Laparoscopic cholecystectomy difficulty degree quantification analysis method and system based on ultrasonic imageomics and electronic equipment

Abstract

The embodiment of the application discloses a laparoscopic cholecystectomy difficulty degree quantification analysis method and system based on ultrasonic imageomics and an electronic device. The method comprises the following steps: adjusting the contrast of the gallbladder region and the background tissue by a local contrast adjustment algorithm, eliminating speckle noise, and obtaining a preprocessed image. The preprocessed image is divided into multiple region blocks, the average gray value of each region block is calculated, and a candidate region is determined. The gallbladder contour feature is screened by using local gradient distribution data, a multi-scale edge detection algorithm is applied to extract preliminary edge information, and region expansion is performed to generate ROI data. The extracted ultrasonic imageomics features of the gallbladder are normalized, core image features are screened, and a quantification analysis model is constructed and optimized in combination with the clinical data of the patient. The clinical data and images of a new patient are used to generate a quantitative score for predicting the difficulty of cholecystectomy. The method can provide accurate difficulty quantification evaluation for clinical cholecystectomy.

Description

Technical Field

[0001] This invention relates to the field of auxiliary equipment for cholecystectomy, and in particular to a method, system and electronic device for quantitatively analyzing the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics. Background Technology

[0002] Laparoscopic cholecystectomy (LC) is the preferred method for treating gallstones, cholecystitis, and other diseases, and has gradually replaced traditional open surgery due to its advantages of minimal invasiveness and rapid recovery. However, the difficulty of LC surgery varies significantly depending on individual patient differences and pathological complexity. Especially in cases with abnormal anatomical structures or severe inflammation, conversion to open surgery or treatment of bile duct injuries may be necessary during the operation, thereby increasing the risk of intraoperative complications and postoperative recovery. Although the widespread adoption of laparoscopic surgery and advancements in technology have significantly increased the success rate of the procedure, predicting the surgical difficulty and preoperative preparation still face multiple challenges.

[0003] Currently, the assessment of the difficulty of gallbladder (LC) surgery relies heavily on the surgeon's experience and subjective judgment. Surgeons assess the procedure preoperatively using imaging data such as ultrasound and CT scans; however, these assessments are primarily based on individual experience rather than standardized quantitative analysis. Therefore, highly subjective judgments can lead to insufficient preoperative preparation, increasing the unpredictability and risk of the surgery. This limitation of subjective judgment is mainly reflected in the complexity of image interpretation. Ultrasound images, due to resolution limitations and noise interference, present challenges for surgeons when interpreting the gallbladder region. Different surgeons may interpret images differently due to varying experience, resulting in inconsistent predictions of surgical difficulty.

[0004] With the development of medical imaging technology and machine learning applications in industry and other fields, radiomics technology has provided new possibilities for extracting and analyzing high-dimensional data from medical images. Radiomics can extract hundreds or even thousands of quantitative features from standard medical images, including shape features, texture features, and statistical features, thereby reflecting subtle differences in images that are not visible to the naked eye. However, regrettably, no research has yet applied ultrasound radiomics to analyze and evaluate the ease or difficulty of laparoscopic cholecystectomy. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a method and system for quantitative analysis of the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics, which facilitates the quantitative analysis and assessment of the difficulty of laparoscopic cholecystectomy.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A quantitative analysis method for the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics includes: performing preoperative ultrasound examinations on a large number of patients, acquiring ultrasound images showing the gallbladder region and saving them in DICOM format; Based on the local contrast adjustment algorithm, the gray level difference of different regions in the ultrasound image is dynamically adjusted to enhance the contrast between the gallbladder region and the background tissue, and to eliminate speckle noise in the ultrasound image to obtain a preprocessed image. The preprocessed image is divided into multiple regions, the average gray value of each region is calculated, and regions whose average gray value exceeds a preset gray value threshold and whose gray values ​​are uniform are marked as candidate regions. Within the candidate region, the local gradient values ​​of the pixels in the candidate region are calculated to form local gradient distribution data; Based on the local gradient distribution data, suspected gallbladder contour regions that meet the gallbladder contour characteristics are screened and marked. For the suspected gallbladder contour region, a multi-scale edge detection algorithm is applied to extract the preliminary edge information of the suspected gallbladder contour region; Based on the preliminary edge information, a region expansion algorithm is executed to expand the preliminary edges to cover the complete outline of the gallbladder. The expanded complete outline is marked as the region of interest of the gallbladder, and ROI data is generated. Based on the ROI data, gallbladder ultrasound radiomics features are extracted, including: shape features, texture features, and grayscale distribution features of the gallbladder region; The radiomics features are normalized to the same scale to form normalized feature data. Based on the normalized feature data, the stability and discriminative power of each feature are calculated, and features that do not meet the stability and discriminative power criteria are removed to obtain the selected target core image omics features. Based on the selected target core radiomics features and the patient's clinical data, a quantitative analysis model for predicting the difficulty of cholecystectomy is constructed. The quantitative analysis model includes an imaging feature model, a clinical feature model, and a comprehensive prediction model combining imaging and clinical features. The sample dataset used for model building is divided into a training set and a test set; the quantitative analysis model is trained and its parameters are tuned on the training set, and the accuracy, sensitivity and specificity of the quantitative analysis model are verified on the test set, and the best quantitative analysis model after training is output. The patient's current imaging features and clinical data are input into the optimal quantitative analysis model to obtain and output a quantitative score of surgical difficulty.

[0007] According to a specific embodiment of the present invention, the step of dynamically adjusting the grayscale differences of different regions in the ultrasound image based on a local contrast adjustment algorithm to enhance the contrast between the gallbladder region and the background tissue, and to eliminate speckle noise in the ultrasound image, to obtain a preprocessed image includes: The ultrasound image is divided into multiple overlapping local regions. For each local region, the gray-level histogram of that region is calculated and gray-level equalization is performed to enhance the gray-level contrast within the local region. For each local region, calculate the local brightness parameter of that local region to characterize the overall gray-level center value of that region; and calculate the gray-level dispersion of each local region to quantify the local contrast level, in order to guide the intensity adjustment of contrast enhancement. Based on the local brightness parameters and grayscale dispersion of each local region, the grayscale values ​​of each pixel in the local region are adjusted to enhance the contrast between local regions, so as to reduce the grayscale difference between the gallbladder region and the background tissue. During the local contrast enhancement process, each local region of the ultrasound image is processed pixel by pixel based on the filtering algorithm to detect and suppress noise signals. The enhancement results of each local region are merged into a complete image, and the pixel values ​​in each overlapping region are weighted and averaged to make the image transition between different overlapping regions smooth, thus obtaining a preprocessed image.

[0008] According to a specific embodiment of the present invention, the step of dividing the preprocessed image into multiple region blocks, calculating the average gray value of each region block, and marking the region blocks whose average gray value exceeds a preset gray value threshold and whose gray value is uniform as candidate regions includes: dividing the preprocessed image into rectangular region blocks of equal size, and calculating the average gray value of each region block; Z-score normalization is applied to the average gray value of each region block to ensure that the gray values ​​of each region block are on a uniform scale. Regions with standardized gray values ​​exceeding a preset gray threshold and uniform gray levels are marked as candidate regions.

[0009] According to a specific embodiment of the present invention, the step of applying a multi-scale edge detection algorithm to the suspected gallbladder contour region to extract preliminary edge information of the suspected gallbladder contour region includes: The resolution of images of suspected gallbladder contour areas is gradually reduced according to a predetermined ratio to generate low-resolution image layers of different scales, forming a multi-scale image sequence. For each scale image in the multi-scale image sequence, the gradient magnitude and gradient direction of each pixel are calculated to identify significant edge features; In the gradient magnitude of the image at each scale, examine the two adjacent pixels along the gradient direction. If the gradient magnitude of a pixel is not greater than the gradient magnitude of its neighboring pixels, then the gradient value of that pixel is set to zero to suppress non-edge responses and create clearer edge lines. Double threshold segmentation is performed on the gradient image after non-maximum suppression to extract the edge map; By layering the edge maps from all scales of the image, a comprehensive edge map containing multi-scale information is obtained, forming the preliminary edge information of the suspected gallbladder contour region.

