Endoscope image processing method based on photodynamic imaging and endoscope imaging system

Abstract

The present application relates to the technical field of image processing, and specifically includes an endoscope image processing method based on photodynamic imaging and an endoscope imaging system, which comprises the following steps: collecting lesion images using a photodynamic imaging endoscope, pre-processing and three-dimensional reconstruction, combining reaction and efficacy database analysis; configuring an image processing module according to a request, and connecting a three-dimensional display; extracting a reaction feature for prediction, and if the enhancement is improved, iteratively optimizing parameters and three-dimensionally visualizing output, which solves the technical problem that endoscope image processing through empirical data is slow, resulting in the inability to quickly and accurately respond to imaging requests, realizes fine processing of endoscope images, extracts photodynamic reaction features, adaptively finds optimal image enhancement parameters, quickly responds to imaging requests and performs image targeted enhancement, thereby accurately three-dimensionally visualizing a lesion area, and significantly improving the technical effect of the output quality of endoscope images.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to an endoscopic image processing method and endoscopic imaging system based on photodynamic imaging. Background Technology

[0002] Photodynamic therapy (PDT) refers to the use of photosensitizers and light of specific wavelengths to activate and generate reactive oxygen species, thereby destroying diseased cells. Light of specific wavelengths also excites photosensitive substances in tissues, producing fluorescence or other optical signals, thus achieving high-resolution imaging of tissue structure and function. Photodynamic imaging refers to the simultaneous visualization output during the PDT process. Commonly, endoscopic images are used to observe changes in diseased tissues under PDT in real time. However, deep learning based on empirical data for endoscopic image processing is time-consuming and cannot quickly respond to imaging requests from the application in practical applications.

[0003] In summary, existing technologies suffer from the problem of slow processing of endoscopic images based on empirical data, resulting in an inability to respond quickly and accurately to imaging requests. Summary of the Invention

[0004] This application provides an endoscopic image processing method and an endoscopic imaging system based on photodynamic imaging, aiming to solve the technical problem in the prior art where the processing speed of endoscopic images based on empirical data is slow, resulting in an inability to respond quickly and accurately to imaging requests.

[0005] In view of the above problems, the technical solution to achieve the present application is as follows:

[0006] In a first aspect, this application provides an endoscopic image processing method based on photodynamic imaging, wherein the method includes: acquiring a lesion area using an endoscopic device with photodynamic imaging technology to obtain raw image information, performing image preprocessing on the raw image information, and reconstructing a three-dimensional structure to obtain a stereoscopic restored structure, wherein the raw image information includes image information corresponding to photodynamic response information;

[0007] Based on the photodynamic reaction database, P initial reaction data sets are set according to the reaction type. The initial reaction data sets include photodynamic agent injection status, light source irradiation conditions, and exposure time.

[0008] Based on the treatment effect evaluation database, Q historical response data sets are set according to the treatment effect. The historical response data sets include pre-treatment image information, mid-treatment image information, and post-treatment image information.

[0009] Upon receiving an imaging request, based on the P sets of initial reaction data, the initial state of the endoscope image processing module is configured, and it is connected to the three-dimensional display terminal corresponding to the stereoscopic reconstruction structure.

[0010] Based on the image information corresponding to the photodynamic response information in the preprocessed image information, photodynamic response features are extracted, including fluorescence intensity features, signal distribution features, and edge features.

[0011] The treatment effect evaluation database is connected, and the treatment effect is simultaneously evaluated and predicted in combination with the photodynamic response characteristics. The predicted evaluation results are obtained, and it is determined whether image-targeted enhancement is needed based on the predicted evaluation results.

[0012] If image targeting enhancement is required, at the start window time point of the preset time window, the image targeting enhancement command is activated in conjunction with the imaging request, and an iterative optimization search is performed in the optimization space corresponding to the Q historical response data sets to obtain image targeting enhancement parameters once.

[0013] Using the image targeting enhancement parameters, the P initial reaction data sets corresponding to the initial state of the endoscopic image processing module are cross-updated to obtain P experimental reaction data sets. This process continues until the termination condition is met. Then, image targeting enhancement is performed at the cutoff window time point in the preset time window, and the results are visualized on the 3D display terminal.

[0014] In a second aspect, this application provides an endoscopic imaging system, wherein the endoscopic imaging system includes: a memory for storing executable instructions; and a processor for executing the executable instructions stored in the memory, wherein, when executing the executable instructions, the processor implements the steps of the endoscopic image processing method based on photodynamic imaging as described in any one of the first aspects.

