: Calculate the cross-sectional area , cross-sectional perimeter and the color feature vector of the profile region As a temporal feature of the profile appearance, the profile mask Masking of the seepage and impregnation area and the proportion of exudation As a constraint feature of the creation edge state, a cross-modal alignment feature set is constructed based on the temporal features of the profile appearance and the constraint features of the creation edge state using a homography matrix mapping and region correspondence matching alignment processing method.

[0021] Preferably, in step S20, the preset wound segmentation model is built using an encoder-decoder network architecture and an attention enhancement network architecture; the wound segmentation model includes an input layer for receiving corrected wound images from a standardized wound time-series dataset; an encoding layer for extracting shallow texture features and deep semantic features of the wound region; a bottleneck feature fusion layer for aggregating wound edge features and regional semantic features at different scales; a decoding layer for upsampling the aggregated features stepwise and restoring the wound spatial resolution; and an output layer for outputting a wound mask.

[0022] Preferably, step S30, which involves performing a preprocessing task based on the cross-modal aligned feature set using a historical pressure hysteresis weighting method to output the fused prediction feature tensor, specifically includes:

[0023] Step S301: For the first Follow-up time Constructing a historical pressure window , ;in, Preset the history window length;

[0024] Step S302: Introduce the lag weighting function , ;in, For any pressure sampling time within the historical pressure window and the first Follow-up time The time interval between The time decay coefficient, It is a natural exponential function; within the historical pressure window Internally, a hysteresis weighting function is used for the wound-related stress time-series features in the cross-modal alignment feature set. Weighted processing is performed to output the lag-weighted pressure load characteristics. ;

[0025] Step S303: Based on the characteristics of hysteresis-weighted pressure load The profile appearance temporal features and creation state constraint features in the cross-modal aligned feature set are subjected to feature splicing and normalization processing to output a fused prediction feature tensor.

[0026] Preferably, step S40, which involves performing a healing trend prediction task based on the fused prediction feature tensor using a joint discrimination method and outputting the healing trend prediction result, specifically includes:

[0027] Step S401: Input the fused prediction feature tensor into the pre-trained dual-stream temporal prediction model. The dual-stream temporal prediction model encodes the wound image-related features and the plantar pressure-related features respectively, and outputs the estimated remaining time for wound closure, the true healing trend score, the stagnation trend score, and the false improvement trend score.

[0028] Step S402: The true healing trend score, stagnation trend score, and false improvement trend score are probabilistically processed to obtain the healing trend prediction result; among them, the dual-stream time series prediction model is built with a two-layer long short-term memory network, and the weighted focus loss function is used for parameter optimization training under class imbalance constraint during the pre-training process.

[0029] Preferably, step S50, which involves identifying false improvement and issuing graded early warnings based on the healing trend prediction results, and outputting a healing trend prediction report, specifically includes:

[0030] Step S501: Obtain the permeation ratio Lag-weighted pressure load characteristics False improvement trend prediction probability stagnation trend prediction probability Estimate the remaining time for wound closure When the healing trend prediction results meet the preset criteria of wound area shrinkage and exudate ratio meets the criteria... Meanwhile, the lag-weighted pressure load characteristics satisfy At that time, the wound was determined to be in a first-degree pseudo-improvement state, characterized by apparent shrinkage but unresolved deep burden; among which, This is the threshold for determining a level one false improvement.

[0031] Step S502: Based on the determination of a Level 1 false improvement state in step S501, the predicted probability of the false improvement trend in the current healing trend prediction result is... satisfy And the probability of stagnation trend prediction satisfy At that time, the wound was determined to be in a level-two false improvement warning state, requiring close follow-up; among which, The threshold for the probability of false improvement. This represents the probability threshold for a stagnant trend.

[0032] Step S503: Based on the determination of a Level II false improvement warning state in step S502, estimate the remaining time for wound closure. satisfy Or the lag-weighted pressure load characteristics satisfy At that time, the wound was determined to be in a level three false improvement warning state, requiring high-risk follow-up intervention; among which, To close the remaining time threshold, The high load risk threshold;

[0033] Step S504: Based on the above false improvement identification and judgment results, generate and output a healing trend prediction report. The healing trend prediction report includes at least the wound area change results, exudate ratio change results, hysteresis weighted pressure load results, healing trend prediction probability results, false improvement status judgment results, and corresponding graded early warning results.

[0034] The present invention also provides a device for predicting the healing trend of diabetic foot based on multi-temporal wound images coupled with plantar pressure distribution, comprising:

[0035] The data standardization construction module is used to acquire multi-temporal wound image sequences and insole-type plantar pressure matrix sequences at at least three follow-up times for the same diabetic foot patient. Combined with a unified standardization processing method, it performs the task of constructing a standardized wound time series dataset and outputs a standardized wound time series dataset.

[0036] The cross-modal alignment feature construction module is used to perform temporal feature extraction tasks based on the standardized profile temporal dataset using a cross-modal alignment feature construction method, and output a cross-modal alignment feature set;

[0037] The fusion feature preprocessing module is used to perform a fusion prediction feature tensor preprocessing task based on the cross-modal aligned feature set using a historical pressure hysteresis weighting processing method, and output the fusion prediction feature tensor.

[0038] The trend prediction module is used to perform a healing trend prediction task based on the fused prediction feature tensor using a joint discrimination method of healing trends, and output the healing trend prediction result.