[0010] According to a specific embodiment of the present invention, the step of progressively reducing the resolution of the image of the suspected gallbladder contour region by a predetermined ratio to generate low-resolution image layers of different scales, forming a multi-scale image sequence, includes: The resolution of the original image of the suspected gallbladder outline area is gradually reduced according to a predetermined ratio to generate low-resolution image layers of different scales. For each image layer, a pixel at a location in the current image layer is selected as the processing center point, and a neighborhood range is set to determine a group of neighboring pixels around that pixel; the size of the neighborhood range is controlled by a radius k, that is, it includes the pixel area around the center point; A weight matrix is ​​generated based on the Gaussian distribution function to form a Gaussian kernel; wherein each element of the Gaussian kernel represents a Gaussian weight value, and the level of the weight value is positively correlated with the distance from the center pixel; The weighted value is obtained by multiplying the gray value of each pixel in the neighborhood of the center point with the corresponding Gaussian weight value. The weighted values ​​are summed to obtain the weighted sum of all pixels in the neighborhood of the center point; Divide the accumulated weighted sum by the sum of all weights to obtain the smoothed grayscale value of the center pixel; The calculated smoothed gray value is assigned to the gray value of the center pixel, thus completing the smoothing process for one pixel. Move the processing center to the next pixel position and repeat the above steps to smooth the entire image pixel by pixel until the smoothing of the entire image is completed. The smoothed current layer image is downsampled to proportionally reduce the image size, thus generating a lower-resolution image for the next layer.

[0011] According to a specific embodiment of the present invention, the step of performing a region expansion algorithm based on the preliminary edge information to expand the preliminary edges to cover the complete contour of the gallbladder, marking the expanded complete contour as the region of interest (ROI) of the gallbladder, and generating ROI data includes: Based on the data carried by the preliminary edge information, a region expansion algorithm is executed to gradually expand from the preliminary edge to the complete outline of the gallbladder region in order to capture the boundary of the gallbladder. During the expansion process, the similarity between the edge and its neighborhood is calculated, and edge points whose similarity does not meet the preset requirements are removed. The expanded, complete contour region is marked as the ROI region of the gallbladder, and the corresponding ROI data is generated.

[0012] According to a specific embodiment of the present invention, the step of calculating the stability and discriminative power of each feature based on the normalized feature data, and removing features that do not meet the stability and discriminative power criteria to obtain the screened target core feature data includes: calculating the inter-group correlation coefficient for each normalized feature, and screening features with a consistency higher than 0.75; Features related to surgical difficulty are selected using statistical methods, and the most representative features are retained using the principles of minimum redundancy and maximum relevance to form the target core feature data.

[0013] According to a specific embodiment of the present invention, the step of constructing a quantitative analysis model for predicting the difficulty of cholecystectomy based on the screened target core feature data and the patient's clinical data includes: Multiple sub-models were constructed from the selected patient clinical data according to feature types; among them... A demographic sub-model is established based on patient demographic data, including patient age, gender, and body mass index. A laboratory test sub-model was established based on the patient's preoperative blood test data, which included white blood cell count, C-reactive protein, and liver function indicators. A medical history sub-model is established based on the patient's medical history information, which includes whether the patient has diabetes, hypertension, and a history of surgery. Based on the prediction results generated by each of the sub-models, a comprehensive clinical model is established; the comprehensive clinical model is configured to assess the patient's overall clinical risk based on a comprehensive analysis of the prediction values ​​of each sub-model. Furthermore, based on the selected target core radiomics features, a corresponding weight is assigned to each radiomics feature, and its weighted value is calculated according to the weight of each feature. The weighted values ​​of each feature are accumulated to generate an image score, which is calculated using the following formula: Rad-score=∑(feature value×feature weight)+b0, where b0 is a cutoff value used to adjust the overall offset of the score. The image score is used to quantitatively evaluate the impact of image features on the difficulty of surgery. The scoring results of the radiomics model are weighted and fused with the predicted values ​​of the integrated clinical model to generate a comprehensive predictive model that combines imaging and clinical features.

[0014] Secondly, embodiments of the present invention also provide a quantitative analysis system for the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics, comprising: The first acquisition unit is used to perform preoperative ultrasound examinations on a large number of patients, acquire ultrasound images showing the gallbladder region and save them in DICOM format. The image enhancement processing unit is used to dynamically adjust the grayscale differences of different regions in the ultrasound image based on a local contrast adjustment algorithm, so as to enhance the contrast between the gallbladder region and the background tissue and eliminate speckle noise in the ultrasound image to obtain a preprocessed image. The first marking unit is used to divide the preprocessed image into multiple regions, calculate the average gray value of each region, and mark the regions whose average gray value exceeds a preset gray value threshold and whose gray values ​​are uniform as candidate regions. The first calculation unit calculates the local gradient values ​​of pixels in the candidate region within the candidate region to form local gradient distribution data. The second marking unit is used to filter and mark the suspected gallbladder contour region that meets the gallbladder contour features based on the local gradient distribution data. The edge information extraction unit is used to extract the preliminary edge information of the suspected gallbladder contour region by applying a multi-scale edge detection algorithm to the suspected gallbladder contour region. The ROI determination unit, based on the preliminary edge information, executes a region expansion algorithm to expand the preliminary edges to cover the complete outline of the gallbladder, marks the expanded complete outline as the region of interest of the gallbladder, and generates ROI data; The radiomics feature extraction unit is used to extract gallbladder ultrasound radiomics features based on the ROI data. The gallbladder ultrasound radiomics features include: shape features, texture features, and grayscale distribution features of the gallbladder region. The normalization processing unit is used to normalize the image omics features, standardizing each feature to the same scale to form normalized feature data; The radiomics feature screening unit is used to calculate the stability and discriminative power of each feature based on the normalized feature data, and to remove features that do not meet the stability and discriminative power criteria in order to obtain the screened target core radiomics features. The model building unit is used to build a quantitative analysis model for predicting the difficulty of cholecystectomy based on the screened target core radiomics features and the patient's clinical data. The quantitative analysis model includes an image feature model, a clinical feature model, and a comprehensive prediction model combining image and clinical features. The model optimization unit is used to divide the sample dataset used for model construction into a training set and a test set; to train and fine-tune the quantitative analysis model on the training set; and to verify the accuracy, sensitivity and specificity of the quantitative analysis model on the test set, and output the best quantitative analysis model after training. The quantitative analysis unit is used to input the current patient's imaging feature data and clinical data into the optimal quantitative analysis model to obtain and output a quantitative score of the difficulty of gallbladder surgery.

[0015] Thirdly, embodiments of the present invention also provide an electronic device, comprising: one or more processors; a memory; wherein the memory stores one or more executable programs, and the one or more processors read the executable program code stored in the memory and run a quantification analysis program for the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics corresponding to the executable program code, so as to execute the quantification analysis method for the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics as described in any of the first aspects.

