A method and system for structured image analysis of pancreatic cystic lesions

Abstract

This invention proposes a structured image analysis method and system for pancreatic cystic lesions, belonging to the field of image processing technology. It collects lesion data through an electronic medical record system, and obtains classification data, basic clinical data, and basic imaging data through interpolation and image correction methods. Based on these classification, clinical, and imaging data, a structured pathological type classification model for pancreatic cystic lesions is constructed to distinguish the pathological types of pancreatic cystic lesions. This invention solves the problems of inaccurate classification due to missing original clinical data, the impact of bias in original imaging data on classification accuracy, the disordered data types and lack of standardized processing, and the difficulty in distinguishing pathological types of pancreatic cystic lesions due to the lack of a systematic approach, achieving accurate and efficient classification of pancreatic cystic tumor pathological types.

Description

Technical Field

[0001] This invention belongs to the field of image processing technology, and in particular relates to a method and system for structured image analysis of pancreatic cystic lesions. Background Technology

[0002] Pancreatic cystic lesions are a highly heterogeneous group of pancreatic tumors, encompassing various pathological types such as IPMN (intraductal papillary mucinous tumor), MCN (mucinous cystic tumor), SCN (serous cystic tumor), SPN (solid pseudopapillary tumor), and NET (cystic neuroendocrine tumor). The biological behavior, treatment options, and prognoses of different pathological types vary significantly, making accurate differentiation of pathological types a crucial prerequisite for clinical diagnosis and treatment decisions. Currently, the structured pathological classification of pancreatic cystic lesions mainly relies on physicians' clinical experience, single imaging examinations, or tumor marker detection. These methods have several shortcomings: First, raw clinical data is prone to numerical deficiencies, such as incomplete patient history records or missing laboratory test results, leading to incomplete input data and directly affecting the input quality and classification accuracy of the model. Second, raw imaging data is affected by factors such as differences in equipment models, scanning parameter settings, and patient positioning during imaging, resulting in data bias. This can easily lead to blurred lesion boundaries and inaccurate extraction of size features, thus affecting lesion detection results and ultimately reducing the accuracy of the diagnosis. The problems include: low classification accuracy; third, the diverse types of data (clinical, imaging, and tumor markers), with clinical data being mostly descriptive, imaging data being image-based, and tumor markers being numerical; and a lack of standardized processing procedures, leading to difficulties in effectively integrating data from different sources, low data utilization, and inconsistencies in judgment standards among physicians, resulting in poor classification consistency; fourth, the lack of a systematic classification model, with existing classification methods relying heavily on physicians' subjective judgments based on their personal experience, easily influenced by the physician's professional level and accumulated clinical experience, leading to missed or incorrect diagnoses, and failing to meet the needs of precise and standardized clinical diagnosis and treatment. Therefore, a technical solution is urgently needed to address these issues and improve the accuracy and consistency of pancreatic cystic tumor pathological classification. Summary of the Invention

[0003] To address the aforementioned problems in existing technologies, this invention proposes a structured image analysis method and system for pancreatic cystic lesions. This method overcomes the deficiencies in existing technologies regarding the classification of pancreatic cystic tumor pathological types, such as insufficient data quality, low classification accuracy, and poor consistency. It enables accurate and efficient classification of pancreatic cystic tumor pathological types, providing reliable support for clinical diagnosis and treatment.

[0004] To achieve the above objectives, the present invention employs the following technical solution: a method for structured imaging analysis of pancreatic cystic lesions, comprising: S1: Collect lesion data through an electronic medical record system; the lesion data includes categorical data and continuous data; the categorical data includes demographic data, which is collected with the patient's consent, and the continuous data includes raw clinical data and raw imaging data. S2: Determine whether there are any missing values ​​in the original clinical data. If so, proceed to S3; otherwise, use the original clinical data as the basic clinical data and proceed to S4. S3: Use the interpolation method to interpolate the missing original clinical data to obtain interpolated clinical data and use the interpolated clinical data as the basic clinical data; S4: Perform image correction on the original imaging data to obtain corrected imaging data and use the corrected imaging data as the basic imaging data; S5: Based on the classification data, basic clinical data and basic imaging data, construct a structured pathological type classification model for pancreatic cystic lesions, and use the structured pathological type classification model for pancreatic cystic lesions to distinguish the pathological types of pancreatic cystic lesions. S4 includes: S41: Acquire raw imaging data, which includes raw CT or MRI multiphase images of pancreatic cystic lesions at different scan slice thicknesses; S42: Perform slice thickness normalization correction on the original image data, resample the original images of all phases to the preset standard slice thickness, and perform preliminary correction on the long diameter of the lesion measured in the original image based on the ratio of the original slice thickness to the standard slice thickness. S43: Perform rigid registration processing on multi-phase images after slice thickness normalization to align the spatial positions of the pancreas and lesions, remove image layers whose spatial displacement exceeds the preset displacement threshold due to respiratory motion, and eliminate the long axis measurement deviation caused by respiratory motion artifacts. S44: Perform three-dimensional segmentation of the pancreatic lesion on the registered image, locate the pancreatic parenchymal region, extract the three-dimensional connected domain of the lesion, and obtain a complete three-dimensional model of the lesion; S45: Based on the three-dimensional model of the lesion, calculate the maximum Euclidean distance of the lesion in three-dimensional space, and use it as the initial corrected major axis of the lesion; S46: Perform volume effect correction on the edge region of the three-dimensional model of the lesion, remove some of the pseudo-enlarged regions caused by the volume effect, and obtain the final corrected long diameter of the pancreatic lesion as the corrected imaging data.