[0039] The report output module is used to identify false improvement and provide graded early warning based on the healing trend prediction results, and output a healing trend prediction report.

[0040] The present invention also provides a device for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution, comprising: a memory, a processor, and a program for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution stored in the memory and executable on the processor. When the program for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution is executed by the processor, the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution is realized.

[0041] The present invention also provides a computer program product, including a diabetic foot healing trend prediction program based on the coupling of multi-temporal wound images and plantar pressure distribution. When the diabetic foot healing trend prediction program based on the coupling of multi-temporal wound images and plantar pressure distribution is executed by a processor, it implements the diabetic foot healing trend prediction method based on the coupling of multi-temporal wound images and plantar pressure distribution.

[0042] The beneficial effects of this invention are as follows: By combining the changes in wound area, wound edge morphology, color evolution, and exudation and maceration status in multi-temporal wound images with the local peak pressure, pressure duration, and historical load accumulation results in plantar pressure distribution, this invention can identify a false improvement state where the wound surface contracts in the short term but the local pressure remains unrelieved, indicating apparent shrinkage but unrelieved deep load. This improves the accuracy of determining the healing trend of diabetic foot and reduces the risk of misjudgment based solely on the apparent changes in a single image.

[0043] This invention constructs a historical pressure window and introduces lag weighting to quantitatively characterize the history of local pressure on the wound during the follow-up period. Combined with a dual-flow time-series prediction model, it outputs healing trend prediction results and graded early warning results. This enables earlier detection of healing stagnation or deterioration risks in outpatient continuous follow-up scenarios, thereby improving the timeliness and reliability of follow-up management for diabetic foot patients. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating the first embodiment of a method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution according to the present invention.

[0046] Figure 2 This is a schematic diagram of the local wound pressure coupling during the first follow-up of the first embodiment of the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution according to the present invention.

[0047] Figure 3 This is a schematic diagram of the local wound pressure coupling during the second follow-up of the first embodiment of the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution according to the present invention.

[0048] Figure 4 This is a schematic diagram of the local wound pressure coupling during the third follow-up of the first embodiment of the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution according to the present invention.

[0049] Figure 5 This is a schematic diagram illustrating the divergent changes in wound area and exudate ratio in a first embodiment of a method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution according to the present invention.

[0050] Figure 6 This diagram illustrates the false improvement identification, scoring, and graded early warning of a first embodiment of a method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution according to the present invention.

[0051] Figure 7 This is a schematic diagram of a device for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution according to the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution according to the present invention. The first embodiment of the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution according to the present invention is presented.

[0054] In the first embodiment, the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution includes:

[0055] Step S10: Obtain multi-temporal wound image sequences and insole-type plantar pressure matrix sequences at at least three follow-up times for the same diabetic foot patient. Combine these with a unified standardization processing method to perform the task of constructing a standardized wound time series dataset and output a standardized wound time series dataset.

[0056] It should be noted that the "unified and standardized processing method" refers to the process of performing unified scaling, unified color reference, unified spatial orientation, unified time stamp, and unified numerical reference on the image-side data and pressure-side data respectively, according to the preset data consistency requirements, for multi-temporal wound image sequences collected at different follow-up times of the same diabetic foot patient and insole-type plantar pressure matrix sequences collected continuously in adjacent follow-up periods. This is to eliminate the data incomparability caused by differences in acquisition equipment, changes in acquisition distance, shooting angle offset, ambient light fluctuations, zero drift of pressure sensing unit, and continuous sampling clock deviation. The unified standardization processing method includes: performing scale reference object identification and scale conversion processing on multi-temporal wound images to ensure that the wound area, wound edge length, and width of the adjacent area at different follow-up times have a unified physical dimension; performing reference color block correction, brightness compensation, and color consistency adjustment on multi-temporal wound images to ensure that the wound color information, exudate area color information, and wound edge maceration area color information at different follow-up times can be compared under the same color reference; performing plantar posture alignment and regional position normalization processing on multi-temporal wound images to ensure that the wound area, wound edge area, and wound edge outward ring area can be positionally correlated under a unified plantar coordinate framework; performing zero-point offset correction, gain consistency correction, abnormal pressure point removal, and timestamp synchronization processing on the insole-type plantar pressure matrix sequence to ensure that the local peak pressure, duration of continuous pressure, and cumulative local load in the continuous pressure matrix can establish a traceable temporal relationship with the corresponding follow-up time of the wound image; and organizing the above-processed image-side data, scale parameters, time stamp information, and pressure-side data together into a standardized wound time-series dataset. Furthermore, the "standardized wound time series dataset" refers to a data set that includes at least corrected wound images, scale conversion parameters corresponding to each wound image, follow-up time identifiers corresponding to each wound image, and a corrected and aligned plantar pressure matrix sequence, so that subsequent steps can directly call this data set to perform wound appearance feature extraction, wound edge status analysis, historical pressure analysis, and healing trend prediction.