[0016] The present invention provides a method, system, and electronic equipment for quantitative analysis of the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics. Through a local contrast adjustment algorithm, it dynamically enhances the grayscale difference between the gallbladder region and background tissue while simultaneously suppressing speckle noise interference, resulting in a preprocessed image used for subsequent candidate region labeling. This solves the problem of blurred images and difficult interpretation often faced by surgeons when judging the difficulty of surgery due to the limitations of traditional ultrasound imaging caused by low resolution and noise interference. It lays a data quality guarantee for accurate gallbladder region identification and quantitative feature extraction. Furthermore, through multi-step image processing of the preprocessed image, the contour region of the gallbladder is automatically identified and segmented. After candidate region screening and preliminary edge information extraction, multi-scale edge detection and region expansion algorithms are used to generate regions of interest (ROIs), ensuring that the complete contour of the gallbladder region is accurately labeled. This solves the problem of inconsistent judgments caused by surgeons' experience and subjective factors, making subsequent analysis more consistent and objective. After obtaining the standardized ROI region, radiomics features, including subtle features such as shape, texture, and grayscale distribution, are extracted. Because radiomics features can reveal multidimensional quantitative information about the gallbladder region, information that is often difficult to identify directly with the naked eye, radiomics features provide a comprehensive feature set that reflects gallbladder imaging information. This provides a high-dimensional data foundation for assessing surgical difficulty. Furthermore, the quantitative analysis model models imaging features and patient clinical data separately, ultimately forming an imaging feature model, a clinical feature model, and a combined predictive model, making surgical difficulty prediction more comprehensive and accurate. Compared to relying solely on the doctor's subjective judgment, the quantitative output of the comprehensive predictive model provides a clear and objective score for surgical difficulty. This reduces the risk of variability and misjudgment caused by individual experience and subtle differences in images that are difficult to identify with the naked eye, facilitating quantitative analysis and assessment of the difficulty of laparoscopic cholecystectomy, thus providing a standardized and quantifiable preoperative assessment reference. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a method for quantitatively analyzing the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the image preprocessing procedure for image enhancement and speckle noise reduction of gallbladder ultrasound images in one embodiment of the present invention; Figure 3 This is a schematic diagram of the preliminary edge information extraction process for the gallbladder contour region in one embodiment of the present invention; Figure 4 This is a schematic diagram of an electronic device structure according to an embodiment of the present invention. Detailed Implementation

[0018] The following is a detailed description of the renal lesion ultrasound analysis and diagnosis auxiliary system based on renal ultrasound imaging provided by the present invention, with reference to the accompanying drawings; This invention provides a method for quantitatively analyzing the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics, applicable to preoperative quantitative analysis and assessment of the difficulty of laparoscopic cholecystectomy in clinical settings. By processing and analyzing the features of preoperative ultrasound images from a large number of patients, a quantitative model is established, aiming to provide an objective and unified analytical standard to assist physicians in preoperative risk assessment and optimization of preoperative decisions.

[0019] See Figure 1 The method for quantitatively analyzing the difficulty of laparoscopic cholecystectomy includes: S110, performing preoperative ultrasound examinations on a large number of patients, obtaining ultrasound images showing the gallbladder area, and saving the ultrasound images in DICOM format.

[0020] In this embodiment, an ultrasound examination was performed on the patient undergoing laparoscopic cholecystectomy using an Aplio i800 ultrasound diagnostic instrument with an i5LX5 probe frequency of 2~5MHz to ensure high-quality grayscale images were obtained.

[0021] During the examination, the patient lies supine to fully expose the area to be examined. The patient fasted for 8 hours prior to the procedure, and the examination was performed by an ultrasound physician with over 5 years of experience to ensure consistent image quality.

[0022] During ultrasound examination, the ultrasound equipment parameters are adjusted to ensure optimal image quality, acquiring ultrasound imaging data of the gallbladder region. Emphasis is placed on saving the clearest, highest-resolution long-axis image of the gallbladder neck, and the image data is saved in DICOM format. DICOM format facilitates the inclusion of complete image information and metadata, such as imaging parameters and patient information, ensuring objectivity and standardization for subsequent analysis.

[0023] S120. Based on the local contrast adjustment algorithm, dynamically adjust the gray level difference of different regions in the ultrasound image to enhance the contrast between the gallbladder region and the background tissue, and eliminate speckle noise in the ultrasound image to obtain a preprocessed image. Specifically, by processing the image through a local contrast adjustment algorithm, the grayscale differences in different regions are dynamically adjusted to enhance the contrast between the gallbladder region and the background tissue, making the gallbladder boundary clearer. Furthermore, speckle noise can be eliminated through a filtering algorithm, which can significantly improve image quality and lay a good foundation for subsequent candidate region recognition and feature extraction.

[0024] Specifically, speckle noise in the ultrasound image can be eliminated using a neighborhood filtering method. For each pixel, the grayscale difference between it and its eight surrounding pixels is calculated. If the grayscale difference is large, the pixel is considered to be noise. The pixel is then smoothed based on the average value of the surrounding pixels, thereby reducing speckle noise and making the image smoother.

[0025] S130. Divide the preprocessed image into multiple regions, calculate the average gray value of each region, and mark the regions whose average gray value exceeds a preset gray value threshold and whose gray values ​​are uniform as candidate regions.

[0026] Specifically, in this embodiment, the preprocessed image can be divided into multiple small regions, such as 16x16 pixels. The average gray value of each small region is calculated, and candidate regions that meet the conditions are selected according to preset brightness and uniformity standards. Regions with gray values ​​exceeding the threshold and good uniformity are initially marked as possible gallbladder candidate regions.

[0027] S140. Within the candidate region, calculate the local gradient values ​​of the pixels in the candidate region to form local gradient distribution data.

[0028] S150. Based on the local gradient distribution data, filter and mark the suspected gallbladder contour regions that meet the gallbladder contour features; within the candidate regions, calculate the local gradient value pixel by pixel to generate local gradient distribution data, and identify and mark the suspected gallbladder contour regions.

[0029] S160. Apply a multi-scale edge detection algorithm to the suspected gallbladder contour region to extract the preliminary edge information of the suspected gallbladder contour region. S170. Based on the preliminary edge information, execute the region expansion algorithm to expand the preliminary edge to cover the complete outline of the gallbladder, mark the expanded complete outline as the region of interest of the gallbladder, and generate ROI data. Alternatively, ITK-SNAP can be used to manually segment the gallbladder region to obtain the Region of Interest (ROI). However, to ensure the stability and consistency of the segmentation results, the segmentation process must be independently performed by at least two sonographers with more than 10 years of experience, and then repeated on N randomly selected cases. The consistency and reproducibility of the segmentation results are evaluated by calculating the intraclass correlation coefficient (ICC), ensuring the reliability of subsequent radiomics feature extraction.

[0030] S180. Based on the ROI data, extract gallbladder ultrasound radiomics features, which include: shape features, texture features, and grayscale distribution features of the gallbladder region.

[0031] S190. Normalize the image omics features by standardizing each feature to the same scale to form normalized feature data. S200. Based on the normalized feature data, calculate the stability and discriminative power of each feature, and remove features that do not meet the stability and discriminative power standards to obtain the selected target core image omics features. In this embodiment, radiomics features can also be extracted from the ROI using the pyradiomics tool, including but not limited to shape features, texture features, and wavelet transform features. These radiomics features need to cover the geometry, texture patterns, and internal structure information of the gallbladder. Through feature extraction, high-dimensional features of the ultrasound image are obtained, such as shape features (e.g., volume and surface area), grayscale features (e.g., pixel intensity distribution), texture features (e.g., grayscale co-occurrence matrix features), and wavelet transform features. These radiomics features provide a rich data foundation for subsequent machine learning model training and also help assess the difficulty of the surgery.

[0032] S210. Based on the screened target core radiomics features and the patient's clinical data, a quantitative analysis model for predicting the difficulty of cholecystectomy is constructed. The quantitative analysis model includes an image feature model, a clinical feature model, and a comprehensive prediction model combining image and clinical features. S220. Divide the sample dataset used for model building into a training set and a test set; train and fine-tune the quantitative analysis model on the training set, and verify the accuracy, sensitivity and specificity of the quantitative analysis model on the test set, and output the best quantitative analysis model after training. S230. Input the current patient's imaging feature data and clinical data into the optimal quantitative analysis model to obtain a quantitative score of surgical difficulty and output it.