[0005] Further, S3 includes: S31: Obtain the missing raw clinical data and convert the raw clinical data into a raw matrix; the raw matrix includes multiple columns; S32: For the first column of the original matrix, extract all non-missing values ​​and calculate the lower quartile Q1 and upper quartile Q3 of all non-missing values; S33: Calculate the product of the difference between Q3 and Q1 and the boundary adjustment coefficient as the first variable. Subtract the first variable from Q1 to obtain the lower bound of the data anomaly. Add the first variable to Q3 to obtain the upper bound of the data anomaly. Find the out-of-bounds data in the first column of the matrix that exceed the upper or lower bound of the data anomaly and mark the out-of-bounds data as missing. S34: Repeat steps S32-S33 for the second column and other columns of the original matrix to obtain the corrected original matrix; S35: Standardize the original modified matrix in S34 using the Zscore function to obtain preprocessed clinical data; and save the mean mu and standard deviation sigma output by the Zscore function. S36: Use interpolation to interpolate the preprocessed clinical data to obtain interpolated clinical data.

[0006] Further, S36 includes: S36-1: Obtain the number of columns of preprocessed clinical data, define N imputation datasets; set preset variance threshold, change threshold, maximum number of iterations, and boundary adjustment coefficient; S36-2: Set the imputation dataset index k, k=1, ..., N. For the kth imputation dataset: assign the preprocessed clinical data to the kth imputation dataset; set an independent random seed to ensure the random perturbation of different datasets is differentiated; for each column of the kth imputation dataset, calculate the median of all non-missing values ​​in the column, generate a random perturbation matching the number of missing values, and use the sum of the median and the random perturbation to fill the missing positions in the column to obtain the initial data matrix imp_mat; S36-3: Perform iterative steps: (1) Perform iterative imputation on the kth imputation dataset: set the loop iteration number iter=1; define the columns with missing values ​​in the preprocessed clinical data as the target column list; (2) Initialize the temporary interpolation matrix temp_mat = imp_mat; (3) Iterate through each target column col in the target column list: a. Separate the non-missing value row index nidx and the missing value row index midx from temp_mat; b. Extract all columns of temp_mat based on nidx and remove the target column col as the prediction variable matrix; c. Calculate the column variance of the predictor variable matrix, retain the columns with variance greater than the preset variance threshold, and update the predictor variable matrix; d. Extract the non-missing values ​​of the target column 'col' in the nidx row as the observations of the target column; e. Input the updated predictor variable matrix and target column observations into the fitlm function to obtain the predictor model; f. Extract all columns of temp_mat based on midx and remove the target column col as the matrix to be interpolated, while retaining the columns selected in step c, to obtain the processed matrix to be interpolated; g. Input the model and the processed interpolation matrix into the predict function, and fill the prediction results into the col column and midx row of temp_mat; (4) Calculate the average of the absolute differences between the locations of all missing values ​​in temp_mat and imp_mat, denoted as delta; (5) If delta is less than the preset change threshold or iter is greater than or equal to the maximum number of iterations, the iteration ends and temp_mat is used as the interpolated matrix; otherwise, iter is set to iter+1, imp_mat is updated to temp_mat, and the iteration returns to step (3). S36-4: After the iteration is completed, the interpolated matrix is ​​used as the interpolation result of the kth interpolated dataset; the mean mu and standard deviation sigma saved in S35 are used column by column to restore the interpolation result by the method of "original value = standardized value × sigma + mu"; S36-5: Repeat S36-2 to S36-4 to obtain the restored interpolation results of N interpolation datasets; S36-6: Take the average of the N restored interpolation results by element to obtain the interpolated clinical data.