[0015] In summary, one or more technical solutions provided in this application achieve the technical effect of refining endoscopic images, extracting photodynamic response features, adaptively finding the optimal image enhancement parameters, quickly responding to imaging requests and performing targeted image enhancement, thereby providing accurate three-dimensional visualization of lesion areas and significantly improving the output quality of endoscopic images. Attached Figure Description

[0016] Figure 1 This application provides a flowchart illustrating the endoscopic image processing method based on photodynamic imaging;

[0017] Figure 2 This application provides a flowchart illustrating the process of obtaining feature distribution calculation results in an endoscopic image processing method based on photodynamic imaging.

[0018] Figure 3 This application provides a schematic diagram of the structure of an endoscopic imaging system.

[0019] Explanation of reference numerals in the attached drawings: Processor 21, Memory 22, Input device 23, Output device 24. Detailed Implementation

[0020] Example 1

[0021] The present application will now be described in detail with reference to the accompanying drawings, such as... Figure 1 As shown, this application provides an endoscopic image processing method based on photodynamic imaging, wherein the method includes:

[0022] S1: An endoscopic device using photodynamic imaging technology acquires the lesion area to obtain raw image information. The raw image information is preprocessed and three-dimensional reconstruction is performed to obtain a three-dimensional restored structure. The raw image information includes image information corresponding to the photodynamic response information.

[0023] S2: Based on the photodynamic reaction database, P initial reaction data sets are set according to the reaction type. The initial reaction data sets include photodynamic agent injection status, light source irradiation conditions, and exposure time.

[0024] S3: Based on the treatment effect evaluation database, set up Q historical response data sets according to the treatment effect. The historical response data sets include pre-treatment image information, mid-treatment image information, and post-treatment image information.

[0025] In practical applications, it has been found that due to the influence of the complex internal environment of organisms, the photodynamic imaging images acquired by endoscopes may have problems such as noise, low contrast, and unclear details, which directly affect the doctor's accurate judgment of the lesion range, the degree of lesion, and the treatment effect. Therefore, it is necessary to process the raw images of endoscopes, such as noise suppression, smoothing, brightness and contrast adjustment, in order to improve the basic image quality.

[0026] However, conventional endoscopic image processing methods often fail to fully utilize the specific information of photodynamic reactions, such as changes in fluorescence intensity and distribution of photodynamic agents. Conventional endoscopic image processing uses all empirical data as training data for deep learning, resulting in a long processing time for the endoscopic image of the trained model, which cannot quickly respond to imaging requests from the application.

[0027] Based on this, this application accurately identifies and enhances the fluorescence signal of the photodynamic response, reduces interference from irrelevant signals, and simultaneously achieves precise display of the three-dimensional structure of the lesion. It can also quantitatively analyze the distribution of the photodynamic agent in three-dimensional space. Considering individual differences and real-time changes during treatment, it combines the photodynamic response database and the treatment effect evaluation database to achieve dynamic optimization of image processing parameters. It can flexibly enhance images according to the actual situation of each patient, adaptively find the best image enhancement parameters, and quickly respond to imaging requests and perform targeted image enhancement. The lesion area is the target area of ​​photodynamic therapy, and image preprocessing refers to basic image processing, such as noise suppression, smoothing, brightness and contrast adjustment.

[0028] The lesion area is precisely observed and images are acquired using an endoscope equipped with photodynamic imaging capabilities to obtain raw image information. The raw image information includes fluorescence signals reflecting the photodynamic response and other related structural information. Image preprocessing operations are performed on the acquired raw image information, such as noise removal, contrast enhancement, brightness correction, and color balance adjustment, to facilitate the subsequent accurate identification of photodynamic response characteristics.

[0029] By using multiple continuously acquired images, a three-dimensional structural model of the lesion area can be constructed through three-dimensional reconstruction using three-dimensional animation modeling software (such as 3DStudio Max, Maya, or any other modeling software). This helps to observe the lesion from different angles and depths, and to determine its boundaries, shape, and relationship with surrounding tissues.

[0030] Based on the photodynamic response database, P different initial response datasets are set according to the response type. Each initial response dataset includes information such as the specific type, dosage, and injection method of the photodynamic agent, as well as the wavelength, power, irradiation time, and mode of the light source. Based on the treatment effect evaluation database, Q historical response datasets are set according to the treatment effect. Each historical response dataset records the image information of the entire process from the early, middle, and late stages of treatment, including physiological indicators such as changes in tumor volume, angiogenesis, and the degree of inflammatory response.

