Automatic evaluation method for laparoscopic cholecystectomy operation skills and related device

Abstract

The invention discloses an automatic evaluation method and related device for laparoscopic cholecystectomy operation skills, and the method comprises the steps: obtaining video data in a laparoscopic cholecystectomy operation scene, and carrying out the framing processing of the video data, and forming framed image data; performing image frame extraction processing on the framed image data based on a time sequence sampling strategy to obtain frame extraction image data; inputting the frame extraction image data into a convergent bleeding feature extraction model to perform bleeding feature extraction processing in the image, and obtaining bleeding feature data corresponding to each frame of image data; and performing score estimation processing on the bleeding feature data corresponding to each frame of image data by using an mGOALS score prediction algorithm based on random forest regression to obtain a score estimation result. In the embodiment of the invention, by focusing the key stage of the operation, the interference of non-key operation is avoided, the evaluation specificity is improved, the automatic evaluation is realized, and the manual intervention demand is obviously reduced.

Description

Technical Field

[0001] The present invention relates to the field of automatic evaluation technology, and in particular to an automatic evaluation method and related device for laparoscopic cholecystectomy operating skills. Background Art

[0002] Laparoscopic cholecystectomy is a basic procedure for hepatobiliary surgeons to build a minimally invasive surgical capability system. The degree of technical mastery directly affects the slope of the learning curve of complex operations. However, although it is considered an entry-level operation, its complexity and potential risks cannot be ignored. Therefore, how to effectively evaluate skills during laparoscopic surgery, shorten the learning curve for beginners, help doctors improve their own skills, and enhance surgical safety is a key and major clinical need to standardize surgical techniques and quality, improve surgical rescue rates, and improve regional inequality in the allocation of medical resources. With the development of artificial intelligence technology, some studies have begun to explore the use of deep learning technology for laparoscopic surgery. The current laparoscopic cholecystectomy (LC) skill assessment relies on subjective scales (such as GOALS), which have problems such as long time consumption (average 38.2 minutes / case) and large inter-rater variability (Δ>22.7%). The bleeding quantification index developed for gastrointestinal surgery cannot be directly applied due to the special spatiotemporal distribution of bleeding in LC surgery (89.6% of bleeding is concentrated in the critical safety field stage). This technical solution aims to build a stage-specific bleeding recognition model to achieve automated and objective surgical skill assessment by quantifying the number of bleeding pixels in each frame during the gallbladder triangle processing stage. Summary of the Invention

[0003] The purpose of the present invention is to overcome the shortcomings of the existing technology. The present invention provides an automatic assessment method and related devices for laparoscopic cholecystectomy operation skills. By focusing on the key stages of the operation, the interference of non-critical operations is avoided, the assessment specificity is improved, and automated assessment is realized, which significantly reduces the need for manual intervention.

[0004] In order to solve the above technical problems, an embodiment of the present invention provides an automatic assessment method for laparoscopic cholecystectomy operating skills, the method comprising:

[0005] Acquiring video data of a laparoscopic cholecystectomy operation scene, and performing frame processing on the video data to form framed image data;

[0006] Performing image extraction processing on the framed image data based on a time sequence sampling strategy to obtain extracted frame image data;

[0007] Inputting the extracted frame image data into a converged bleeding feature extraction model to perform bleeding feature extraction processing in the image, and obtaining bleeding feature data corresponding to each frame of image data, wherein the bleeding feature extraction model adopts a PoolNet network architecture;

[0008] The mGOALS score prediction algorithm based on random forest regression was used to perform score estimation processing on the bleeding feature data corresponding to each frame of image data to obtain the score estimation result.

[0009] Optionally, performing frame extraction processing on the framed image data to obtain frame extracted image data includes:

[0010] Based on the time series sampling strategy, the anatomical process of the gallbladder triangle in the key safety field of view is focused on to perform image frame processing and obtain frame image data.