[0057] It is understood that in this step, the present invention first performs unified standardization processing on multi-temporal wound images and plantar pressure matrix sequences. The technical effect is to provide a consistent, comparable, and continuous data foundation for subsequent steps. Since diabetic foot patients typically do not complete image and pressure acquisition under completely constant acquisition environments during outpatient follow-ups, data obtained from the same wound at different follow-up times often exhibit significant external disturbances. Directly comparing wound images acquired at different times by area, color, or boundary can easily lead to misinterpreting changes in shooting distance, illumination, and foot posture as changes in wound condition. Similarly, directly mechanically correlating a continuous plantar pressure matrix sequence with a wound image at a specific moment can easily result in temporal and localized mismatches. Through this unified standardization processing of image and pressure data, the wound area changes, wound edge contraction changes, color changes, exudation and maceration changes, and corresponding wound pressure changes extracted in subsequent steps are all established under unified reference conditions, thereby significantly improving the comparability of data between different follow-up times.

[0058] It should be understood that, compared to traditional techniques that only measure the area of ​​a single wound image or perform independent statistical analysis on continuous pressure data, the unified standardization process used in this invention does not simply stitch the two types of data together. Instead, it corrects key factors affecting the authenticity and comparability of the data before feature extraction, thus achieving superior technical results. Traditionally, one approach focuses solely on the wound image itself, estimating the wound area through manual outlining, fixed pixel conversion, or simple image segmentation, and then using area reduction or expansion as the basis for judgment. However, this approach struggles to distinguish between area changes caused by actual tissue repair and apparent changes caused by variations in shooting angle, distance, and lighting. Another approach, while incorporating pressure monitoring, typically only provides the average pressure of the entire foot, local peak pressure, or pressure statistics for a specific measurement day, lacking precise correspondence with the wound image acquisition time, wound spatial location, and adjacent areas. Therefore, it cannot effectively reflect the continuous load state borne by the wound area. In this step, the invention avoids area errors caused by traditional fixed-pixel conversion through standardized scale processing; avoids color feature drift caused by insufficient brightness compensation through standardized color processing; avoids spatial inconsistencies caused by simple scaling through standardized pose processing; and avoids zero-drift and time misalignment errors caused by direct use of traditional pressure data through pressure sequence correction and timestamp synchronization processing. Especially in the context of diabetic foot clinic follow-up, the imaging conditions, foot positioning, and actual walking status of patients may differ at each visit. Traditional methods often struggle to maintain stable results in complex acquisition environments. However, the standardized processing in this step of the invention can significantly reduce the impact of external interference factors on the subsequent healing trend judgment, thereby making the joint analysis based on multi-temporal wound images and plantar pressure distribution closer to the actual wound recovery process and improving the accuracy of identifying false improvement states.

[0059] For example, in an outpatient follow-up scenario, a diabetic foot patient's wound is located below the metatarsal head of the forefoot. During the first follow-up, medical staff immediately captured an image of the wound after changing the dressing. During the second follow-up, due to different lighting conditions in the examination room and a closer shooting distance than the first follow-up, the wound area in the captured image appears slightly smaller and slightly more reddish. If the traditional method of directly comparing the area and color of a single image is used, it is easy to judge that the patient's healing trend is continuously improving. However, in this step of the present invention, firstly, the two wound images are uniformly scaled according to a scale reference, so that wound images at different shooting distances can be converted into data at the same physical scale; then, the two images are color-corrected according to a reference color block to correct color deviations under different lighting conditions; furthermore, the wound position in the two wound images is mapped to a unified foot coordinate frame through foot posture alignment. Simultaneously, the insole-type plantar pressure matrix sequence continuously recorded between the two follow-up visits was zero-point corrected and timestamped to obtain the local pressure history within the time period corresponding to the second wound image. After the above standardization processing, it was found that although the wound surface area of ​​the patient was slightly reduced compared to the previous follow-up visit, the corresponding area of ​​the wound was still under high pressure exposure for several consecutive days, and the local pressure center did not deviate significantly from the center area of ​​the wound. It can be seen that the standardized wound time-series dataset constructed in this step of the present invention can effectively eliminate the interference caused by the shooting environment and pressure acquisition deviation, so that subsequent steps can determine whether the patient belongs to the case of apparent improvement but deep load not being relieved based on real, continuous, and unified data. For example, for another patient, if the first and second follow-up images show obvious color difference without standardization processing, traditional methods may mistakenly interpret the darkening of the wound edge area as increased inflammatory exudation; however, after the unified color correction of the present invention, it can be confirmed that the color difference mainly comes from the change of the shooting environment rather than the change of the wound itself, thereby avoiding subsequent misjudgment.

[0060] Step S20: Based on the standardized profile time series dataset, perform the time series feature extraction task using the cross-modal alignment feature construction method, and output the cross-modal alignment feature set;

[0061] It should be noted that the "cross-modal alignment feature construction method" refers to the process of associating wound image data and plantar pressure data based on the standardized wound time-series dataset output in step S10, according to unified time identifiers, unified plantar spatial coordinates, and unified feature organization rules, to form a feature set that can be directly called upon for subsequent historical pressure lag weighting processing and joint discrimination of healing trends. The standardized wound time-series dataset includes at least the corrected wound image, scale normalization parameters, follow-up time identifier information, and the corrected and synchronized plantar pressure matrix sequence. The time-series feature extraction task includes: extracting wound area, wound perimeter, wound region color features, exudate and maceration areas in the outer ring region of the wound edge, and the exudate percentage from the corrected wound image; establishing a correspondence between the wound region and local regions of the plantar pressure matrix based on the wound's position in the unified plantar coordinate system; and extracting the corresponding pressure time-series features of the wound based on the correspondence to form a cross-modal alignment feature set.