[0007] Further, the specific process of slice thickness normalization correction in step S42 is as follows: set the standard slice thickness to 1mm, and calculate the preliminary correction major diameter according to the formula L1=Lmeans×(Tslice / Tstandard), where Lmeans is the lesion major diameter measured in the original image, Tslice is the scan slice thickness of the original image, and Tstandard is the preset standard slice thickness.

[0008] Furthermore, the rigid registration process in step S43 employs a registration algorithm based on mutual information, using the venous phase image as a reference image, and aligning the arterial phase and equilibrium phase images with the reference image; the preset displacement threshold is 2 voxels.

[0009] Furthermore, the specific process of three-dimensional segmentation of pancreatic lesions in step S44 is as follows: using U-Net network or improved U-Net network, first segment the pancreatic parenchyma region, then segment the abnormally enhanced or low-enhanced lesion region within the pancreatic parenchyma region, extract the complete three-dimensional model of the lesion through 3D connected component analysis, and remove isolated artifact regions. The calculation process of the maximum Euclidean distance in the three-dimensional space in step S45 is as follows: traverse all pixels in the three-dimensional model of the lesion, calculate the Euclidean distance between any two pixels, and select the maximum Euclidean distance as the initial corrected major axis of the lesion.

[0010] Furthermore, the specific process of volume effect correction in step S46 is as follows: the edge region of the lesion three-dimensional model is shrunk by a distance of 0.5 voxels, the false region with blurred edges is removed, the core region of the lesion is retained, the maximum Euclidean distance of the shrunk lesion three-dimensional model is recalculated to obtain the final corrected major axis, and the final corrected major axis is used as the corrected imaging data.

[0011] Further, S5 includes: S51: For different structured pathological types of pancreatic cystic lesions, set different weights for classification data, basic clinical data, and basic imaging data; S52: Using the weighted weights, perform a weighted summation of the classification data, basic clinical data, and basic imaging data, and take the pancreatic cystic lesion type with the highest weighted summation score as the pancreatic cystic lesion pathological type.

[0012] This invention also proposes a structured image analysis system for pancreatic cystic lesions, used to perform the aforementioned structured image analysis method for pancreatic cystic lesions. The system includes a data acquisition module, a data processing module, and a result output module. The data acquisition module is connected to the data processing module, and the data processing module is connected to the result output module. The data acquisition module collects lesion data through an electronic medical record system. The data processing module generates classification data, basic clinical data, and basic imaging data based on the lesion data. The result output module constructs a structured pathological type classification model for pancreatic cystic lesions based on the classification data, basic clinical data, and basic imaging data, and uses this model to distinguish the pathological types of pancreatic cystic lesions.

[0013] The beneficial technical effects of this invention compared with the prior art are as follows: (1) By supplementing the missing original clinical data through the interpolation method, the problem of inaccurate classification caused by the lack of original clinical data is solved, and the incomplete model input due to missing data is avoided, thereby improving the reliability of the classification model; (2) By processing the original imaging data through image correction, the problem of the deviation of the original imaging data affecting the classification accuracy is solved, and the interference in the image acquisition process (such as equipment error, imaging deviation, etc.) is eliminated, ensuring the accuracy of the basic imaging data and providing data support for accurate classification; (3) Clearly distinguishing between classification data (demographic characteristics) and continuous data (original clinical and imaging data), and performing targeted processing (missing interpolation, image correction) on the two types of data respectively, solving the problem of messy data types and lack of standardized processing, realizing the standardization and structuring of data, and laying a standardized data foundation for model construction; (4) By integrating various types of data after structured processing, a special pathological type classification model is constructed, providing a standardized and feasible classification process, replacing the traditional classification method that relies on experience judgment or single data, improving the accuracy and consistency of pathological type differentiation, and solving the problem of the difficulty in differentiating pathological types of pancreatic cystic lesions and the lack of systematic methods. Attached Figure Description

[0014] To more clearly illustrate the embodiments of the present invention or the technical solutions in 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 merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0015] Figure 1 This is a simplified flowchart of a structured imaging analysis method for pancreatic cystic lesions according to the present invention. Figure 2 This is a simplified structural diagram of a structured imaging analysis system for pancreatic cystic lesions according to the present invention. Detailed Implementation

[0016] 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.