[0011] Optionally, the process of training the bleeding feature extraction model to a converged bleeding feature extraction model includes:

[0012] Obtain a multi-source heterogeneous laparoscopic cholecystectomy video dataset, wherein the multi-source heterogeneous laparoscopic cholecystectomy video dataset includes N columns of laparoscopic cholecystectomy video datasets in an international standard library and M columns of real clinical laparoscopic cholecystectomy video datasets;

[0013] For each column of laparoscopic cholecystectomy video data in the multi-source heterogeneous laparoscopic cholecystectomy video dataset, a time-series sampling strategy is used to perform frame extraction processing on the gallbladder triangle dissection process during the focusing on the key safety field of view to form a sample frame-extracted image dataset;

[0014] Annotating each sample frame image data in the sample frame image data set based on a double-blind manual annotation method to obtain bleeding pixel point data corresponding to each sample frame image data, wherein the annotation includes three types of morphological features: acute ejective bleeding, tissue oozing, and coagulation deposition;

[0015] Based on the HSV color space enhancement algorithm, a comparative learning framework is constructed for the bleeding pixel data corresponding to each sample frame image data to obtain a comparative learning framework corresponding to each sample frame image data;

[0016] The bleeding pixel data corresponding to each sample frame image data and the contrast learning framework corresponding to each sample frame image data are mixed and randomly divided into training data set, verification data set and test data set in a ratio of 8:1:1;

[0017] The training data set, the validation data set, and the test data set are sequentially input into a bleeding feature extraction model for training, validation, and testing until a converged bleeding feature extraction model is formed.

[0018] Optionally, the PoolNet network architecture of the bleeding feature extraction model integrates a VGG-16 network and a ResNet-50 network as the backbone network, and a transfer learning strategy is used to initialize the training weight values ​​during the training process.

[0019] Optionally, during the testing process, the bleeding feature extraction model evaluates the test results based on multi-dimensional quantitative indicators, wherein the multi-dimensional quantitative indicators include the Dice similarity coefficient to evaluate segmentation accuracy, the Precision rate to measure the positive predictive value, and the Recall rate to characterize the sensitivity, as follows:

[0020]

[0021] Among them, TP represents the number of correctly identified blood pixels, that is, true positives; FP represents the number of non-blood pixels incorrectly identified as blood pixels, that is, false positives; FN represents the number of unidentified bleeding pixels, that is, false negatives.

[0022] Optionally, the mGOALS score prediction algorithm based on random forest regression is used to perform score estimation processing on the bleeding feature data corresponding to each frame of image data to obtain a score estimation result, including:

[0023] Performing multidimensional bleeding volume feature vector construction processing using bleeding feature data corresponding to each frame of image data to obtain multidimensional bleeding volume feature data corresponding to the frame of image data, wherein the multidimensional bleeding volume feature data is composed of total bleeding volume feature data, peak bleeding volume feature data, and average bleeding volume feature data;

[0024] The multidimensional bleeding volume feature data is scored and estimated using the mGOALS score prediction algorithm based on random forest regression to obtain a score estimation result; wherein the mGOALS score prediction algorithm based on random forest regression realizes automatic association of skill scores through feature extraction, model training, verification and optimization.

[0025] Optionally, the step of constructing a multidimensional bleeding volume feature vector using the bleeding feature data corresponding to each frame of image data to obtain the multidimensional bleeding feature data corresponding to the frame of image data includes:

[0026] The bleeding volume characteristic data is calculated and processed by taking the bleeding pixel count as the core feature of the bleeding characteristic data corresponding to each frame of image data to obtain the total bleeding volume characteristic data, the peak bleeding volume characteristic data and the average bleeding volume characteristic data corresponding to the frame of image data;

[0027] A multidimensional bleeding volume feature vector is constructed based on the total bleeding volume feature data, peak bleeding volume feature data and average bleeding volume feature data corresponding to the frame-extracted image data to obtain the multidimensional bleeding volume feature data corresponding to the frame-extracted image data.

[0028] In addition, an embodiment of the present invention further provides an automatic evaluation device for laparoscopic cholecystectomy operating skills, the device comprising:

[0029] Framing module: used to obtain video data in the laparoscopic cholecystectomy operation scene, and perform frame processing on the video data to form framed image data;

[0030] Frame extraction module: used for performing frame extraction processing on the framed image data based on a time sequence sampling strategy to obtain frame extracted image data;

[0031] Feature extraction module: used to input the sampled frame image data into the converged bleeding feature extraction model to extract the bleeding features in the image and obtain the bleeding feature data corresponding to each frame of image data. The bleeding feature extraction model adopts the PoolNet network architecture;

[0032] Score estimation module: used to use the mGOALS score prediction algorithm based on random forest regression to perform score estimation processing on the bleeding feature data corresponding to each frame of image data to obtain the score estimation result.

[0033] In addition, an embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above methods when executing the computer program.

[0034] In addition, an embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the above methods are implemented.