[0033] Understandably, during ultrasound imaging, factors such as patient respiration and changes in probe angle can easily cause interference from speckle noise and artifacts, leading to unstable image quality. This makes segmentation of the gallbladder region in ultrasound images difficult and inconsistent in quality. Furthermore, the contrast between the gallbladder and surrounding tissues, such as the liver and bile ducts, is low in ultrasound images, especially in cases of high fat content or inflammation, where the gallbladder boundaries become more blurred, further increasing the difficulty of segmentation.

[0034] In this embodiment of the invention, a local contrast adjustment algorithm dynamically enhances the grayscale difference between the gallbladder region and the background tissue while simultaneously suppressing speckle noise interference, resulting in a preprocessed image. This image is then used for subsequent candidate region labeling. This solves the problem of blurred images and difficult interpretation often faced by surgeons when assessing surgical difficulty due to the limitations of traditional ultrasound imaging, such as low resolution and noise interference from speckle and artifacts. It lays a foundation for accurate gallbladder region identification and quantitative feature extraction, ensuring data quality.

[0035] Furthermore, through multi-step image processing of the preprocessed images, the contour region of the gallbladder is automatically identified and segmented. After candidate region selection and preliminary edge information extraction, multi-scale edge detection and region expansion algorithms are used to generate regions of interest (ROIs), ensuring that the complete contour of the gallbladder region is accurately marked. This solves the problem of inconsistency in judgment caused by doctors' experience and subjective factors, making subsequent analysis more consistent and objective. After obtaining the standardized ROI regions, radiomics features are extracted, including subtle features such as shape, texture, and grayscale distribution. Because radiomics features can reveal multi-dimensional quantitative information about the gallbladder region, this information is usually difficult to directly identify with the naked eye. Unlike subjective assessments where doctors rely on limited image information, radiomics feature extraction establishes a feature set that comprehensively reflects gallbladder image information, providing a high-dimensional data foundation for surgical difficulty assessment. Moreover, in the quantitative analysis model, image features and patient clinical data are modeled separately, ultimately forming an image feature model, a clinical feature model, and a comprehensive prediction model combining the two, making surgical difficulty prediction more comprehensive and accurate. Compared to relying solely on the doctor's subjective judgment, the quantitative output of a comprehensive predictive model can provide a clear and objective score for the difficulty of the surgery. This can reduce the risk of discrepancies and misjudgments in the analysis results caused by the doctor's individual experience and the inability to recognize subtle differences in images with the naked eye. It facilitates the quantitative analysis and assessment of the difficulty of laparoscopic cholecystectomy, thus providing a standard and quantifiable preoperative assessment reference analysis.

[0036] In ultrasound image processing, ultrasound images often suffer from low grayscale contrast and severe speckle noise due to factors such as patient size, tissue density, and probe angle variations, making image analysis difficult. In some other applications, existing technologies typically employ simple global contrast adjustment and noise filtering, but they cannot achieve precise contrast enhancement and noise suppression for the gallbladder region. This may result in the gallbladder region being unable to effectively distinguish from background tissue, thereby affecting the accuracy of segmentation and feature extraction.

[0037] In this embodiment, local adjustment and other ultrasound image preprocessing methods are used to dynamically enhance the contrast between the gallbladder region and the background tissue, and effectively suppress noise interference.

[0038] See Figure 2 In some embodiments, the step of dynamically adjusting the grayscale differences in different regions of the ultrasound image based on a local contrast adjustment algorithm to enhance the contrast between the gallbladder region and the background tissue, and to eliminate speckle noise in the ultrasound image to obtain a preprocessed image (step S120) includes: S121. The ultrasound image is divided into multiple overlapping local regions. For each local region, a gray-level histogram is calculated, and gray-level equalization is performed to enhance the gray-level contrast within the local region. Furthermore, image detail information can be preserved simultaneously. S123. For each local region, a local brightness parameter is calculated to characterize the overall gray-level center value of the region. Additionally, the gray-level dispersion of each local region is calculated to quantify the local contrast level, which guides the intensity adjustment of contrast enhancement.

[0039] S125. Based on the local brightness parameters and grayscale dispersion of each local area, adjust the grayscale values ​​of each pixel in the local area to enhance the contrast between local areas, so as to reduce the grayscale difference between the gallbladder area and the background tissue.

[0040] Specifically, for low-brightness, high-dispersion areas, the contrast is enhanced; while for high-brightness, low-dispersion areas, grayscale expansion is reduced to avoid excessive noise enhancement.

[0041] S127. During the local contrast enhancement process, each local region of the ultrasound image is processed pixel by pixel based on the filtering algorithm to detect and suppress noise signals.

[0042] In this embodiment, the aforementioned neighborhood filtering algorithm can be used to perform noise detection and suppression on a pixel-by-pixel basis in the local area, thereby eliminating speckle noise.

[0043] S129. The enhancement results of each local region are merged into a complete image. The pixel values ​​in each overlapping region are weighted and averaged to achieve a smooth transition between different overlapping regions, resulting in a preprocessed image. Specifically, the enhanced local regions are stitched and merged, and the pixel values ​​of the overlapping parts are weighted and averaged to achieve a smooth transition of the image, ultimately generating a complete preprocessed image. Not only is the overall contrast improved, but there are also no obvious boundary breaks or noise interference, providing a high-quality image foundation for subsequent segmentation.

[0044] In some embodiments, dividing the preprocessed image into multiple region blocks, calculating the average gray value of each region block, and marking region blocks with average gray values ​​exceeding a preset gray threshold and uniform gray values ​​as candidate regions includes: dividing the preprocessed image into rectangular region blocks of equal size, and calculating the average gray value of each region block; Z-score normalization is applied to the average gray value of each region block to ensure that the gray values ​​of each region block are on a uniform scale. By applying Z-score normalization, the differences in gray values ​​between regions caused by factors such as imaging conditions and brightness variations are eliminated, thereby improving the reliability and accuracy of candidate region selection.

[0045] Regions with standardized grayscale values ​​exceeding a preset grayscale threshold and exhibiting uniform grayscale are marked as candidate regions. Specifically, based on the standardized grayscale values, regions meeting the following criteria are selected as candidate regions: Standardized grayscale values ​​exceeding a preset grayscale threshold: These are marked as high-brightness regions, used for initial screening to identify areas that may contain the gallbladder. The threshold can be dynamically adjusted based on actual data, for example, set to regions exceeding 1.5 times the standard deviation.

[0046] For regions that meet the grayscale value standard, their grayscale uniformity is further calculated. Specifically, the standard deviation of the pixel grayscale values ​​within the region is calculated. If the standard deviation is lower than a set threshold, the grayscale distribution of the region is considered to be relatively uniform.

[0047] In this embodiment, the above screening can accurately identify candidate regions that may contain the gallbladder area, avoiding mislabeling of other background tissues or high-noise areas, which helps to improve the accuracy of gallbladder segmentation.

[0048] See Figure 3 In some embodiments, the step of applying a multi-scale edge detection algorithm to the suspected gallbladder contour region to extract preliminary edge information of the suspected gallbladder contour region (step S160) includes: S161. The resolution of the image of the suspected gallbladder contour area is gradually reduced according to a predetermined ratio to generate low-resolution image layers of different scales, forming a multi-scale image sequence. In this embodiment, the suspected gallbladder contour region is decomposed into multiple scales. The image of the suspected gallbladder contour region is progressively reduced in size according to a certain ratio to generate low-resolution image layers of different scales. By repeatedly performing smoothing and downsampling operations, a series of images with progressively decreasing resolution are obtained, which is called a multi-scale image sequence.