[0031] It should be noted that the types of reactions mentioned include, but are not limited to, photoinduced fluorescence reactions (fluorescence reactions emitted by specific molecules after being excited by light of a specific wavelength, and the corresponding fluorescence signals can be used to detect and locate target substances), photoinduced quenching reactions (the fluorescence intensity of some substances changes after being exposed to light (e.g., quenching)), photodynamic oxidation reactions (in photodynamic therapy, some photosensitizers can generate oxidative stress under light, leading to cell or tissue damage), photoinduced electron transfer reactions (involving the electron transfer process that occurs in photosensitizers under light, which helps to understand the mechanism of action of photosensitizers, thereby optimizing their application in photodynamic imaging), photoisomerization reactions (some molecules undergo structural changes (isomerization) under light, and this reaction data helps to analyze the stability and activity of molecules in photodynamic imaging), and photodynamic drug release reactions (some photosensitizers or drug carriers can release drugs under specific light, which is used to evaluate the efficiency and specificity of drug release).

[0032] S4: Upon receiving an imaging request, based on the P sets of initial reaction data, configure the initial state of the endoscope image processing module and connect it to the three-dimensional display terminal corresponding to the stereoscopic reconstruction structure.

[0033] S5: Based on the image information corresponding to the photodynamic response information in the preprocessed image information, extract the photodynamic response features, which include fluorescence intensity features, signal distribution features, and edge features;

[0034] Based on the different photodynamic agents and light source parameters corresponding to the P initial reaction data sets, the working parameters of the endoscope image processing module are adjusted to accurately identify and analyze the corresponding photodynamic reaction characteristics; the endoscope image processing module is connected to the three-dimensional display terminal to ensure that the processed image information can be transmitted to the three-dimensional display interface in real time, so as to intuitively observe and analyze the three-dimensional reconstruction structure;

[0035] Based on the preprocessed image information, key regions reflecting the photodynamic response are located; the parts related to the photodynamic response are segmented, with a focus on the image information of the fluorescence signal portion. Further, the fluorescence intensity values ​​of each pixel or region of interest in the image are quantitatively analyzed. By calculating parameters such as average intensity, maximum intensity, and intensity distribution curves, the activity level of the photodynamic response is revealed. The spatial distribution patterns of the fluorescence signal are analyzed, such as concentration, diffusion range, and morphological characteristics. These signal distribution characteristics are used to determine the penetration and uniformity of drug distribution within the lesion. The clarity of the fluorescence signal boundary is detected, and an edge detection algorithm is used to extract the boundary contour between the lesion area and normal tissue to determine the effective range of the photodynamic response and whether the lesion boundary is clear.

[0036] The extracted fluorescence intensity features, signal distribution features, and edge features are integrated into the stereoscopic reconstruction structure to achieve visual annotation and dynamic presentation of the photodynamic response features on the stereoscopic reconstruction structure. On the three-dimensional display terminal, representative response features are highlighted through color coding, transparency adjustment, and other methods, providing technical support for more accurate diagnosis and treatment planning.

[0037] S6: Connect to the treatment effect evaluation database, combine the photodynamic response characteristics to perform synchronous evaluation and prediction of treatment effect, obtain the prediction evaluation results, and determine whether image-targeted enhancement is needed based on the prediction evaluation results;

[0038] S7: If image targeting enhancement is required, at the starting window time point of the preset time window, the image targeting enhancement command is activated in conjunction with the imaging request, and an iterative optimization search is performed in the optimization space corresponding to the Q historical response data sets to obtain image targeting enhancement parameters once.

[0039] S8: Using the image targeting enhancement parameters, cross-update the P initial reaction data sets corresponding to the initial state of the endoscope image processing module to obtain P experimental reaction data sets. Continue the loop until the termination condition is met, then perform image targeting enhancement at the cutoff window time point in the preset time window and output the visualization on the three-dimensional display terminal.

[0040] The acquired photodynamic response images of the current lesion area are integrated with the treatment effect evaluation database. Using machine learning or deep learning models, combined with photodynamic response characteristics, the current treatment effect is simultaneously evaluated and predicted to obtain the prediction evaluation results, which include the treatment effect level, the degree of lesion improvement, and potential risk assessment. Based on the prediction evaluation results, it is determined whether it is necessary to further improve the image quality for more accurate evaluation. If the prediction results show that there is insufficient treatment or the condition is complex and the image information is not clear enough, the image targeted enhancement process is initiated.