[0035] In an embodiment of the present invention, the fully automatic calculation of the bleeding pixel count in the gallbladder triangle is realized, and its evaluation efficiency is verified to be significantly negatively correlated with the expert score; the problems of strong subjectivity and long time consumption of traditional scale evaluation are solved; by focusing on the key surgical stages, the interference of non-critical operations is avoided and the evaluation specificity is improved; "bleeding pixels per frame" rather than the total bleeding volume is used as an indicator, which enhances the stability and comparability of the results; and end-to-end automation of the entire process from video input to skill scoring is achieved, which significantly reduces the need for manual intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0037] Figure 1 1 is a flow chart of an automatic evaluation method for laparoscopic cholecystectomy operating skills in an embodiment of the present invention;

[0038] Figure 2 1 is a schematic diagram of the structure of an automatic evaluation device for laparoscopic cholecystectomy operating skills in an embodiment of the present invention;

[0039] Figure 3 It is a schematic diagram of the structure of an electronic device in an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0041] For example 1, please refer to Figure 1 , Figure 1 4 is a flow chart of an automatic evaluation method for laparoscopic cholecystectomy operating skills in an embodiment of the present invention.

[0042] like Figure 1 As shown, a method for automatically evaluating laparoscopic cholecystectomy operating skills comprises:

[0043] S11: acquiring video data of a laparoscopic cholecystectomy operation scene, and performing frame processing on the video data to form frame image data;

[0044] In the specific implementation process of the present invention, when performing a laparoscopic cholecystectomy operation, video capture is performed in the surgical operation scene through a preset camera device, thereby obtaining video data in the laparoscopic cholecystectomy operation scene, and then the video data is framed according to the acquisition exposure frequency of the camera device to form framed image data.

[0045] S12: performing frame sampling processing on the framed image data based on a time sequence sampling strategy to obtain frame-sampled image data;

[0046] In the specific implementation process of the present invention, the image frame processing is performed on the framed image data based on the time series sampling strategy to obtain the framed image data, including: focusing on the anatomical process of the gallbladder triangle in the key safety field of view stage based on the time series sampling strategy to perform image frame processing to obtain the framed image data.

[0047] Specifically, in order to reduce the subsequent calculation amount and improve the prediction accuracy, the framed image data needs to be processed accordingly. Therefore, it is necessary to perform image extraction processing on the framed image data according to the time sampling strategy of 25 frames / second and focus on the gallbladder triangle anatomy process of the gallbladder triangle anatomy stage of the critical safety field (CVS) to obtain the extracted image data; by constructing a CVS bleeding recognition model, it can be used to identify bleeding conditions during the core operation of LC (laparoscopic cholecystectomy), that is, image frame data with bleeding conditions can be extracted from the framed image data, so that each frame of image data in the extracted image data is image data with bleeding conditions.

[0048] Since the standardized establishment of the critical view of safety (CVS) is a core evaluation indicator in the laparoscopic cholecystectomy safety assessment system, and given that the risk of intraoperative bleeding is significantly negatively correlated with the surgeon's skill level, a structured video analysis method was used to focus on the gallbladder triangle dissection stage of the critical view of safety (CVS), which is the surgical process from serosal incision to complete exposure of the triangle.

[0049] S13: Inputting the extracted frame image data into a converged bleeding feature extraction model to perform bleeding feature extraction processing in the image to obtain bleeding feature data corresponding to each frame of image data, wherein the bleeding feature extraction model adopts a PoolNet network architecture;

[0050] In the specific implementation process of the present invention, the process of training the bleeding feature extraction model to a convergent bleeding feature extraction model includes: obtaining a multi-source heterogeneous laparoscopic cholecystectomy video dataset, wherein the multi-source heterogeneous laparoscopic cholecystectomy video dataset includes N columns of laparoscopic cholecystectomy video datasets in an international standard library and M columns of real clinical laparoscopic cholecystectomy video datasets; performing frame extraction processing on the gallbladder triangle anatomy process of each column of laparoscopic cholecystectomy video data in the multi-source heterogeneous laparoscopic cholecystectomy video dataset in the focus on the key safety field stage using a time series sampling strategy to form a sample frame image dataset; labeling each sample frame image data in the sample frame image dataset based on a double-blind manual labeling method to obtain bleeding pixel point data corresponding to each sample frame image data, wherein the labeling includes three types of morphological features: acute spurting bleeding, tissue oozing and coagulation deposition; constructing a comparative learning framework for the bleeding pixel point data corresponding to each sample frame image data based on the HSV color space enhancement algorithm to obtain a comparative learning framework corresponding to each sample frame image data;

[0051] The bleeding pixel data corresponding to each sample frame image data and the comparative learning framework corresponding to each sample frame image data are mixed and randomly divided into a training data set, a verification data set, and a test data set in a ratio of 8:1:1; the training data set, the verification data set, and the test data set are sequentially input into the bleeding feature extraction model for training, verification, and testing until a converged bleeding feature extraction model is formed.