[0049] For each scale of the image, the image smoothing process can be performed using the following formula: ;in, Indicates the first Location in layer-scale image pixel values, This represents the lower-resolution image from the previous layer. Let k be a two-dimensional Gaussian kernel, where k is the radius of the kernel. Represents the distance relative to the center pixel within a filter window or convolution kernel. The offset distance, (x+u, y+v), represents the coordinates of all pixels within the current filtering window; during the filtering operation based on this filtering algorithm formula, the algorithm will rotate around the center pixel. The neighboring pixels (x+u, y+v) are traversed one by one, and the gray values ​​of the neighboring pixels are weighted and calculated in combination with the filter weights to obtain a smoothing or denoising effect at the center position.

[0050] S163. For each scale image in the multi-scale image sequence, calculate the gradient magnitude and gradient direction of each pixel to identify significant edge features. The gradient direction includes the horizontal gradient. and vertical gradient The calculation of the gradient magnitude and gradient direction for each pixel includes: based on the horizontal gradient... The calculation formula is Calculate the horizontal gradient; based on the vertical gradient... The calculation formula is Calculate the gradient in the vertical direction.

[0051] The gradient magnitude G(x,y) and gradient direction θ(x,y) of each pixel are calculated based on Gx and Gy.

[0052] ; .

[0053] S165. In the gradient magnitude of the image at each scale, check the two adjacent pixels in the gradient direction. S167. If the gradient magnitude of the pixel is not greater than the gradient magnitude of the adjacent pixel, then set the gradient value of the pixel to zero to suppress non-edge responses and form clearer edge lines. S168. Perform double threshold segmentation on the gradient image after non-maximum suppression to extract the edge map; Specifically, set a high threshold. and low threshold If the pixel gradient magnitude is greater than If the pixel gradient magnitude is between 0, it is marked as a strong edge point; and If the pixel gradient magnitude is less than a certain value, it is marked as a weak edge point; If a strong edge point is not a strong edge point, it is discarded. An edge image is generated by tracking the weak edge points adjacent to the strong edge points, and the strong edge points and their adjacent weak edge points are considered as edges.

[0054] S169. Overlay the edge images from all scales layer by layer to obtain a comprehensive edge map containing multi-scale information, forming the preliminary edge information of the suspected gallbladder contour region.

[0055] During image overlay, pixels at the same location that are marked as edges in either image at any scale are also marked as edges in the overlay image. The overlay formula is as follows: ; Where E(x,y) represents the pixel value at position (x,y) in the final synthesized edge map, E i (x,y) represents the edge image in the i-th scale image.

[0056] In some existing image processing methods, due to the low resolution and high noise interference of ultrasound images, it is difficult to extract detailed information of the gallbladder region clearly and stably. In traditional image processing methods, the boundary of the target region is usually identified by edge detection at a single scale. However, single-scale processing cannot take into account both coarse and fine edges in the image, which may lead to loss of details or incomplete edges during boundary detection.

[0057] In some embodiments, the step of progressively reducing the resolution of the image of the suspected gallbladder contour region by a predetermined ratio to generate low-resolution image layers of different scales, forming a multi-scale image sequence, includes: The resolution of the original image of the suspected gallbladder outline area is gradually reduced according to a predetermined ratio to generate low-resolution image layers of different scales. For each image layer, a pixel at a location in the current image layer is selected as the processing center point, and a neighborhood range is set to determine a group of neighboring pixels around that pixel; the size of the neighborhood range is controlled by a radius k, that is, it includes the pixel area around the center point; A weight matrix is ​​generated based on the Gaussian distribution function to form a Gaussian kernel; where each element of the Gaussian kernel represents a Gaussian weight value, and the level of the weight value is positively correlated with the distance from the center pixel; Wherein, the Gaussian distribution function is σ is the standard deviation of the Gaussian distribution function, which controls the degree of smoothing.

[0058] Multiply the gray value of each pixel in the neighborhood of the center point by the corresponding Gaussian weight value to obtain a weighted value; sum all the weighted values ​​to obtain the weighted sum of all pixels in the neighborhood of the center point. Divide the accumulated weighted sum by the sum of all weights to obtain the smoothed grayscale value of the center pixel; The calculated smoothed gray value is assigned to the gray value of the center pixel, thus completing the smoothing process for one pixel. Move the processing center to the next pixel position and repeat the above steps to smooth the entire image pixel by pixel until the smoothing of the entire image is completed. The smoothed current layer image is downsampled to proportionally reduce the image size, thus generating a lower-resolution image for the next layer.

[0059] This embodiment proposes a specific edge detection scheme based on multi-scale image processing. By progressively reducing the resolution of the suspected gallbladder contour region layer by layer according to a predetermined ratio to generate multi-scale low-resolution image layers, and then forming a multi-scale image sequence through layer-by-layer smoothing and downsampling, this effectively solves the problem of easily losing detailed information in traditional single-scale edge detection, greatly improving the accuracy and stability of gallbladder region edge detection. Furthermore, since the multi-scale image sequence can simultaneously preserve the coarse and fine edge information of the gallbladder region, it facilitates the provision of a multi-layered and clear image foundation for assessing surgical difficulty during subsequent feature extraction and quantitative analysis.

[0060] In some embodiments, the step of performing a region expansion algorithm based on the preliminary edge information to expand the preliminary edges to cover the complete outline of the gallbladder, marking the expanded complete outline as the region of interest (ROI) of the gallbladder, and generating ROI data includes: performing a region expansion algorithm based on the data carried by the preliminary edge information, gradually expanding from the preliminary edge to the complete outline of the gallbladder region to capture the boundary of the gallbladder; calculating the similarity between the edge and its neighborhood during the expansion process, and removing edge points whose similarity does not meet a preset requirement; marking the expanded complete outline region as the ROI region of the gallbladder, and generating the corresponding ROI data.

[0061] Specifically, before executing the region expansion algorithm, preliminary edge information of the gallbladder is first obtained from the previous step. This preliminary edge information typically includes initially detected gallbladder contour points, which serve as reference starting points for subsequent region expansion. Based on this preliminary edge information, the contour points of the preliminary gallbladder edge are selected as the starting positions for expansion. The region expansion algorithm is invoked, starting from these initial contour points and expanding outwards pixel by pixel to cover the complete contour of the gallbladder region. During the expansion process, adjacent regions of the preliminary edge are checked pixel by pixel, and their grayscale values, texture features, and other information are used to determine whether they belong to the gallbladder region. The determination of each new boundary pixel during the expansion process is based on its similarity calculation with the current edge to ensure the accuracy of the expanded region. Whenever a new edge point is expanded to, the similarity between that edge point and its adjacent edges is calculated, and edge points that do not meet the preset similarity threshold are removed. For example, if the grayscale value difference or directional feature of a new edge point deviates from the existing edge features beyond the preset threshold, the point is considered an anomaly and excluded from the expanded region to ensure the continuity and accuracy of the edges. After expansion, a complete and closed gallbladder contour region is obtained. This complete outline region is marked as the region of interest (ROI) of the gallbladder, and the boundary data and location coordinates of this region are saved to generate corresponding ROI data for use in subsequent radiomics feature extraction and analysis.

[0062] In some embodiments, the step of calculating the stability and discriminative power of each feature based on the normalized feature data, and removing features that do not meet the stability and discriminative power criteria to obtain the filtered target core feature data includes: calculating the inter-group correlation coefficient (ICC) for each normalized feature, and filtering features with a consistency higher than 0.75.