[0041] At the start window time point within the preset time window, an image targeting enhancement command activated by the imaging request is received; in the optimization space composed of Q sets of historical response data, an iterative optimization algorithm (such as genetic algorithm, particle swarm optimization, etc.) is used to find the image targeting enhancement parameters corresponding to the optimization space, involving gain adjustment, filter parameter adjustment, and contrast / brightness optimization;

[0042] After obtaining the corresponding image-targeting enhancement parameters after one iteration of optimization, the P initial response data sets of the endoscopic image processing module are compared with the image-targeting enhancement parameters and the initial state of the endoscopic image processing module. The sets are cross-updated to form new P experimental response data sets, which aim to simulate the changes in the image under different enhancement strategies, thereby finding the optimal enhancement scheme. The above optimization and update process is repeated until the preset termination conditions are met, such as reaching the predetermined image quality standard, the upper limit of the number of iterations, or the convergence of the optimization index. At the cutoff time point of the preset time window, the determined optimal image-targeting enhancement parameters are used to enhance the original image. After the enhanced image is 3D reconstructed, it is visualized in real time on a 3D display terminal. The refresh rate corresponding to the preset time window is not less than 30fps (below 30fps will cause a decrease in visual smoothness and affect the observation of details in the endoscopic image). The optimal image enhancement strategy is found quickly in a short time, which significantly improves the output quality of the endoscopic image.

[0043] Furthermore, the method of this application involves iterative optimization search within the optimization space corresponding to the Q sets of historical response data.

[0044] S71: In the endoscopic image processing module, historical poor quality scores corresponding to historical evaluation results are obtained from the treatment effect evaluation database to obtain M poor quality score data sets.

[0045] S72: By comparing the M sets of poor quality scores, extract N core poor quality parameter features, and set the search probability based on the M poor quality scores corresponding to the M sets of poor quality scores;

[0046] S73: Based on the N core inferior parameter features, the M inferior score data sets and the search probability, the data is clustered again according to the data correlation to establish an optimization space.

[0047] In the endoscopic image processing module, the evaluation results of historical treatment cases are statistically analyzed based on the stored treatment effect evaluation database. In particular, each historical treatment record is evaluated in detail, and a historical poor score is calculated for each case based on a preset evaluation system (such as phototoxicity, side effects, complications, etc.). The historical poor scores of all cases are summarized to form M poor score data sets, each set representing a specific historical treatment case.

[0048] For the above M sets of poor quality scores, in-depth mining and data analysis were conducted to identify N core poor quality parameters that are significantly related to the treatment effect. Among them, the core poor quality parameters include poor quality parameters corresponding to phototoxicity (the ability of different photosensitizers to generate reactive oxygen species, the degree of cell damage, the extent of tissue necrosis, etc. after photoactivation), poor quality parameters corresponding to side effects (recording adverse reactions that may occur during and after photodynamic therapy, such as skin photosensitivity, pain, edema, bleeding, infection, etc.), and poor quality parameters corresponding to complications (cases of complications specific to different tissue layers (such as muscle layer, etc.).

[0049] Based on the scores in each set of poor quality scores, a corresponding search probability is set. In general, cases with higher poor quality scores (i.e., cases with poor efficacy or large side effects) are assigned a higher search probability because they need to be considered during the optimization process. For the N core poor quality parameters extracted, the intrinsic relationship between the features and their correlation with the treatment effect are analyzed based on the data correlation between the N core poor quality parameters and historical poor quality scores.

[0050] Clustering algorithms (such as K-means, hierarchical clustering, DBSCAN, etc.) are used to re-divide the data, forming several clusters with similar characteristics and poor quality scores. This constructs an optimization space based on historical data, which includes various possible combinations of image processing parameters and their corresponding historical performance, providing a basic framework for subsequent iterative optimization searches. Within the constructed optimization space, appropriate optimization algorithms (such as genetic algorithms, particle swarm optimization, simulated annealing, etc.) are used for iterative optimization searches according to the set search probability weights. The goal of the search is to find a set of optimal photodynamic response parameters so that the endoscopic image processing module can better highlight lesion features and improve image quality in subsequent processing.

[0051] Furthermore, based on the N core inferiority parameter features, the M inferiority score data sets, and the search probability, the data is clustered and divided again according to the data correlation. The method of this application includes:

[0052] S731: Based on the M sets of poor quality scoring data and the search probability, perform optimization space search configuration to obtain search barrier index parameters, which include search minimum interval index parameters and barrier search boundary constraint index parameters.

[0053] S732: Perform an optimization configuration of the optimization space using the search barrier index parameters;

[0054] S733: Based on the N core inferior parameter features, the feature distribution is calculated in combination with the M inferior score data sets, and the optimization space is configured for secondary optimization according to the feature distribution calculation results.

[0055] Using the obtained M sets of poor quality scores and their corresponding search probabilities, the optimization space is configured for search. This configuration refers to setting the search path and priority based on the poor quality score of each historical case and its probability of being searched, thus determining the direction and intensity of the search. Search barrier parameters are defined. These parameters aim to avoid entering regions with high poor quality scores (i.e., poor performance) during the optimization process, instead favoring searches towards regions with lower poor quality scores (i.e., better performance). The search barrier parameters include a minimum search interval parameter and a search boundary constraint parameter. The minimum search interval parameter sets the minimum distance between adjacent parameter combinations during the search process to prevent overly dense searches. The search boundary constraint parameter limits the search range to avoid entering obviously ineffective parameter regions.