[0017] The concepts involved in this application will first be described with reference to the accompanying drawings. It should be noted that the following descriptions of various concepts are only for the purpose of making the content of this application easier to understand and do not constitute a limitation on the scope of protection of this application; furthermore, the embodiments and features in the embodiments of this application can be combined with each other unless otherwise specified. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0018] Refer to the instruction manual. Figure 1 This invention proposes a structured imaging analysis method for pancreatic cystic lesions, comprising: S1: collecting lesion data through an electronic medical record system; the lesion data includes categorical data and continuous data; the categorical data includes demographic data, collected with the patient's consent, including age and sex; the categorical data also includes whether there is a capsule, whether the common bile duct is dilated, whether the cyst wall is enhanced, the thickness of the cyst wall, whether the lymph nodes are enlarged, the intracystic septa, the enhancement of the intracystic septa, the lesion morphology, the lesion location, the number of lesions (single or multiple), whether the main pancreatic duct is dilated, whether the pancreatic duct is connected, whether the wall nodules are enhanced, whether there are wall nodules, whether there are wall nodules in the main pancreatic duct, whether the pancreatic parenchyma is atrophied, whether the solid components are enhanced, whether the cyst wall is homogeneous, whether the septa are homogeneous, and whether the perivascular area is clear; the continuous data includes raw clinical data and raw imaging data, the clinical data including serum carcinoembryonic antigen (CEA) and serum carbohydrate antigen 19-9. 19-9) Level, etc., the above-mentioned original imaging data may include MRI or CT image data such as the diameter of the common bile duct (mm), the diameter of the main pancreatic duct (mm), the long diameter of the largest mural nodule (mm), the long diameter of the lesion (mm), the long diameter of the solid component (mm), the short diameter of the largest lymph node (mm), the short diameter of the largest mural nodule (mm), the short diameter of the lesion (mm), and the short diameter of the solid component (mm); S2: Determine whether there are missing values ​​in the original clinical data. If so, proceed to S3; otherwise, use the original clinical data as the basic clinical data and proceed to S4. S3: Imput the missing original clinical data using the interpolation method to obtain the interpolated clinical data, and use the interpolated clinical data as the basic clinical data; S4: Perform image correction on the original imaging data to obtain the corrected imaging data, and use the corrected imaging data as the basic imaging data; S5: Based on the classification data, basic clinical data, and basic imaging data, construct a structured pathological type classification model for pancreatic cystic lesions, and use the structured pathological type classification model for pancreatic cystic lesions to distinguish the pathological types of pancreatic cystic lesions; S4 includes: S41: Acquire raw imaging data, which includes raw CT or MRI multiphase images of pancreatic cystic lesions at different scan slice thicknesses; S42: Perform slice thickness normalization correction on the original image data, resample the original images of all phases to the preset standard slice thickness, and perform preliminary correction on the long diameter of the lesion measured in the original image based on the ratio of the original slice thickness to the standard slice thickness. S43: Perform rigid registration processing on multi-phase images after slice thickness normalization to align the spatial positions of the pancreas and lesions, remove image layers whose spatial displacement exceeds the preset displacement threshold due to respiratory motion, and eliminate the long axis measurement deviation caused by respiratory motion artifacts. S44: Perform three-dimensional segmentation of the pancreatic lesion on the registered image, locate the pancreatic parenchymal region, extract the three-dimensional connected domain of the lesion, and obtain a complete three-dimensional model of the lesion; S45: Based on the three-dimensional model of the lesion, calculate the maximum Euclidean distance of the lesion in three-dimensional space, and use it as the initial corrected major axis of the lesion; S46: Perform volume effect correction on the edge region of the three-dimensional model of the lesion, remove some of the pseudo-enlarged regions caused by the volume effect, and obtain the final corrected long diameter of the pancreatic lesion as the corrected imaging data.

[0019] The interpolation method in S3 described above may include: S31: Obtain the missing original clinical data and convert the original clinical data into an original matrix; the original matrix includes multiple columns; it is understood that the above multiple columns may include serum carcinoembryonic antigen data columns, serum carbohydrate antigen 19-9 data columns, etc.

[0020] S32: For the first column of the original matrix, extract all non-missing values ​​and calculate the lower quartile Q1 and upper quartile Q3 of all non-missing values; S33: Calculate the product of the difference between Q3 and Q1 and the boundary adjustment coefficient as the first variable. Subtract the first variable from Q1 to obtain the lower bound of the data anomaly. Add the first variable to Q3 to obtain the upper bound of the data anomaly. Find the out-of-bounds data in the first column of the matrix that exceed the upper or lower bound of the data anomaly and mark the out-of-bounds data as missing. S34: Repeat steps S32-S33 for the second column and other columns of the original matrix to obtain the corrected original matrix; through the above steps, out-of-bounds values ​​in the original clinical data can be removed to avoid the out-of-bounds values ​​affecting the subsequent iterative model.