[0063] Specifically, for each normalized feature, the Intra-Class Correlation Coefficient (ICC) is calculated using the following formula to assess the consistency of the feature across different ROI splits or multiple measurements. ; The variance of the features across different segmentation results or measurement groups is used to characterize the variance. This represents the within-group variance of the feature. A higher ICC value indicates greater consistency of the feature across different segmentation results or measurement groups. Studies have shown that features with an ICC value greater than or equal to 0.75 generally exhibit higher stability under varying measurement or segmentation conditions. Features with an ICC value below 0.75 are considered unstable and are discarded.

[0064] Features related to surgical difficulty were screened using statistical methods, and the most representative features were retained using the principles of minimum redundancy and maximum relevance to form the target core feature data. Specifically, features with P ≥ 0.05 and features with ICC values ​​≤ 0.80 were removed using the Independent Samples Test and the Mann-Whitney U Test. The 10 most relevant features were selected using the Maximum Relevance Minimum Redundancy (mRMR) algorithm, and further feature selection was performed using the Minimum Absolute Shrinkage and Selection Operator (LASSO) algorithm. 10-fold cross-validation was used to select the optimal value of the penalty coefficient λ, thereby retaining the most representative features that contribute the most to the prediction.

[0065] In some embodiments, constructing a quantitative analysis model for predicting the difficulty of cholecystectomy based on the screened target core feature data and patient clinical data includes: constructing multiple sub-models based on feature types from the screened patient clinical data; wherein, A demographic sub-model is established based on patient demographic data, including patient age, sex, and body mass index; a laboratory sub-model is established based on preoperative blood test data, including white blood cell count, C-reactive protein, and liver function indicators; a medical history sub-model is established based on patient medical history information, including whether the patient has diabetes, hypertension, and a history of surgery; and a comprehensive clinical model is established based on the prediction results generated by each sub-model; the comprehensive clinical model is configured to assess the patient's overall clinical risk based on a comprehensive analysis of the prediction values ​​of each sub-model. Furthermore, based on the selected target core radiomics features, each radiomics feature is assigned a corresponding weight, and its weighted value is calculated according to the weight of each feature; the weighted values ​​of each feature are accumulated to generate an image score, which is calculated by the following formula: Rad-score=∑(feature value × feature weight)+b0, where b0 is a cutoff value used to adjust the overall offset of the score. For example, suppose there are three features used to calculate the image score: gallbladder volume, surface area, and texture contrast. The specific parameters of these features are as follows: gallbladder volume, corresponding feature value (measured value): 30 cubic centimeters; corresponding feature weight: 0.8, indicating that this feature is relatively important in predicting surgical difficulty; gallbladder surface area, feature value: 50 square centimeters, feature weight: 0.4; texture contrast, feature value: 0.7, feature weight: 0.9; cutoff value: determined to be 2.5 during model training, representing the adjusted baseline value. Substituting these values ​​into the image score formula: Rad-score = (30 × 0.8) + (50 × 0.4) + (0.7 × 0.9) + 2.5, the calculated result is: Rad-score = 47.13. The image score (Rad-score) can reflect the comprehensive influence of multiple image features on the difficulty of surgery, providing clinicians with a valuable quantitative assessment of surgical difficulty.

[0066] The scoring results of the radiomics model and the predicted values ​​of the comprehensive clinical model (i.e., clinical scores) are used as input features and weighted and fused to construct a comprehensive prediction model that combines imaging and clinical features.

[0067] Specifically, the radiomics model is used to provide an objective quantitative assessment of surgical difficulty, reflecting the importance of ultrasound imaging features in surgical prediction. During the construction of the radiomics model, corresponding weights are assigned based on the importance of each radiomics feature. The weight of each feature is obtained through the coefficients or regularization learning of a linear regression model, used to quantify the contribution or importance of each feature in predicting surgical difficulty. The weighted value of each feature is calculated, i.e., each image feature value is multiplied by its corresponding weight to obtain the weighted feature value. The weighted values ​​of all image features are summed to generate an image score (Rad-score). The formula for calculating the image score is: Rad-score = ∑(feature value × feature weight) + b0; where b0 is a cutoff value used to adjust the overall bias of the score. The image score is used to quantitatively assess the impact of image features on the degree of surgical difficulty.

[0068] After obtaining the scoring results from the radiomics model and the predicted values ​​from the integrated clinical model, the two are weighted and fused to generate a comprehensive predictive model combining imaging and clinical features. Specifically, the scoring results from the radiomics model and the predicted values ​​from the integrated clinical model are assigned weights, which reflect the relative importance of imaging and clinical features in predicting surgical difficulty. The imaging score and the clinical predicted value are weighted and averaged to obtain a comprehensive score. The relationship between this comprehensive score and the degree of surgical difficulty is used to train and optimize the comprehensive predictive model combining imaging and clinical features to determine the comprehensive predictive model. This comprehensive predictive model can comprehensively integrate the patient's imaging and clinical information, providing a more accurate assessment of surgical difficulty.

[0069] In some embodiments, during the construction of the comprehensive prediction model, image scores and clinical scores are assigned certain weights. The comprehensive prediction model learns the relationship between the weighted input features and the surgical difficulty label to more comprehensively reflect the combined effect of image features and clinical features, and finally outputs a comprehensive score of surgical difficulty, thereby providing doctors with a standardized and objective quantitative result of surgical difficulty.

[0070] Next, iterative training and optimization of the model are required. Specifically, the sample dataset is divided into training and testing sets, typically in a 7:3 or 8:2 ratio, to reserve some data for testing model performance while training the model. The image feature model, clinical feature model, and comprehensive prediction model are trained separately on the training set to ensure that each model can learn the correlation between image and clinical features and surgical difficulty from the training data. During training, the image feature model and clinical feature model can be trained using basic algorithms such as logistic regression, random forest, and support vector machine, and cross-validation is used to select the optimal parameters. For example, the penalty parameter of the support vector machine model and the number of trees in the random forest model need to be tuned.

[0071] For the comprehensive prediction model, the weighting coefficients of image scores and clinical scores are adjusted during training. By optimizing the weighting coefficients, the model can better balance the contributions of image features and clinical features, thereby maximizing the model's predictive performance on the training data.

[0072] After training, the trained image feature model, clinical feature model, and comprehensive prediction model were validated on a test set to evaluate their generalization ability. Key validation metrics included accuracy, sensitivity, and specificity. The validation results revealed the model's strengths and weaknesses, providing a reference for further optimization. Finally, the model with the best overall performance was selected as the final optimal quantitative analysis model based on the training and testing results.

[0073] In this embodiment, a quantitative predictive analysis model that better reflects the actual surgical difficulty scenario is formed by weighted fusion of images and clinical scores. This model can provide doctors with more consistent multi-dimensional assessment results in predicting the difficulty of surgery, helping them to effectively judge preoperative risks and formulate corresponding surgical plans. This can, to a certain extent, help improve the comprehensiveness and accuracy of surgical difficulty prediction. Example 2