[0056] Based on the search barrier index parameters, a rough optimization configuration of the optimization space is performed, setting the initial search step size, search starting point, and boundary conditions to form a preliminary search path and strategy. Based on N core inferior parameter features, feature distribution calculations are performed on M inferior score datasets to analyze the distribution patterns of each feature variable in different score intervals. Based on the results of the feature distribution calculations, a more detailed and precise secondary configuration of the optimization space is performed. Furthermore, based on the trend and density distribution of feature variables in different score intervals, the search focus and direction are adjusted to concentrate the search on parameter regions that are strongly correlated with excellent treatment effects. At this point, it is also necessary to adjust the search step size, update the search starting point, or change the constraints of the search boundary to find the optimal solution or near-optimal solution more quickly during the optimization process.

[0057] By combining factors such as historical data quality scores, core parameter features and their correlation, and search probability, the configuration of the search space is gradually refined and optimized, thereby effectively finding the optimal parameter setting scheme in photodynamic imaging endoscopy image processing.

[0058] Furthermore, based on the N core inferiority parameter features, combined with the M inferiority score data sets, feature distribution calculation is performed, and the optimization space is configured for secondary optimization according to the feature distribution calculation results. The method of this application includes:

[0059] S7331: Taking the N core inferior quality parameter features as the center, perform distribution feature analysis based on the first inferior quality score data set, and construct the first feature distribution matrix;

[0060] S7332: Traverse the M sets of poor quality rating data to obtain the first feature distribution matrix, the second feature distribution matrix, ..., the Mth feature distribution matrix;

[0061] S7333: Based on the first feature distribution matrix, the second feature distribution matrix, ... the Mth feature distribution matrix, obtain the feature distribution calculation result.

[0062] N parameters closely related to treatment efficacy are extracted from M sets of poor-quality scoring data as core poor-quality parameter features. Taking the first set of poor-quality scoring data as an example, a detailed distribution feature analysis is performed on the N core parameter features. For example, the mean, variance, skewness, kurtosis and other statistics of each parameter are calculated, or histograms, scatter plots and other charts are drawn to intuitively show the parameter distribution. By comparing the parameter distribution, a first feature distribution matrix is ​​constructed. The first feature distribution matrix reflects the numerical distribution and interrelationship of each core parameter feature under the first set of poor-quality scoring data.

[0063] Traverse the remaining M-1 poor quality score data sets, perform the above distribution feature analysis on each set, and construct the corresponding feature distribution matrix, namely the second feature distribution matrix to the Mth feature distribution matrix. Each feature distribution matrix is ​​a multi-dimensional array, where the rows of the feature distribution matrix represent different core parameter features, and the columns of the feature distribution matrix represent the instance distribution of the corresponding feature in a specific poor quality score data set.

[0064] After obtaining M feature distribution matrices, the distribution patterns, variation trends, and interrelationships of each core parameter feature under different scoring levels are investigated. Based on the results of feature distribution calculation, the optimization space is refined. For example, if a certain core parameter feature is determined to show good treatment effect within a specific scoring interval, then this parameter region will be given priority consideration when the optimization algorithm searches for optimization. Through in-depth analysis of the feature distribution matrices, the search strategy can be optimized based on the original optimization space, such as adjusting the search step size, priority, or adding additional constraints, in order to quickly approach the optimal solution and improve the efficiency of endoscopic image processing.

[0065] Furthermore, such as Figure 2 As shown, based on the first feature distribution matrix, the second feature distribution matrix, ..., the Mth feature distribution matrix, the feature distribution calculation result is obtained. The method of this application includes:

[0066] S73331: Based on the damage parameter characteristics, the first feature distribution matrix, the second feature distribution matrix, ... the Mth feature distribution matrix are vertically stacked and synthesized into the first overall damage feature distribution matrix;

[0067] S73332: Based on the concurrency parameter characteristics, the first feature distribution matrix, the second feature distribution matrix, ... the Mth feature distribution matrix are horizontally stacked and synthesized into the first overall concurrency feature distribution matrix;

[0068] S73333: Calculate the first feature distribution characteristic factor using the first overall damage feature distribution matrix and the first overall concurrency feature distribution matrix;

[0069] S73334: Referencing the first characteristic distribution factor, calculate the second characteristic distribution factor, and obtain the characteristic distribution calculation result through the first characteristic distribution factor and the second characteristic distribution factor.