[0062] It should be understood that, compared to traditional techniques that extract area or color information from a single wound image or extract full-foot statistical indicators from plantar pressure distribution, the improvement in technical effectiveness of this step lies primarily in the depth of establishing cross-modal correspondences and the completeness of the temporal analysis foundation. Traditional image analysis schemes typically treat the wound as an independent two-dimensional visual object, focusing only on the increase or decrease in wound surface area, color intensity, or the regularity of its boundaries, while ignoring the local mechanical load continuously borne by the plantar area where the wound is located during walking. Traditional pressure analysis schemes often focus on the average or peak pressure of the entire forefoot, arch, or heel region, but do not precisely correspond them to specific wound areas, thus failing to demonstrate whether a certain local high pressure truly acts on the wound itself. In contrast, this step does not simply stitch together image features and pressure features, but first establishes a correspondence between the wound area and local areas of the plantar pressure matrix based on a unified plantar coordinate system, and then extracts the apparent temporal features of the wound, the wound edge state constraint features, and the corresponding pressure temporal features of the wound based on this correspondence. The resulting technical benefits are twofold: firstly, it avoids the one-sided judgment in traditional methods that equates wound shrinkage with improved healing; secondly, it avoids the ambiguity in traditional pressure analysis where "high pressure across the entire foot cannot be confirmed to be acting on the wound area." Especially in the context of diabetic foot outpatient follow-up, short-term surface contraction of the wound does not necessarily mean that the deep load has been relieved. Without the cross-modal aligned feature set established in this step, it is difficult to accurately identify the false improvement state of "apparent improvement but localized persistent pressure" in subsequent analysis. Therefore, compared with traditional techniques, this step significantly improves the specificity, continuity, and reliability of wound condition analysis, and creates a more solid feature foundation for subsequent trend prediction and graded early warning.

[0063] For example, such as Figure 2 As shown, this figure illustrates the spatial distribution of the wound area and its corresponding local pressure area within a unified plantar coordinate system. The solid white line represents the wound boundary, the dashed red line represents the exudative and macerated boundary identified within the outer circumferential zone of the wound edge, and the background heat map and its contour lines represent the normalized pressure distribution within the corresponding local area of ​​the wound. It can be seen that the wound boundary is relatively large at this stage, and a certain range of exudative and macerated area already exists within the outer circumferential zone of the wound edge. Simultaneously, a significant pressure heat zone has formed within the corresponding local area of ​​the wound, and the center of the pressure heat zone overlaps significantly with the center of the wound. This figure demonstrates that in the initial follow-up phase, the wound surface condition and the state of continuous local pressure coexist, providing a foundation for subsequent analysis of the coupling relationship between apparent changes in the wound and the pressure history. Figure 3 As shown, the wound boundary has shrunk overall, indicating that if only the surface contour of the wound is observed, it might be assumed that the wound has entered a certain degree of recovery. However, it can also be seen that the exudate and maceration boundary, indicated by the red dashed line, has not decreased synchronously with the wound boundary; instead, it remains clearly distributed in the vicinity of the wound edge, and the local pressure heat zone still overlaps significantly with the wound area. This shows that although the apparent size of the wound has decreased, the local condition of the wound edge has not improved synchronously, and the area where the wound is located continues to bear a high pressure load. Therefore, judging the healing status solely based on changes in wound area easily overlooks the important factor of continuous deep pressure. Figure 4 As shown, the wound boundary is compared to Figure 2 and Figure 3 Further shrinkage is observed, outwardly manifesting as continuous contraction; however, simultaneously, the boundary of exudation and maceration at the wound edge remains relatively clear, and the intensity of the local pressure-heat zone further increases, with the center of the pressure-heat zone still highly overlapping with the local area where the wound is located. This figure shows that during continuous follow-up, although the wound exhibits a change of "surface shrinkage," the corresponding local area's continuous pressure state has not been relieved, and even shows a tendency to intensify. This reflects the pseudo-improvement scenario of "apparent shrinkage but unrelieved deep load," which is the key focus of this invention. Figure 5 As shown, one curve represents the change in wound area over time, and the other curve represents the change in exudate percentage over time. It is clear that as the number of follow-up visits increases, the wound area continuously decreases, while the exudate percentage continuously increases. This result indicates that wound surface shrinkage does not necessarily accompany a synchronous improvement in the wound edge condition, especially when persistent wetting, maceration, or abnormal exudation exists in the vicinity of the wound edge. Therefore, judging "improved healing" solely based on a reduction in wound area has significant limitations. Figure 6As shown, the invention presents a comprehensive risk score for pseudo-improvement calculated based on indicators such as changes in wound area, changes in exudate percentage, historical weighted pressure load, and the proportion of high-pressure duration, along with corresponding Level 1, Level 2, and Level 3 warning thresholds. It is evident that at the first follow-up, the comprehensive risk score was at a relatively low level; at the second follow-up, the score had already crossed the Level 1 warning threshold, indicating a need for closer follow-up observation; and at the third follow-up, the comprehensive risk score further approached or reached a higher-level threshold, indicating that the wound was more likely in a high-risk pseudo-improvement state, requiring more proactive follow-up management or decompression intervention. This demonstrates that the invention can not only identify pseudo-improvement states but also classify and express their risk levels, thus providing more specific quantitative evidence for continuous clinical follow-up management.