[0074] Based on the same technical concept as the foregoing embodiments, a quantitative analysis system for the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics is also provided, including: The first acquisition unit is used to perform preoperative ultrasound examinations on a large number of patients, acquire ultrasound images showing the gallbladder region and save them in DICOM format. The image enhancement processing unit is used to dynamically adjust the grayscale differences of different regions in the ultrasound image based on a local contrast adjustment algorithm, so as to enhance the contrast between the gallbladder region and the background tissue and eliminate speckle noise in the ultrasound image to obtain a preprocessed image. The first marking unit is used to divide the preprocessed image into multiple regions, calculate the average gray value of each region, and mark the regions whose average gray value exceeds a preset gray value threshold and whose gray values ​​are uniform as candidate regions. The first calculation unit calculates the local gradient values ​​of pixels in the candidate region within the candidate region to form local gradient distribution data. The second marking unit is used to filter and mark the suspected gallbladder contour region that meets the gallbladder contour features based on the local gradient distribution data. The edge information extraction unit is used to extract the preliminary edge information of the suspected gallbladder contour region by applying a multi-scale edge detection algorithm to the suspected gallbladder contour region. The ROI determination unit, based on the preliminary edge information, executes a region expansion algorithm to expand the preliminary edges to cover the complete outline of the gallbladder, marks the expanded complete outline as the region of interest of the gallbladder, and generates ROI data; The radiomics feature extraction unit is used to extract gallbladder ultrasound radiomics features based on the ROI data. The gallbladder ultrasound radiomics features include: shape features, texture features, and grayscale distribution features of the gallbladder region. The normalization processing unit is used to normalize the image omics features, standardizing each feature to the same scale to form normalized feature data; The radiomics feature screening unit is used to calculate the stability and discriminative power of each feature based on the normalized feature data, and to remove features that do not meet the stability and discriminative power criteria in order to obtain the screened target core radiomics features. The model building unit is used to build a quantitative analysis model for predicting the difficulty of cholecystectomy based on the screened target core radiomics features and the patient's clinical data. The quantitative analysis model includes an image feature model, a clinical feature model, and a comprehensive prediction model combining image and clinical features. The model optimization unit is used to divide the sample dataset used for model construction into a training set and a test set; to train and fine-tune the quantitative analysis model on the training set; and to verify the accuracy, sensitivity and specificity of the quantitative analysis model on the test set, and output the best quantitative analysis model after training. The quantitative analysis unit is used to input the current patient's imaging feature data and clinical data into the optimal quantitative analysis model to obtain and output a quantitative score of the difficulty of gallbladder surgery.

[0075] The system in this embodiment can be used to execute Figure 1 The technical solution of the method embodiment shown is similar in principle and technical effect to that of Embodiment 1, and will not be described in detail here. They can be referred to each other.

[0076] It is understood that the system of this embodiment can also be used to execute other embodiments in Embodiment 1. For the sake of clarity and simplicity, the other embodiments will not be described in detail, but can be referred to each other.

[0077] Another embodiment of the present invention provides an electronic device, including: one or more processors; a memory; wherein the memory stores one or more executable programs, and the one or more processors read the executable program code stored in the memory and run a quantification analysis program for the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics corresponding to the executable program code, so as to execute the quantification analysis method for the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics as described in any of the embodiments of the present invention.

[0078] Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention, which can implement any of the methods described in Embodiment 1 of the present invention, such as... Figure 4 As shown, as an optional embodiment, the above-mentioned electronic device may include: a housing 41, a processor 42, a memory 43, a circuit board 44, and a power supply circuit 45, wherein the circuit board 44 is disposed inside the space enclosed by the housing 41, and the processor 42 and the memory 43 are disposed on the circuit board 44; the power supply circuit 45 is used to supply power to the various circuits or devices of the above-mentioned electronic device; the memory 43 is used to store executable program code; the processor 42 runs the program corresponding to the executable program code by reading the executable program code stored in the memory 43, for executing the aforementioned quantitative analysis method for the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics as described in any of the embodiments in Example 1.

[0079] For details on the specific execution process of the above steps by the processor 42 and the steps further executed by the processor 42 through running executable program code, please refer to the description of Embodiment 1 of the network space data fusion analysis method of the present invention, which will not be repeated here.

[0080] It should be understood that the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0081] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for quantitatively analyzing the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics, characterized in that, include: Preoperative ultrasound examinations were performed on a large number of patients to obtain ultrasound images showing the gallbladder area and save them in DICOM format. Based on the local contrast adjustment algorithm, the gray level difference of different regions in the ultrasound image is dynamically adjusted to enhance the contrast between the gallbladder region and the background tissue, and to eliminate speckle noise in the ultrasound image to obtain a preprocessed image. The preprocessed image is divided into multiple regions, the average gray value of each region is calculated, and regions whose average gray value exceeds a preset gray value threshold and whose gray values ​​are uniform are marked as candidate regions. Within the candidate region, the local gradient values ​​of the pixels in the candidate region are calculated to form local gradient distribution data; Based on the local gradient distribution data, suspected gallbladder contour regions that meet the gallbladder contour characteristics are screened and marked. For the suspected gallbladder contour region, a multi-scale edge detection algorithm is applied to extract the preliminary edge information of the suspected gallbladder contour region; Based on the preliminary edge information, a region expansion algorithm is executed to expand the preliminary edges to cover the complete outline of the gallbladder. The expanded complete outline is marked as the region of interest of the gallbladder, and ROI data is generated. Based on the ROI data, gallbladder ultrasound radiomics features are extracted, including: shape features, texture features, and grayscale distribution features of the gallbladder region; The radiomics features are normalized to the same scale to form normalized feature data. Based on the normalized feature data, the stability and discriminative power of each feature are calculated, and features that do not meet the stability and discriminative power criteria are removed to obtain the selected target core image omics features. Based on the selected target core radiomics features and the patient's clinical data, a quantitative analysis model for predicting the difficulty of cholecystectomy is constructed. The quantitative analysis model includes an imaging feature model, a clinical feature model, and a comprehensive prediction model combining imaging and clinical features. The sample dataset used for model building is divided into a training set and a test set; the quantitative analysis model is trained and its parameters are tuned on the training set, and the accuracy, sensitivity and specificity of the quantitative analysis model are verified on the test set, and the best quantitative analysis model after training is output. The patient's current imaging features and clinical data are input into the optimal quantitative analysis model to obtain and output a quantitative score of surgical difficulty.

2. The method for quantitative analysis of the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics according to claim 1, characterized in that, The local contrast adjustment algorithm dynamically adjusts the grayscale differences in different regions of the ultrasound image to enhance the contrast between the gallbladder region and the background tissue, and eliminates speckle noise in the ultrasound image, resulting in a preprocessed image including: The ultrasound image is divided into multiple overlapping local regions. For each local region, the gray-level histogram of that region is calculated and gray-level equalization is performed to enhance the gray-level contrast within the local region. For each local region, calculate the local brightness parameter of that local region to characterize the overall gray-level center value of that region; and calculate the gray-level dispersion of each local region to quantify the local contrast level, in order to guide the intensity adjustment of contrast enhancement. Based on the local brightness parameters and grayscale dispersion of each local region, the grayscale values ​​of each pixel in the local region are adjusted to enhance the contrast between local regions, so as to reduce the grayscale difference between the gallbladder region and the background tissue. During the local contrast enhancement process, each local region of the ultrasound image is processed pixel by pixel based on the filtering algorithm to detect and suppress noise signals. The enhancement results of each local region are merged into a complete image, and the pixel values ​​in each overlapping region are weighted and averaged to make the image transition between different overlapping regions smooth, thus obtaining a preprocessed image.

3. The method for quantitative analysis of the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics according to claim 1, characterized in that, The step of dividing the preprocessed image into multiple region blocks, calculating the average gray value of each region block, and marking the region blocks whose average gray value exceeds a preset gray value threshold and whose gray value is uniform as candidate regions includes: dividing the preprocessed image into rectangular region blocks of equal size and calculating the average gray value of each region block; Z-score normalization is applied to the average gray value of each region block to ensure that the gray values ​​of each region block are on a uniform scale. Regions with standardized gray values ​​exceeding a preset gray threshold and uniform gray levels are marked as candidate regions.