[0021] S35: Use the Zscore function in the MATLAB toolkit to standardize the original modified matrix in S34 to obtain preprocessed clinical data; and save the mean mu and standard deviation sigma output by the Zscore function. S36: Use interpolation to interpolate the preprocessed clinical data to obtain interpolated clinical data.

[0022] S36 includes: S36-1: Obtain the number of preprocessed clinical data columns and define N imputation datasets; set preset variance threshold, change threshold, maximum number of iterations, and boundary adjustment coefficient; here, the preset variance threshold, change threshold, maximum number of iterations, and boundary adjustment coefficient are all predefined constant variables, and are not limited here.

[0023] S36-2: Set the imputation dataset index k, k=1, ..., N. For the kth imputation dataset: assign the preprocessed clinical data to the kth imputation dataset; set an independent random seed to ensure the random perturbation of different datasets is differentiated; for each column of the kth imputation dataset, calculate the median of all non-missing values ​​in the column, generate a random perturbation matching the number of missing values, and use the sum of the median and the random perturbation to fill the missing positions in the column to obtain the initial data matrix imp_mat; S36-3: Perform iterative steps: (1) Perform iterative imputation on the kth imputation dataset: set the loop iteration number iter=1; define the columns with missing values ​​in the preprocessed clinical data as the target column list; (2) Initialize the temporary interpolation matrix temp_mat = imp_mat; (3) Iterate through each target column col in the target column list: a. Separate the non-missing value row index nidx and the missing value row index midx from temp_mat; b. Extract all columns of temp_mat based on nidx and remove the target column col as the prediction variable matrix; c. Calculate the column variance of the predictor variable matrix, retain the columns with variance greater than the preset variance threshold, and update the predictor variable matrix; d. Extract the non-missing values ​​of the target column 'col' in the nidx row as the observations of the target column; e. Input the updated predictor variable matrix and target column observations into the fitlm function to obtain the predictor model; f. Extract all columns of temp_mat based on midx and remove the target column col as the matrix to be interpolated, while retaining the columns selected in step c, to obtain the processed matrix to be interpolated; g. Input the model and the processed interpolation matrix into the predict function, and fill the prediction results into the col column and midx row of temp_mat; (4) Calculate the average of the absolute differences between the locations of all missing values ​​in temp_mat and imp_mat, denoted as delta; (5) If delta is less than the preset change threshold or iter is greater than or equal to the maximum number of iterations, the iteration ends and temp_mat is used as the interpolated matrix; otherwise, iter is set to iter+1, imp_mat is updated to temp_mat, and the iteration returns to step (3). S36-4: After the iteration is completed, the interpolated matrix is ​​used as the interpolation result of the kth interpolated dataset; the mean mu and standard deviation sigma saved in S35 are used column by column to restore the interpolation result by the method of "original value = standardized value × sigma + mu"; S36-5: Repeat S36-2 to S36-4 to obtain the restored interpolation results of N interpolation datasets; S36-6: Take the average of the N restored interpolation results by element to obtain the interpolated clinical data.

[0024] The specific process of slice thickness normalization correction in step S42 is as follows: A standard slice thickness of 1 mm is set, and the preliminary corrected long diameter is calculated according to the formula L1 = Lmeans × (Tslice / Tstandard), where Lmeans is the lesion long diameter measured in the original image, Tslice is the original image's scan slice thickness, and Tstandard is the preset standard slice thickness. Through the above slice thickness normalization correction, preliminary image correction is achieved, reducing measurement errors caused by different scan slice thicknesses. The rigid registration processing in step S43 uses a registration algorithm based on mutual information. A venous phase image with relatively stable imaging quality and clear anatomical structure is used as a reference image. The arterial phase and equilibrium phase images are spatially aligned with the reference image, thereby reducing the measurement error of the lesion long diameter caused by misalignment between multiple phase images due to the patient's respiratory movements. Simultaneously, a preset displacement threshold of 2 voxels is set. During the registration process, the spatial displacement of each image layer is judged. When the displacement of a certain layer exceeds this threshold, it is determined that the layer is excessively affected by respiratory movements and is discarded to further ensure the accuracy of subsequent measurements. The specific process of three-dimensional segmentation of pancreatic lesions in step S44 is as follows: The registered image is segmented sequentially using a U-Net network or an improved U-Net network. First, the pancreatic parenchyma region is segmented and located. Then, within the pancreatic parenchyma region, lesion regions with abnormal enhancement or low enhancement features are further segmented. Based on the completion of two-dimensional segmentation, the spatially continuous parts of the lesion are merged through 3D connected component analysis to extract a complete three-dimensional model of the lesion. Artifact regions or noise regions that are too small or isolated are removed to ensure that the three-dimensional model can truly reflect the spatial morphology of the lesion. The calculation process of the maximum Euclidean distance in three-dimensional space in step S45 is as follows: All voxel points in the three-dimensional model of the lesion are traversed. The Euclidean distance between any two voxel points in three-dimensional space is calculated. The maximum value among all Euclidean distances is selected as the initial corrected major diameter of the lesion. The maximum distance in three-dimensional space replaces the maximum diameter on the traditional two-dimensional slice, which can more realistically and comprehensively reflect the actual size of the lesion. The specific process of volume effect correction in step S46 is as follows: the edge region of the lesion three-dimensional model is shrunk by a distance of 0.5 voxels to remove some of the "false enlargement region" caused by the volume effect, while retaining the core region of the lesion. The maximum Euclidean distance of the shrunk lesion three-dimensional model is recalculated to obtain the final corrected major axis, and the final corrected major axis is used as the corrected imaging data.