[0064] Step S30: Based on the cross-modal aligned feature set, perform the fusion prediction feature tensor preprocessing task using the historical pressure lag weighting method, and output the fusion prediction feature tensor;

[0065] It should be noted that the "historical pressure lag weighting method" means that, for the wound state corresponding to the k-th follow-up time, not only is the local pressure information near the current follow-up time used, but also the historical pressure exposure information within a preset time range before the follow-up time is further introduced. According to the principle that "the closer the historical pressure data is to the current follow-up time, the greater the impact on the current wound state, and the farther the historical pressure data is from the current follow-up time, the smaller the impact on the current wound state", the historical pressure sequence is subjected to attenuation weighting processing, thereby constructing a fusion prediction feature that reflects the cumulative effect of local load on the wound.

[0066] Understandably, this step transforms the localized wound pressure information, originally scattered across different time points and sampling frequencies, into continuous discriminative features that reflect the "cumulative impact of load." This allows subsequent healing trend predictions to move beyond the instantaneous state at a single moment and reflect the actual pressure background experienced by the wound over a historical period. In diabetic foot outpatient follow-up scenarios, whether a wound has truly entered a recovery state is usually not solely determined by the appearance of the images taken on the current follow-up day. It is also closely related to whether the wound has been continuously subjected to localized high-pressure stimulation in the preceding days, whether there has been prolonged repetitive pressure, and whether the pressure center consistently coincides with the wound location. This step, by applying lag-weighted processing to the wound's corresponding pressure sequence features within the historical pressure window, allows pressure exposure information closer to the current follow-up time to occupy a higher weight in subsequent predictions, while preserving the continuous impact of earlier pressure exposures on the wound microenvironment and tissue recovery status. This results in lag-weighted pressure load features that more accurately characterize the deep load state of the wound. Furthermore, this step integrates the lagged weighted pressure load characteristics with the apparent temporal characteristics of the wound surface and the constraint characteristics of the wound edge state into a fusion prediction feature tensor, enabling the subsequent model to simultaneously receive two types of input information: "wound surface changes" and "historical pressure changes," thereby improving the ability to distinguish between true healing trends, stagnant trends, and pseudo-improvement trends.

[0067] Step S40: Based on the fused prediction feature tensor, perform the healing trend prediction task using the joint discrimination method of healing trend, and output the healing trend prediction result;

[0068] It should be noted that the "joint discrimination method for healing trends" refers to the following: the fused prediction feature tensor output in step S30 is input into the pre-trained trend prediction model, and the trend prediction model performs joint analysis on changes in wound appearance, changes in wound edge state, and changes in corresponding local pressure on the wound, and outputs the prediction result of the current wound under a preset trend category. In this embodiment, the trend prediction model is preferably a dual-stream temporal prediction model, which includes at least an image-related feature encoding branch and a pressure-related feature encoding branch. The image-related feature encoding branch is used to receive temporal features of wound appearance and constraints of wound edge state, and the pressure-related feature encoding branch is used to receive temporal features of corresponding pressure on the wound and hysteresis-weighted pressure load features. After the two branches complete the temporal feature encoding respectively, joint discrimination processing is performed to output the true healing trend score, the stagnation trend score, and the false improvement trend score. Furthermore, after performing probabilistic processing on the above three trend scores, the healing trend prediction results are obtained. The healing trend prediction results include at least the true healing trend prediction probability, the stagnation trend prediction probability, and the false improvement trend prediction probability. In some implementations, the estimated remaining wound closure time can also be output simultaneously for subsequent graded early warning. The dual-stream time-series prediction model preferably adopts a two-layer long short-term memory network. Its pre-training process preferably uses a weighted focus loss function for parameter optimization training under class imbalance constraints to enhance the ability to identify subtle differences between false improvement samples and stagnation samples.

[0069] Understandably, the technical effect of this step lies in the following: based on the organization of wound image-side and pressure-side information into a fused predictive feature tensor in step S30, a joint discriminant model is further used to achieve collaborative analysis of multi-source temporal features, thereby avoiding trend judgments based solely on a single feature or simple threshold rules. Because the recovery state of diabetic foot wounds exhibits significant temporal and multi-factor coupling, the relationship between wound area reduction, wound perimeter changes, color changes, exudation changes, and local pressure changes is not always linearly consistent. Therefore, simply using area thresholds, color thresholds, or peak pressure thresholds often fails to accurately reflect the true healing trend of the current wound. This step encodes image-side and pressure-side features separately using a dual-stream temporal prediction model. On the one hand, it preserves the continuous evolution information of the wound's apparent changes; on the other hand, it preserves the continuous evolution information of the historical pressure changes in the corresponding local area of ​​the wound. Then, through joint discriminant analysis, three types of trend results are obtained, allowing for the simultaneous consideration of the relationship between visual improvement of the wound and continuous local pressure within the same discriminant framework. Therefore, the subsequent healing trend prediction results are no longer isolated area judgments or pressure judgments, but trend probability results based on the comprehensive analysis of multiple time-series features. Therefore, they are more suitable for dynamic risk identification in outpatient continuous follow-up scenarios.