4. The method for quantitative analysis of the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics according to claim 1, characterized in that, The suspected gallbladder contour region is processed using a multi-scale edge detection algorithm to extract preliminary edge information, including: The resolution of images of suspected gallbladder contour areas is gradually reduced according to a predetermined ratio to generate low-resolution image layers of different scales, forming a multi-scale image sequence. For each scale image in the multi-scale image sequence, the gradient magnitude and gradient direction of each pixel are calculated to identify significant edge features; In the gradient magnitude of the image at each scale, examine the two adjacent pixels along the gradient direction. If the gradient magnitude of a pixel is not greater than the gradient magnitude of its neighboring pixels, then the gradient value of that pixel is set to zero to suppress non-edge responses and create clearer edge lines. Double threshold segmentation is performed on the gradient image after non-maximum suppression to extract the edge map; By layering the edge maps from all scales of the image, a comprehensive edge map containing multi-scale information is obtained, forming the preliminary edge information of the suspected gallbladder contour region.

5. The method for quantitative analysis of the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics according to claim 4, characterized in that, The process of progressively reducing the resolution of images of suspected gallbladder contour regions according to a predetermined ratio to generate low-resolution image layers of different scales, forming a multi-scale image sequence, includes: The resolution of the original image of the suspected gallbladder outline area is gradually reduced according to a predetermined ratio to generate low-resolution image layers of different scales. For each image layer, a pixel at a location in the current image layer is selected as the processing center point, and a neighborhood range is set to determine a group of neighboring pixels around that pixel; the size of the neighborhood range is controlled by a radius k, that is, it includes the pixel area around the center point; A weight matrix is ​​generated based on the Gaussian distribution function to form a Gaussian kernel; wherein each element of the Gaussian kernel represents a Gaussian weight value, and the level of the weight value is positively correlated with the distance from the center pixel; The weighted value is obtained by multiplying the gray value of each pixel in the neighborhood of the center point with the corresponding Gaussian weight value. The weighted values ​​are summed to obtain the weighted sum of all pixels in the neighborhood of the center point; Divide the accumulated weighted sum by the sum of all weights to obtain the smoothed grayscale value of the center pixel; The calculated smoothed gray value is assigned to the gray value of the center pixel, thus completing the smoothing process for one pixel. Move the processing center to the next pixel position and repeat the above steps to smooth the entire image pixel by pixel until the smoothing of the entire image is completed. The smoothed current layer image is downsampled to proportionally reduce the image size, thus generating a lower-resolution image for the next layer.

6. The method for quantitative analysis of the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics according to claim 1, characterized in that, Based on the preliminary edge information, a region expansion algorithm is executed to expand the preliminary edges to cover the complete outline of the gallbladder. The expanded complete outline is marked as the region of interest (ROI) of the gallbladder, and ROI data is generated, including: Based on the data carried by the preliminary edge information, a region expansion algorithm is executed to gradually expand from the preliminary edge to the complete outline of the gallbladder region in order to capture the boundary of the gallbladder. During the expansion process, the similarity between the edge and its neighborhood is calculated, and edge points whose similarity does not meet the preset requirements are removed. The expanded, complete contour region is marked as the ROI region of the gallbladder, and the corresponding ROI data is generated.

7. The method for quantitative analysis of the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics according to claim 1, characterized in that, The step of calculating the stability and discriminative power of each feature based on the normalized feature data, and removing features that do not meet the stability and discriminative power criteria to obtain the filtered target core feature data includes: calculating the inter-group correlation coefficient for each normalized feature and filtering features with a consistency higher than 0.

75. Features related to surgical difficulty are selected using statistical methods, and the most representative features are retained using the principles of minimum redundancy and maximum relevance to form the target core feature data.

8. The method for quantitative analysis of the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics according to claim 1, characterized in that, Based on the screened target core feature data and the patient's clinical data, a quantitative analysis model for predicting the difficulty of cholecystectomy is constructed, including: Multiple sub-models were constructed from the selected patient clinical data according to feature types; among them... A demographic sub-model is established based on patient demographic data, including patient age, gender, and body mass index. A laboratory test sub-model was established based on the patient's preoperative blood test data, which included white blood cell count, C-reactive protein, and liver function indicators. A medical history sub-model is established based on the patient's medical history information, which includes whether the patient has diabetes, hypertension, and a history of surgery. Based on the prediction results generated by each of the sub-models, a comprehensive clinical model is established; the comprehensive clinical model is configured to assess the patient's overall clinical risk based on a comprehensive analysis of the prediction values ​​of each sub-model. Furthermore, based on the selected target core radiomics features, a corresponding weight is assigned to each radiomics feature, and its weighted value is calculated according to the weight of each feature. The weighted values ​​of each feature are accumulated to generate an image score, which is calculated using the following formula: Rad-score=∑(feature value×feature weight)+b0, where b0 is a cutoff value used to adjust the overall offset of the score. The image score is used to quantitatively evaluate the impact of image features on the difficulty of surgery. The scoring results of the radiomics model and the predicted values ​​of the integrated clinical model are used as input features for weighted fusion to construct an integrated prediction model that combines imaging and clinical features.

9. A quantitative analysis system for the difficulty of laparoscopic cholecystectomy based on ultrasound radiomics, characterized in that, include: The first acquisition unit is used to perform preoperative ultrasound examinations on a large number of patients, acquire ultrasound images showing the gallbladder region and save them in DICOM format. The image enhancement processing unit is used to dynamically adjust the grayscale differences of different regions in the ultrasound image based on a local contrast adjustment algorithm, so as to enhance the contrast between the gallbladder region and the background tissue and eliminate speckle noise in the ultrasound image to obtain a preprocessed image. The first marking unit is used to divide the preprocessed image into multiple regions, calculate the average gray value of each region, and mark the regions whose average gray value exceeds a preset gray value threshold and whose gray values ​​are uniform as candidate regions. The first calculation unit calculates the local gradient values ​​of pixels in the candidate region within the candidate region to form local gradient distribution data. The second marking unit is used to filter and mark the suspected gallbladder contour region that meets the gallbladder contour features based on the local gradient distribution data. The edge information extraction unit is used to extract the preliminary edge information of the suspected gallbladder contour region by applying a multi-scale edge detection algorithm to the suspected gallbladder contour region. The ROI determination unit, based on the preliminary edge information, executes a region expansion algorithm to expand the preliminary edges to cover the complete outline of the gallbladder, marks the expanded complete outline as the region of interest of the gallbladder, and generates ROI data; The radiomics feature extraction unit is used to extract gallbladder ultrasound radiomics features based on the ROI data. The gallbladder ultrasound radiomics features include: shape features, texture features, and grayscale distribution features of the gallbladder region. The normalization processing unit is used to normalize the image omics features, standardizing each feature to the same scale to form normalized feature data; The radiomics feature screening unit is used to calculate the stability and discriminative power of each feature based on the normalized feature data, and to remove features that do not meet the stability and discriminative power criteria in order to obtain the screened target core radiomics features. The model building unit is used to build a quantitative analysis model for predicting the difficulty of cholecystectomy based on the screened target core radiomics features and the patient's clinical data. The quantitative analysis model includes an image feature model, a clinical feature model, and a comprehensive prediction model combining image and clinical features. The model optimization unit is used to divide the sample dataset used for model construction into a training set and a test set; to train and fine-tune the quantitative analysis model on the training set; and to verify the accuracy, sensitivity and specificity of the quantitative analysis model on the test set, and output the best quantitative analysis model after training. The quantitative analysis unit is used to input the current patient's imaging feature data and clinical data into the optimal quantitative analysis model to obtain and output a quantitative score of the difficulty of gallbladder surgery.

10. An electronic device, characterized in that, include: One or more processors; Memory; The memory stores one or more executable programs. The one or more processors read the executable program code stored in the memory and run the laparoscopic cholecystectomy difficulty quantification analysis program based on ultrasound radiomics corresponding to the executable program code, so as to execute the laparoscopic cholecystectomy difficulty quantification analysis method based on ultrasound radiomics as described in any one of claims 1 to 8.

Images