[0070] Based on the characteristics of damage parameters (such as the wavelength, power density, and illumination duration of different light sources (lasers, LEDs, etc.) corresponding to phototoxicity and side effects), the first feature distribution matrix to the Mth feature distribution matrix are stacked vertically. For each sample, the data of each feature distribution matrix in the dimension of damage parameters are stacked together in order to form a multidimensional array or long vector, thereby constructing the first overall damage feature distribution matrix. Similarly, based on the characteristics of concurrency parameters (such as the wavelength, power density, and illumination duration of different light sources (lasers, LEDs, etc.) corresponding to complications), the first feature distribution matrix to the Mth feature distribution matrix are stacked horizontally. The feature distribution matrices under different concurrency states under the same damage parameter are merged column by column to form a new two-dimensional matrix, namely the first overall concurrency feature distribution matrix.

[0071] By analyzing the overall characteristic distribution matrix of the first injury and the overall characteristic distribution matrix of the first concurrency, the first characteristic distribution factor reflecting the inherent laws and characteristics is extracted and calculated. The first characteristic distribution factor includes various statistical indicators, such as mean, variance, covariance, entropy, etc. Referring to the first characteristic distribution factor that has been obtained, similar or related calculation methods are used to explore and calculate the second characteristic distribution factor from other perspectives or at a deeper level, so as to comprehensively understand and quantify the characteristic distribution.

[0072] By combining the first and second characteristic distribution factors and employing objective weighting methods such as the coefficient of variation method, a weighted fusion of the first and second characteristic distribution factors is performed to calculate the feature distribution result. This approach comprehensively considers the impact of damage and concurrent states on the feature distribution, and through multi-perspective and in-depth analysis, improves the quantification accuracy and comprehensiveness of the feature distribution.

[0073] Furthermore, by using the image targeting enhancement parameters, the P initial response data sets corresponding to the initial state of the endoscopic image processing module are cross-updated to obtain P experimental response data sets. This process continues until the termination condition is met. The method of this application includes:

[0074] S81: Cross-update the first initial reaction data set in the P initial reaction data sets corresponding to the initial state of the endoscope image processing module to obtain the first test reaction data set, and repeat to obtain P test reaction data sets;

[0075] S82: Based on the P sets of experimental response data, generate prominent variation features, including contrast features, image texture features, and color characteristics;

[0076] S83: Determine whether to continue iterative optimization based on the termination condition corresponding to the preset optimization target.

[0077] Based on the first initial reaction data set among the P initial reaction data sets corresponding to the initial state of the endoscopic image processing module, the first image targeting enhancement parameters are applied to perform image targeting enhancement. Image targeting enhancement includes adjusting various attributes of the image, such as brightness, contrast, sharpness, color balance, etc., to improve image quality and lesion recognition ability. The processed results constitute the first experimental reaction data set.

[0078] Then, the process of adjusting the remaining (P-1) initial reaction data sets one by one is repeated P-1. A new experimental reaction data set is generated after each processing, until the cross-update of all P sets is completed. Feature extraction is performed on the P experimental reaction data sets obtained after cross-update to generate prominent change features. The prominent change features include, but are not limited to, contrast features (by comparing the gray level difference of the image before and after processing), image texture features (such as edge sharpness, patch distribution, etc.), and color characteristics (such as changes in chroma and saturation, etc.).

[0079] Based on the preset optimization objective, the generated prominent change features are used for evaluation; it is determined whether the current set of P experimental response data meets the preset termination condition, which may be reaching a certain image quality improvement standard, feature improvement threshold, or iteration limit; if the termination condition is not met, it returns to using new or adjusted image targeting enhancement parameters to perform a new round of cross-update optimization on all initial response data sets until the termination condition is met.

[0080] Furthermore, by cross-updating the P initial reaction data sets corresponding to the initial state of the endoscopic image processing module to obtain P experimental reaction data sets, this application also includes:

[0081] After activating the image targeting enhancement command, real-time image sequences are acquired through synchronous monitoring.

[0082] Receive the demand information and compare it with the real-time image sequence to determine whether to issue an interrupt command;

[0083] If the interrupt command is issued, cross-update backtracking is performed using the time information of the interrupt command.

[0084] The endoscope image processing module is started, and the image-targeted enhancement command is activated. Simultaneously, the image sequence acquired by the endoscope is monitored and acquired in real time. Continuous image frames from the endoscope are received and cached in real time to form a real-time image sequence. Specific requirements from the user are received, including requirements for detail preservation, processing speed, etc. The image enhancement distortion is compared between the currently cross-updated experimental response dataset and the received real-time image sequence to calculate the difference between the enhancement effect and the original image, and whether it meets the requirements in the request information.