[0070] It should be understood that, compared to traditional techniques that use single-rule classification, single-static feature classification, or single-modal prediction models to judge wound status, the improvement in technical effectiveness of this step is mainly reflected in the completeness of the trend discrimination dimension and the fineness of the discrimination boundary. In traditional solutions, a common approach is to make hard judgments about wound status based on several manually set thresholds. For example, a continuously shrinking wound area is considered a healing trend, while a unchanged or increasing area is considered a stagnant or deteriorating trend. However, this type of method struggles to identify complex situations such as "shrinking area but worsening wound edge condition, and unrelieved local load." Another approach, while introducing machine learning models, often limits input features to single images or single-moment pressure statistics, lacking simultaneous modeling of historical changes on both the image and pressure sides. This can easily lead to trend discrimination biased towards one modality while ignoring important information from another. This step uses a dual-stream temporal prediction model to jointly encode the fusion prediction feature tensor, enabling continuous expression of wound appearance changes and local pressure changes within their respective branches. Multimodal fusion is then completed in the joint discrimination stage, thus more accurately distinguishing between states of "true healing but effective relief of local pressure," stagnant states of "no significant improvement in area and persistent exudation," and pseudo-improvement states of "surface contraction but unrelieved deep-seated burden." Especially when the number of pseudo-improvement samples is relatively small and close to the boundary of stagnant samples, pre-training under class imbalance constraints using a weighted focus loss function allows the model to pay more attention to difficult-to-classify samples during training, thereby improving the sensitivity to identifying pseudo-improvement trends. Therefore, compared to traditional techniques, this step not only improves the accuracy of trend classification results but also enhances the ability to identify high-risk clinical boundary states.

[0071] Step S50: Based on the healing trend prediction results, identify false improvement and output graded early warning, and output a healing trend prediction report.

[0072] It should be noted that "false improvement identification and graded early warning output" refers to the process of identifying whether the current wound is in a false improvement state of "apparent shrinkage but unrelieved deep load" based on the healing trend prediction result output in step S40 and combined with the key auxiliary features formed in steps S20 and S30, and generating corresponding early warning results according to preset risk levels. In this embodiment, false improvement identification is completed at least based on changes in wound area, changes in exudate ratio, hysteretic weighted pressure load characteristics, false improvement trend prediction probability, and, in some implementation methods, the estimated remaining wound closure time. Specifically, when the wound area shows a shrinking trend, while the exudate ratio does not decrease or even increases, and the hysteretic weighted pressure load characteristic is higher than a preset threshold, it can be determined that the current wound is in a false improvement state of apparent shrinkage but unrelieved deep load. On this basis, different early warning levels such as ordinary prompts, key follow-up prompts, and high-risk re-examination prompts can be distinguished based on the magnitude of the false improvement trend prediction probability, the magnitude of the stagnation trend prediction probability, and the estimated remaining closure time or high pressure load level. Furthermore, the "Healing Trend Prediction Report" refers to a structured set of results output at the current follow-up time. The structured set of results includes at least the results of changes in wound area, changes in exudate ratio, changes in pressure characteristics of the corresponding local area of ​​the wound, historical weighted pressure load, probability results of three types of trend predictions, results of false improvement identification, and warning level results.

[0073] It should be understood that, compared to traditional techniques that only output single-level judgments such as "whether it has healed" or "whether it has worsened," or only provide single-item prompts such as whether a certain pressure index exceeds the standard or whether a certain area index has changed, the improvement in technical effectiveness of this step is mainly reflected in the comprehensiveness of risk identification, the hierarchical nature of result expression, and the targeted nature of clinical application. Traditional methods often only provide fragmented conclusions, such as indicating improvement when the wound area decreases, indicating observation when the wound area remains unchanged, and indicating decompression when the local pressure is high. Although such single conclusions are intuitive, they cannot explain the combined relationship between multiple indicators, nor can they explain why a wound, although its area has decreased, is still in a high-risk state. This step, by introducing a false improvement identification logic, incorporates factors such as wound area shrinkage, increased exudate ratio, local persistent high pressure, and trend probability output into the same judgment chain, which can provide a more accurate explanation for the complex state of "visual improvement but deep risks still exist." Furthermore, traditional methods rarely provide a tiered early warning mechanism, often failing to differentiate between patients requiring routine follow-up, intensive follow-up, and high-risk re-examination. This step, however, uses tiered early warning outputs to assign different management actions to wound conditions at varying risk levels, thereby improving the rationality of follow-up resource allocation. Especially in outpatient settings with a large number of patients and a fast-paced follow-up process, the lack of structured tiered reports provided by this step makes it difficult to prioritize and treat high-risk patients within a limited timeframe. Therefore, compared to traditional techniques, this step not only improves the completeness of the results but also enhances the support of the output information for subsequent clinical decision-making.