[0025] S5 includes: S51: setting different weights for classification data, basic clinical data, and basic imaging data for different structured pathological types of pancreatic cystic lesions; S52: using the weights to perform a weighted summation of classification data, basic clinical data, and basic imaging data, and taking the pancreatic cystic lesion type with the highest weighted summation score as the pancreatic cystic lesion pathological type.

[0026] Specifically, for example, for IPMN pathology types, weighted averages can be set for categorical data, basic clinical data, and basic imaging data to obtain an IPMN scoring model: .

[0027] Where FL represents categorical data, FL i1 This represents the i1th categorical data point, where i1 represents the categorical data index, and n1 represents the total number of categorical data points. It includes binary categorical variables such as gender and whether the cyst wall is uniform (a binary categorical variable has two types, each with different values), and multivariate categorical variables such as age, vascular invasion, and intracystic septa (a multivariate categorical variable has multiple different values, with multiple segments defined and assigned values ​​for each segment); LC represents the basic clinical data. i2 This represents the i2th basic clinical data point, where i2 represents the basic clinical data index and n2 represents the total number of basic clinical data points. These clinical data typically require laboratory testing and other methods to obtain. YX represents basic imaging data. i3 This represents the i3rd basic imaging data, where i3 represents the sequence number of the basic imaging data, and n3 represents the total number of basic imaging data. Basic imaging data requires the use of imaging instruments (CT scanner, MRI scanner, etc.) to obtain lesion characteristic data; A1, A i1 B i2 and C i3 The above weighting is used for the weighting.

[0028] For MCN, SCN, SPN, and NET pathological types, different weightings need to be set for classification data, basic clinical data, and basic imaging data compared to IPMN pathological types. The same weighted items are summed to obtain the corresponding weighted sum score, and the pathological type with the highest weighted sum score is determined as the pancreatic cystic lesion pathological type.

[0029] Refer to the instruction manual. Figure 2This invention also proposes a structured image analysis system for pancreatic cystic lesions, used to execute the aforementioned structured image analysis method for pancreatic cystic lesions. The system includes a data acquisition module, a data processing module, and a result output module. The data acquisition module is connected to the data processing module, and the data processing module is connected to the result output module. The data acquisition module collects lesion data through an electronic medical record system. The data processing module generates classification data, basic clinical data, and basic imaging data based on the lesion data. The result output module constructs a structured pathological type classification model for pancreatic cystic lesions based on the classification data, basic clinical data, and basic imaging data, and uses this model to distinguish the pathological types of pancreatic cystic lesions. The data acquisition module, data processing module, and result output module can be implemented on big data processing platforms such as Hadoop, embedded systems such as FPGA and DSP, or web development frameworks such as HTML.

[0030] The embodiments and / or implementation methods described above are merely preferred embodiments and / or implementation methods for implementing the technology of the present invention, and are not intended to limit the implementation methods of the technology of the present invention in any way. Any person skilled in the art can make some modifications or alterations to other equivalent embodiments without departing from the scope of the technical means disclosed in the content of the present invention, but they should still be regarded as the technology or embodiments that are substantially the same as the present invention.

[0031] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. The above descriptions are only preferred embodiments of this application. It should be noted that due to the limitations of written expression, while there are objectively infinite specific structures, those skilled in the art can make several improvements, modifications, or changes without departing from the principles of this application, and can also combine the above technical features in an appropriate manner. These improvements, modifications, changes, or combinations, or the direct application of the inventive concept and technical solution to other situations without modification, should all be considered within the scope of protection of this application.