[0085] If excessive distortion or failure to meet diagnostic requirements is detected during image enhancement, the system automatically determines whether to issue an interrupt command. Upon receiving an interrupt command, the system records the specific time information of the interrupt command. This time information is used as a reference point to achieve rapid backtracking during cross-updates. Specifically, based on the recorded time point, the system restores the state of the last valid test response data set before the interruption, or selects the corresponding intermediate result based on the timestamp. This ensures that the image processing flow can restart from the correct position, preventing further propagation and enhancement of invalid results. This ensures flexible handling of various emergencies during cross-updates in the image processing module and effectively guarantees image quality.

[0086] In summary, the beneficial effects of the embodiments of this application are:

[0087] 1. By integrating the photodynamic response database and the treatment effect evaluation database, and using advanced image preprocessing technology and three-dimensional reconstruction algorithms, a three-dimensional reconstruction structure of the lesion area can be generated, enabling real-time tracking of the photodynamic therapy effect.

[0088] 2. Within a preset time window, quickly find the optimal enhancement parameters and perform targeted image enhancement to dynamically adjust the output effect of the display terminal.

[0089] 3. By employing a method that simultaneously monitors and acquires real-time image sequences after activating the image-targeted enhancement command; receives demand information and compares it with the real-time image sequence to determine whether to issue an interrupt command; if an interrupt command is issued, cross-updates and backtracks are performed using the interrupt command's timing information. This ensures that the image processing flow can restart from the correct position, avoiding further propagation of invalid results and ensuring flexible handling of various unforeseen situations during cross-updates in the image processing module, effectively guaranteeing image quality.

[0090] Example 2

[0091] like Figure 3 The diagram shown is a structural schematic of an endoscopic imaging system provided in an embodiment of this application. Figure 3 The endoscopic imaging system shown is merely an example and should not be construed as limiting the functionality or scope of the embodiments of the present invention. Figure 3 As shown, the endoscopic imaging system includes a processor 21, a memory 22, an input device 23, and an output device 24; wherein the number of processors 21 can be one or more. Figure 3 Taking a processor 21 as an example, the processor 21, memory 22, input device 23, and output device 24 can be connected via a bus or other means. Figure 3 Taking the example of a connection between China and Israel via a bus.

[0092] The memory 22 can be used to store software programs, computer-executable programs, and modules, such as the program instructions / modules corresponding to the photodynamic imaging-based endoscopic image processing method in this embodiment. The processor 21 executes various functional applications and data processing of the computer device by running the software programs, instructions, and modules stored in the memory 22, thereby realizing the aforementioned photodynamic imaging-based endoscopic image processing method.

[0093] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0094] In summary, any step can be stored as a computer instruction or program in an unrestricted computer memory and can be called and recognized by an unrestricted computer processor; no further restrictions are imposed here.

[0095] Furthermore, the above technical solutions only embody the preferred technical solutions of the embodiments of this application. Any changes that those skilled in the art may make to certain parts of these solutions embody the novel principles of the embodiments of this application. Obviously, those skilled in the art can make various modifications and variations to this application without departing from the scope of this application.

Claims

1. An endoscopic image processing method based on photodynamic imaging, characterized in that, The method includes: Endoscopic devices using photodynamic imaging technology acquire images of the lesion area to obtain raw image information. The raw image information is then preprocessed and three-dimensional reconstruction is performed to obtain a three-dimensional restored structure. The raw image information includes image information corresponding to the photodynamic response information. Based on the photodynamic reaction database, P initial reaction data sets are set according to the reaction type. The initial reaction data sets include photodynamic agent injection status, light source irradiation conditions, and exposure time. Based on the treatment effect evaluation database, Q historical response data sets are set according to the treatment effect. The historical response data sets include pre-treatment image information, mid-treatment image information, and post-treatment image information. Upon receiving an imaging request, based on the P sets of initial reaction data, the initial state of the endoscope image processing module is configured, and it is connected to the three-dimensional display terminal corresponding to the stereoscopic reconstruction structure. Based on the image information corresponding to the photodynamic response information in the preprocessed image information, photodynamic response features are extracted, including fluorescence intensity features, signal distribution features, and edge features. The treatment effect evaluation database is connected, and the treatment effect is simultaneously evaluated and predicted in combination with the photodynamic response characteristics. The predicted evaluation results are obtained, and it is determined whether image-targeted enhancement is needed based on the predicted evaluation results. If image targeting enhancement is required, at the start window time point of the preset time window, the image targeting enhancement command is activated in conjunction with the imaging request, and an iterative optimization search is performed in the optimization space corresponding to the Q historical response data sets to obtain image targeting enhancement parameters once. Using the image targeting enhancement parameters, the P initial reaction data sets corresponding to the initial state of the endoscopic image processing module are cross-updated to obtain P experimental reaction data sets. This process continues until the termination condition is met. Then, image targeting enhancement is performed at the cutoff window time point in the preset time window, and the results are visualized on the 3D display terminal.