[0052] Furthermore, the PoolNet network architecture of the bleeding feature extraction model integrates the VGG-16 network and the ResNet-50 network as the backbone network, and uses a transfer learning strategy to initialize the training weight values ​​during the training process.

[0053] Furthermore, during the testing process, the bleeding feature extraction model evaluates the test results based on multi-dimensional quantitative indicators, including the Dice similarity coefficient to evaluate segmentation accuracy, the Precision rate to measure the positive predictive value, and the Recall rate to characterize the sensitivity, as follows:

[0054]

[0055] Among them, TP represents the number of correctly identified blood pixels, that is, true positives; FP represents the number of non-blood pixels incorrectly identified as blood pixels, that is, false positives; FN represents the number of unidentified bleeding pixels, that is, false negatives.

[0056] Specifically, the first step is to construct a bleeding feature extraction model, which adopts the PoolNet network architecture and integrates the VGG-16 network and ResNet-50 network as the backbone network. During the training process, the transfer learning strategy is used to initialize the training weight values; it is used to segment bleeding pixels in the gallbladder triangle (average DSC = 0.803 ± 0.015, Precision = 0.860 ± 0.022, Recall = 0.759 ± 0.018).

[0057] When training the bleeding feature extraction model, it is necessary to first build the corresponding data set. Therefore, it is necessary to combine the international standard library (cholec80) with real clinical data (30 cases from Zhujiang Hospital) to ensure that the data can cover a variety of surgical scenarios.

[0058] For these video data, a 25-frame-per-second temporal sampling strategy was adopted, focusing on the gallbladder triangle anatomy during the critical visual field of safety (CVS) phase. A dataset of bleeding pixels was constructed through double-blind manual annotation, with annotation categories including three morphological features: acute spurting bleeding, tissue hemorrhage, and coagulation deposition. To address the issue of color misclassification between liver parenchyma and bleeding areas, an HSV color space enhancement algorithm was introduced to construct a contrastive learning framework. A collaborative training strategy of positive and negative samples was used to suppress artifact interference, significantly improving the model's sensitivity to bleeding areas. The dataset was randomly divided into training, validation, and test sets in an 8:1:1 ratio.

[0059] Since the standardized establishment of the Critical View of Safety (CVS) is a core evaluation indicator in the laparoscopic cholecystectomy safety assessment system, and given that the risk of intraoperative bleeding is significantly negatively correlated with the surgeon's skill level, a structured video analysis method was used to focus on the gallbladder triangle dissection stage - the surgical process from serosal incision to complete exposure of the triangle. Video processing followed the following technical procedures: First, Wondershare Filmora software was used to export the video to a standard format (resolution 1920×1080, frame rate 25fps, bit rate 5999kbps). Subsequently, video frame extraction was performed in the PyCharm Community Edition development environment, extracting key frames every 25 frames (i.e., per second). Finally, the bleeding sites were manually labeled using a double-blind method using the LabelMe annotation system.

[0060] In image processing, based on the principles of digital image processing, the intraoperative bleeding area can be quantitatively identified through video annotation and pixel segmentation technology; the morphological standards of bleeding pixels are strictly defined, and according to the computer vision classification criteria, they mainly include the following three morphological features: (1) acute jet bleeding (sudden pulsating bleeding); (2) tissue oozing (low-flow exudative bleeding); (3) coagulation deposition (localized blood accumulation); the above three types of features are all included in the multimodal annotation system. To address the risk of misjudgment caused by the similar color of the liver parenchyma and the bleeding area in the laparoscopic surgical field, this embodiment will establish a double verification mechanism: by systematically collecting non-bleeding frame images as a counterexample training set, using the HSV color space enhancement algorithm to construct a contrast learning framework, effectively improving the sensitivity and specificity of bleeding feature recognition; this embodiment uses a positive and negative sample collaborative training strategy to significantly reduce the interference of tissue artifacts on the semantic segmentation model.

[0061] The bleeding feature extraction model adopts a dual-stream feature extraction mechanism: VGG-16 and ResNet-50 are integrated as the backbone network, the pre-trained weights (based on the ImageNet dataset) are initialized through a transfer learning strategy, and the newly constructed layers are initialized using Xavier.