Claims

1. A method for structured imaging analysis of pancreatic cystic lesions, characterized in that: include: S1: Collect lesion data through an electronic medical record system; the lesion data includes categorical data and continuous data; the categorical data includes demographic data, and the continuous data includes raw clinical data and raw imaging data; S2: Determine whether there are any missing values ​​in the original clinical data. If so, proceed to S3; otherwise, use the original clinical data as the basic clinical data and proceed to S4. S3: Preprocess the missing original clinical data to obtain imputed clinical data, and use the imputed clinical data as the basic clinical data; S4: Perform image correction on the original imaging data to obtain corrected imaging data and use the corrected imaging data as the basic imaging data; S5: Based on the classification data, basic clinical data and basic imaging data, construct a structured pathological type classification model for pancreatic cystic lesions, and use the structured pathological type classification model for pancreatic cystic lesions to distinguish the pathological types of pancreatic cystic lesions. S4 includes: S41: Acquire raw imaging data, which includes raw CT or MRI multiphase images of pancreatic cystic lesions at different scan slice thicknesses; S42: Perform slice thickness normalization correction on the original image data, resample the original images of all phases to the preset standard slice thickness, and perform preliminary correction on the long diameter of the lesion measured in the original image based on the ratio of the original slice thickness to the standard slice thickness. S43: Perform rigid registration processing on multi-phase images after slice thickness normalization to align the spatial positions of the pancreas and lesions, remove image layers whose spatial displacement exceeds the preset displacement threshold due to respiratory motion, and eliminate the long axis measurement deviation caused by respiratory motion artifacts. S44: Perform three-dimensional segmentation of the pancreatic lesion on the registered image, locate the pancreatic parenchymal region, extract the three-dimensional connected domain of the lesion, and obtain a complete three-dimensional model of the lesion; S45: Based on the three-dimensional model of the lesion, calculate the maximum Euclidean distance of the lesion in three-dimensional space, and use it as the initial corrected major axis of the lesion; S46: Perform volume effect correction on the edge region of the three-dimensional model of the lesion, remove some of the pseudo-enlarged regions caused by the volume effect, and obtain the final corrected long axis of the pancreatic lesion as the corrected imaging data. In S3, the clinical data obtained after intercalation include: S36-1: Obtain the number of columns of preprocessed clinical data, define N imputation datasets; set preset variance threshold, change threshold, maximum number of iterations, and boundary adjustment coefficient; S36-2: Set the imputation dataset index k, k=1, ..., N. For the kth imputation dataset: assign the preprocessed clinical data to the kth imputation dataset; set an independent random seed to ensure the random perturbation of different datasets is differentiated; for each column of the kth imputation dataset, calculate the median of all non-missing values ​​in the column, generate a random perturbation matching the number of missing values, and use the sum of the median and the random perturbation to fill the missing positions in the column to obtain the initial data matrix imp_mat; S36-3: Perform iterative steps: (1) Perform iterative imputation on the kth imputation dataset: set the loop iteration number iter=1; define the columns with missing values ​​in the preprocessed clinical data as the target column list; (2) Initialize the temporary interpolation matrix temp_mat = imp_mat; (3) Iterate through each target column col in the target column list: a. Separate the non-missing value row index nidx and the missing value row index midx from temp_mat; b. Extract all columns of temp_mat based on nidx and remove the target column col as the prediction variable matrix; c. Calculate the column variance of the predictor variable matrix, retain the columns with variance greater than the preset variance threshold, and update the predictor variable matrix; d. Extract the non-missing values ​​of the target column 'col' in the nidx row as the observations of the target column; e. Input the updated predictor variable matrix and target column observations into the fitlm function to obtain the predictor model; f. Extract all columns of temp_mat based on midx and remove the target column col as the matrix to be interpolated, while retaining the columns selected in step c, to obtain the processed matrix to be interpolated; g. Input the model and the processed interpolation matrix into the predict function, and fill the prediction results into the col column and midx row of temp_mat; (4) Calculate the average of the absolute differences between the locations of all missing values ​​in temp_mat and imp_mat, denoted as delta; (5) If delta is less than the preset change threshold or iter is greater than or equal to the maximum number of iterations, the iteration ends and temp_mat is used as the interpolated matrix; otherwise, iter is updated to increment it by 1, imp_mat is updated to temp_mat, and the iteration returns to step (3). S36-4: After the iteration is completed, the interpolated matrix is ​​used as the interpolation result of the kth interpolated dataset; the mean mu and standard deviation sigma saved in S35 are used column by column to restore the interpolation result by the method of "original value = standardized value × sigma + mu"; S36-5: Repeat S36-2 to S36-4 to obtain the restored interpolation results of N interpolation datasets; S36-6: Take the average of the N restored interpolation results by element to obtain the interpolated clinical data.