2. The endoscopic image processing method based on photodynamic imaging as described in claim 1, characterized in that, The method involves iterative optimization search within the optimization space corresponding to the Q sets of historical response data, including: In the endoscopic image processing module, historical poor scores corresponding to historical evaluation results are obtained based on the treatment effect evaluation database to obtain M poor score data sets; By comparing the M sets of poor quality rating data, N core poor quality parameter features are extracted, and the search probability is set based on the M poor quality ratings corresponding to the M sets of poor quality rating data; Based on the N core inferior parameter features, the M inferior score data sets, and the search probability, the data is clustered again according to the data correlation to establish an optimization space.

3. The endoscopic image processing method based on photodynamic imaging as described in claim 2, characterized in that, Based on the N core inferiority parameter features, the M inferiority score datasets, and the search probability, the data is clustered again according to the data correlation to establish an optimization space. The method includes: Based on the M sets of poor quality scores and the search probability, the optimization space search configuration is performed to obtain search barrier index parameters, which include search minimum interval index parameters and barrier search boundary constraint index parameters. The search space is configured for optimization using the search barrier index parameters. Based on the N core inferior parameter features, the feature distribution is calculated in combination with the M inferior score data sets, and the optimization space is configured for secondary optimization according to the feature distribution calculation results.

4. The endoscopic image processing method based on photodynamic imaging as described in claim 3, characterized in that, Based on the N core inferiority parameter features, and combined with the M inferiority score data sets, feature distribution is calculated, and the optimization space is configured for secondary optimization according to the feature distribution calculation results. The method includes: Using the N core inferiority parameter features as the center, a distribution feature analysis is performed based on the first inferiority score data set to construct a first feature distribution matrix; By traversing the M sets of poor quality rating data, we obtain the first feature distribution matrix, the second feature distribution matrix, ..., the Mth feature distribution matrix; Based on the first feature distribution matrix, the second feature distribution matrix, ..., the Mth feature distribution matrix, the feature distribution calculation result is obtained.

5. The endoscopic image processing method based on photodynamic imaging as described in claim 4, characterized in that, Based on the first feature distribution matrix, the second feature distribution matrix, ..., the Mth feature distribution matrix, the method for obtaining the feature distribution calculation result includes: Based on the damage parameter characteristics, the first feature distribution matrix, the second feature distribution matrix, ... the Mth feature distribution matrix are stacked vertically and synthesized into the first overall damage feature distribution matrix; Based on the concurrency parameter characteristics, the first feature distribution matrix, the second feature distribution matrix, ... the Mth feature distribution matrix are horizontally stacked and synthesized into the first overall concurrency feature distribution matrix; The first feature distribution characteristic factor is calculated using the first overall damage feature distribution matrix and the first overall concurrency feature distribution matrix. The second characteristic factor is calculated by referring to the first characteristic distribution characteristic factor, and the characteristic distribution calculation result is obtained by using the first characteristic distribution characteristic factor and the second characteristic distribution characteristic factor.

6. The endoscopic image processing method based on photodynamic imaging as described in claim 1, characterized in that, Using the image targeting enhancement parameters, the P initial response data sets corresponding to the initial state of the endoscopic image processing module are cross-updated to obtain P experimental response data sets. This process continues until a termination condition is met. The method includes: The first initial reaction data set in the P initial reaction data sets corresponding to the initial state of the endoscope image processing module is cross-updated to obtain the first experimental reaction data set, and this process is repeated to obtain P experimental reaction data sets. Based on the P sets of experimental response data, prominent variation features are generated, including contrast features, image texture features, and color characteristics. Based on the termination condition corresponding to the preset optimization goal, determine whether it is necessary to continue iterative optimization.

7. The endoscopic image processing method based on photodynamic imaging as described in claim 6, characterized in that, The method further includes cross-updating P initial response data sets corresponding to the initial state of the endoscopic image processing module to obtain P experimental response data sets. After activating the image targeting enhancement command, real-time image sequences are acquired through synchronous monitoring. Receive the demand information and compare it with the real-time image sequence to determine whether to issue an interrupt command; If the interrupt command is issued, cross-update backtracking is performed using the time information of the interrupt command.

8. An endoscopic imaging system, characterized in that, The endoscopic imaging system includes: Memory, used to store executable instructions; The processor, when executing executable instructions stored in the memory, is used to implement the steps of the endoscopic image processing method based on photodynamic imaging as described in any one of claims 1-7.

Images