[0074] Example 2: Furthermore, the present invention provides a diabetic foot healing trend prediction device based on the coupling of multi-temporal wound images and plantar pressure distribution. This device employs a method for predicting diabetic foot healing trends based on the coupling of multi-temporal wound images and plantar pressure distribution as described in the above embodiments, thus solving the technical problem of predicting diabetic foot healing trends based on the coupling of multi-temporal wound images and plantar pressure distribution. Compared with the prior art, the beneficial effects of the diabetic foot healing trend prediction device based on the coupling of multi-temporal wound images and plantar pressure distribution provided by the present invention are the same as those of the method for predicting diabetic foot healing trends based on the coupling of multi-temporal wound images and plantar pressure distribution provided in the above embodiments. Other technical features of the diabetic foot healing trend prediction device based on the coupling of multi-temporal wound images and plantar pressure distribution are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0075] Example 3: This invention provides a device for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution. Please refer to... Figure 7A device for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to execute the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution described in Embodiment 1 above. The device for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution in this embodiment of the invention may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. This device for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the invention. A diabetic foot healing trend prediction device based on multi-temporal wound images coupled with plantar pressure distribution may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the diabetic foot healing trend prediction device based on multi-temporal wound images coupled with plantar pressure distribution. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An I / O interface 1006 is also connected to the bus. Typically, the following devices can be connected to I / O interface 1006: input devices 1007 including, for example, a touchscreen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows a diabetic foot healing trend prediction device based on multi-temporal wound image coupling with plantar pressure distribution to wirelessly or wiredly communicate with other devices to exchange data. Although a diabetic foot healing trend prediction device based on multi-temporal wound image coupling with plantar pressure distribution with various devices is shown in the figure, it should be understood that implementation or possession of all the devices shown is not required.It can be implemented alternatively or with more or fewer devices.

[0076] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the above-described method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution. The computer program product provided by this invention can solve the technical problem of predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution provided in the above embodiments, and will not be repeated here.

[0077] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0078] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0079] 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 predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution, characterized in that, The methods include: Step S10: Obtain multi-temporal wound image sequences and insole-type plantar pressure matrix sequences at at least three follow-up times for the same diabetic foot patient. Combine these with a unified standardization processing method to perform the task of constructing a standardized wound time series dataset and output a standardized wound time series dataset. Step S20: Based on the standardized profile time series dataset, perform the time series feature extraction task using the cross-modal alignment feature construction method, and output the cross-modal alignment feature set; Step S30: Based on the cross-modal aligned feature set, perform the fusion prediction feature tensor preprocessing task using the historical pressure lag weighting method, and output the fusion prediction feature tensor; Step S40: Based on the fused prediction feature tensor, perform the healing trend prediction task using the joint discrimination method of healing trend, and output the healing trend prediction result; Step S50: Based on the healing trend prediction results, identify false improvement and output graded early warning, and output a healing trend prediction report.

2. The method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution as described in claim 1, characterized in that, Step S10 involves acquiring multi-temporal wound image sequences from at least three follow-up times for the same diabetic foot patient, as well as insole-type plantar pressure matrix sequences from adjacent follow-up times. This is combined with a unified standardization processing method to construct a standardized wound time-series dataset, and the resulting output includes the following steps: Step S101: Obtain multi-temporal wound image sequences from at least three follow-up times for the same diabetic foot patient, as well as insole-type plantar pressure matrix sequences from adjacent follow-up times. Obtain the sequence from the multi-temporal wound image sequence... Wound images at each follow-up time. Identify wound images The scale reference in the text is based on the actual side length of the scale reference. With corresponding pixel length Calculate the first Standardized parameters corresponding to each follow-up time point , ; Step S102: Obtain the preset reference color block truth vector and the wound image. The observed color patch vectors in the image are used to perform color correction and plantar posture correction on the wound image based on the reference color patch ground truth vector and the observed color patch vectors, resulting in a posture-uniform corrected wound image. and standardize the scale parameters. With corrected wound images Perform associative storage and output a standardized subset of image-associated data. Step S103: Perform zero-point correction, gain correction, and image timestamp and pressure timestamp alignment processing on the insole-type plantar pressure matrix sequence to obtain the synchronized corrected pressure matrix sequence, and combine the image standardized associated data subset and the corrected pressure matrix sequence to form a standardized wound time series dataset.

3. The method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution as described in claim 2, characterized in that, Step S20, which involves performing a time-series feature extraction task based on the standardized profile time-series dataset using a cross-modal alignment feature construction method and outputting a cross-modal alignment feature set, specifically includes: Step S201: Correct the profile images from the standardized profile time-series dataset. Input a preset profile segmentation model, and the profile segmentation model outputs a profile mask. Based on the profile mask and scale standardization parameters Calculate the cross-sectional area and cross-sectional perimeter ,in, ; and extract the profile mask. The corresponding profile region color feature vector ; Step S202: Based on profile mask Construct the outer annular region of the fossa, based on the profile mask. The bleed area mask is extracted using a bleed area extraction method based on HSV color space threshold segmentation and GLCM texture constraints. And calculate the permeation ratio. ,in, ; Step S203: Calculate the cross-sectional area , cross-sectional perimeter and the color feature vector of the profile region As a temporal feature of the profile appearance, the profile mask Masking of the seepage and impregnation area and the proportion of exudation As a constraint feature of the creation edge state, a cross-modal alignment feature set is constructed based on the temporal features of the profile appearance and the constraint features of the creation edge state using a homography matrix mapping and region correspondence matching alignment processing method.

4. The method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution as described in claim 3, characterized in that, In step S20, the preset wound segmentation model is built using an encoder-decoder network architecture and an attention enhancement network architecture. The wound segmentation model includes an input layer, which is used to receive corrected wound images from a standardized wound time series dataset; and an encoding layer, which is used to extract shallow texture features and deep semantic features of the wound region. The bottleneck feature fusion layer is used to aggregate wound edge features and regional semantic features at different scales; The decoding layer is used to upsample the aggregated features step by step and restore the spatial resolution of the wound; the output layer is used to output the wound mask.