2. The method for structured imaging analysis of pancreatic cystic lesions according to claim 1, characterized in that, Preprocessing of missing raw clinical data includes: S31: Obtain the missing raw clinical data and convert the raw clinical data into a raw matrix; the raw matrix includes multiple columns; S32: For the first column of the original matrix, extract all non-missing values ​​and calculate the lower quartile Q1 and upper quartile Q3 of all non-missing values; S33: Calculate the product of the difference between Q3 and Q1 and the boundary adjustment coefficient as the first variable. Subtract the first variable from Q1 to obtain the lower bound of the data anomaly. Add the first variable to Q3 to obtain the upper bound of the data anomaly. Find the out-of-bounds data in the first column of the matrix that exceed the upper or lower bound of the data anomaly and mark the out-of-bounds data as missing. S34: Repeat steps S32-S33 for the second column and other columns of the original matrix to obtain the corrected original matrix; S35: Standardize the original modified matrix in S34 using the Zscore function to obtain preprocessed clinical data; and save the mean mu and standard deviation sigma output by the Zscore function.

3. The method for structured imaging analysis of pancreatic cystic lesions according to claim 1, characterized in that, The specific process of slice thickness normalization correction in step S42 is as follows: set the standard slice thickness to 1mm, and calculate the preliminary correction major diameter according to the formula L1=Lmeans×(Tslice / Tstandard), where Lmeans is the lesion major diameter measured in the original image, Tslice is the scan slice thickness of the original image, and Tstandard is the preset standard slice thickness.

4. The method for structured imaging analysis of pancreatic cystic lesions according to claim 1, characterized in that, The rigid registration process in step S43 uses a registration algorithm based on mutual information, taking the venous phase image as the reference image and aligning the arterial phase and equilibrium phase images with the reference image; the preset displacement threshold is 2 voxels.

5. The method for structured imaging analysis of pancreatic cystic lesions according to claim 1, characterized in that, The specific process of three-dimensional segmentation of pancreatic lesions in step S44 is as follows: using U-Net network or improved U-Net network, first segment the pancreatic parenchyma region, then segment the abnormally enhanced or low-enhanced lesion region within the pancreatic parenchyma region, extract the complete three-dimensional model of the lesion through 3D connected component analysis, and remove isolated artifact regions. The calculation process of the maximum Euclidean distance in the three-dimensional space in step S45 is as follows: traverse all pixels in the three-dimensional model of the lesion, calculate the Euclidean distance between any two pixels, and select the maximum Euclidean distance as the initial corrected major axis of the lesion.

6. The method for structured imaging analysis of pancreatic cystic lesions according to claim 1, characterized in that, The specific process of volume effect correction in step S46 is as follows: shrink the edge region of the lesion three-dimensional model by a distance of 0.5 voxels, remove the false regions with blurred edges, retain the core region of the lesion, recalculate the maximum Euclidean distance of the shrunken lesion three-dimensional model, obtain the final corrected major axis, and use the final corrected major axis as the corrected imaging data.

7. The method for structured imaging analysis of pancreatic cystic lesions according to claim 1, characterized in that, S5 includes: S51: For different structured pathological types of pancreatic cystic lesions, set different weights for classification data, basic clinical data, and basic imaging data; S52: Using the weighted weights, perform a weighted summation of the classification data, basic clinical data, and basic imaging data, and take the pancreatic cystic lesion type with the highest weighted summation score as the pancreatic cystic lesion pathological type.

8. A structured image analysis system for pancreatic cystic lesions, used to execute the structured image analysis method for pancreatic cystic lesions according to any one of claims 1-7, characterized in that, The system includes a data acquisition module, a data processing module, and a result output module. The data acquisition module is connected to the data processing module, and the data processing module is connected to the result output module. The data acquisition module is used to collect lesion data through an electronic medical record system. The data processing module is used to generate classification data, basic clinical data, and basic imaging data based on the lesion data. The result output module is used to construct a structured pathological type classification model for pancreatic cystic lesions based on the classification data, basic clinical data, and basic imaging data, and to distinguish the pathological types of pancreatic cystic lesions using the structured pathological type classification model for pancreatic cystic lesions.

Images