5. The method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution as described in claim 1, characterized in that, Step S30, which involves performing a preprocessing task based on the cross-modal aligned feature set using a historical pressure hysteresis weighting method and outputting a fused prediction feature tensor, specifically includes: Step S301: For the first Follow-up time Constructing a historical pressure window , ;in, Preset the history window length; Step S302: Introduce the lag weighting function , ;in, For any pressure sampling time within the historical pressure window and the first Follow-up time The time interval between The time decay coefficient, It is a natural exponential function; within the historical pressure window Internally, a hysteresis weighting function is used for the wound-related stress time-series features in the cross-modal alignment feature set. Weighted processing is performed to output the lag-weighted pressure load characteristics. ; Step S303: Based on the characteristics of hysteresis-weighted pressure load The profile appearance temporal features and creation state constraint features in the cross-modal aligned feature set are subjected to feature splicing and normalization processing to output a fused prediction feature tensor.

6. The method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution as described in claim 1, characterized in that, Step S40, which involves performing a healing trend prediction task based on the fused prediction feature tensor using a joint discrimination method and outputting the healing trend prediction result, specifically includes: Step S401: Input the fused prediction feature tensor into the pre-trained dual-stream temporal prediction model. The dual-stream temporal prediction model encodes the wound image-related features and the plantar pressure-related features respectively, and outputs the estimated remaining time for wound closure, the true healing trend score, the stagnation trend score, and the false improvement trend score. Step S402: The true healing trend score, stagnation trend score, and false improvement trend score are probabilistically processed to obtain the healing trend prediction result; among them, the dual-stream time series prediction model is built with a two-layer long short-term memory network, and the weighted focus loss function is used for parameter optimization training under class imbalance constraint during the pre-training process.

7. The method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution as described in claim 6, characterized in that, Step S50, which involves identifying false improvement and issuing graded early warnings based on the healing trend prediction results, and outputting a healing trend prediction report, specifically includes: Step S501: Obtain the permeation ratio Lag-weighted pressure load characteristics False improvement trend prediction probability stagnation trend prediction probability Estimate the remaining time for wound closure When the healing trend prediction results meet the preset criteria of wound area shrinkage and exudate ratio meets the criteria... Meanwhile, the lag-weighted pressure load characteristics satisfy At that time, the wound was determined to be in a first-degree pseudo-improvement state, characterized by apparent shrinkage but unresolved deep burden; among which, The threshold for determining a first-degree false improvement; Step S502: Based on the determination of a Level 1 false improvement state in step S501, the predicted probability of the false improvement trend in the current healing trend prediction result is... satisfy And the probability of stagnation trend prediction satisfy At that time, the wound was determined to be in a level-two false improvement warning state, requiring close follow-up; among which, The threshold for the probability of false improvement. This represents the probability threshold for a stagnant trend. Step S503: Based on the determination of a Level II false improvement warning state in step S502, estimate the remaining time for wound closure. satisfy Or the lag-weighted pressure load characteristics satisfy At that time, the wound was determined to be in a level three false improvement warning state, requiring high-risk follow-up intervention; among which, To close the remaining time threshold, The high load risk threshold; Step S504: Based on the above false improvement identification and judgment results, generate and output a healing trend prediction report. The healing trend prediction report includes at least the wound area change results, exudate ratio change results, hysteresis weighted pressure load results, healing trend prediction probability results, false improvement status judgment results, and corresponding graded early warning results.

8. A device for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution, applied to the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution as described in any one of claims 1 to 7, characterized in that, The diabetic foot healing trend prediction device based on multi-temporal wound images coupled with plantar pressure distribution includes: The data standardization construction module is used to acquire multi-temporal wound image sequences and insole-type plantar pressure matrix sequences at at least three follow-up times for the same diabetic foot patient. Combined with a unified standardization processing method, it performs the task of constructing a standardized wound time series dataset and outputs a standardized wound time series dataset. The cross-modal alignment feature construction module is used to perform temporal feature extraction tasks based on the standardized profile temporal dataset using a cross-modal alignment feature construction method, and output a cross-modal alignment feature set; The fusion feature preprocessing module is used to perform a fusion prediction feature tensor preprocessing task based on the cross-modal aligned feature set using a historical pressure hysteresis weighting processing method, and output the fusion prediction feature tensor. The trend prediction module is used to perform a healing trend prediction task based on the fused prediction feature tensor using a joint discrimination method of healing trends, and output the healing trend prediction result. The report output module is used to identify false improvement and provide graded early warning based on the healing trend prediction results, and output a healing trend prediction report.

9. A device for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution, characterized in that, The device for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution includes: a memory, a processor, and a program for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution stored in the memory and executable on the processor. When the program for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution is executed by the processor, it implements the method for predicting the healing trend of diabetic foot based on the coupling of multi-temporal wound images and plantar pressure distribution as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a diabetic foot healing trend prediction program based on the coupling of multi-temporal wound images and plantar pressure distribution. When the diabetic foot healing trend prediction program based on the coupling of multi-temporal wound images and plantar pressure distribution is executed by the processor, it implements the diabetic foot healing trend prediction method based on the coupling of multi-temporal wound images and plantar pressure distribution as described in any one of claims 1 to 7.

Images