[0062] During the training process, the input training data set is obtained through structured video sampling, and a static image data set is generated following the time-series sampling rule of 25 frames per second. The data annotation and processing process strictly follows the evidence-based medicine standards: according to the predefined hemorrhage morphology classification system, the annotation coordinates are mapped to the HSV color space for standardized conversion, and the pixel density of the hemorrhage area is automatically calculated by the convolutional neural network. The data set is randomly divided into training set, validation set and test set in a stratified ratio of 8:1:1 to ensure the statistical validity of the model generalization ability evaluation; the model performance evaluation uses multi-dimensional quantitative indicators: including the Dice similarity coefficient (DSC) to evaluate segmentation accuracy, the precision rate (Precision) to measure the positive predictive value, and the recall rate (Recall) to characterize the sensitivity. A special adaptive feature fusion mechanism is constructed to optimize the multi-scale hemorrhage feature extraction through spatial pyramid pooling. The mathematical definitions of these three indicators are as follows:

[0063]

[0064] Among them, TP represents the number of correctly identified blood pixels, that is, true positives; FP represents the number of non-blood pixels incorrectly identified as blood pixels, that is, false positives; FN represents the number of unidentified bleeding pixels, that is, false negatives.

[0065] S14: Using the mGOALS score prediction algorithm based on random forest regression, the bleeding feature data corresponding to each frame of image data is scored and estimated to obtain a score estimation result.

[0066] In the specific implementation process of the present invention, the mGOALS score prediction algorithm based on random forest regression is used to perform score estimation processing on the bleeding feature data corresponding to each frame of image data to obtain a score estimation result, including: using the bleeding feature data corresponding to each frame of image data to construct a multidimensional bleeding volume feature vector to obtain multidimensional bleeding volume feature data corresponding to the frame image data, wherein the multidimensional bleeding volume feature data is composed of total bleeding volume feature data, peak bleeding volume feature data and average bleeding volume feature data; using the mGOALS score prediction algorithm based on random forest regression to perform score estimation processing on the multidimensional bleeding volume feature data to obtain a score estimation result; wherein the mGOALS score prediction algorithm based on random forest regression realizes automatic association of skill scores through feature extraction, model training, verification and optimization.

[0067] Furthermore, the method of using the bleeding characteristic data corresponding to each frame of image data to construct a multidimensional bleeding volume feature vector to obtain the multidimensional bleeding feature data corresponding to the frame-by-frame image data includes: calculating the bleeding volume feature data using the bleeding pixel count as the core feature of the bleeding characteristic data corresponding to each frame of image data to obtain the total bleeding volume feature data, peak bleeding volume feature data and average bleeding volume feature data corresponding to the frame-by-frame image data; and constructing a multidimensional bleeding volume feature vector based on the total bleeding volume feature data, peak bleeding volume feature data and average bleeding volume feature data corresponding to the frame-by-frame image data to obtain the multidimensional bleeding volume feature data corresponding to the frame-by-frame image data.

[0068] Specifically, an mGOALS score prediction algorithm based on random forest regression was established, which achieved automatic association of skill scores through feature extraction, model training, verification and optimization.

[0069] Specifically, the average bleeding pixel count per frame was used as the core feature, combined with auxiliary indicators such as total bleeding volume and peak bleeding volume to construct a multidimensional bleeding behavior feature vector. A video analysis framework was constructed using 70 internal datasets, and each video was independently encoded and input into the deep learning system. The model executed a fully automated processing flow: video preprocessing - bleeding area segmentation - quantitative indicator output, ultimately generating three sets of key parameters: (1) total bleeding volume (total number of pixels); (2) peak bleeding volume (maximum number of pixels per frame); and (3) average number of bleeding pixels per frame (number of pixels / frame).

[0070] A random forest regression model was used to capture the nonlinear relationship between bleeding characteristics and mGOALS scores through decision tree ensemble learning. Total video bleeding volume was normalized (total bleeding pixel count / total number of frames) to establish a Spearman rank correlation model with the total mGOALS score and each dimension. The Mann-Whitney U test was used to compare the statistical differences in bleeding indicators between the high- and low-skill groups. End-to-end automation was achieved throughout the entire process, from video input to skill scoring, significantly reducing the need for manual intervention. Validation and optimization: The model was validated in an internal test set (70 cases) and an external test set (30 cases). The Kruskal-Wallis test was used to compare the two data sets, and the Fleiss kappa test was used to evaluate the consistency of the scores of three expert surgeons. The results showed that the correlation coefficient (ρ) between the predicted score and the expert score was -0.681 (P < 0.001), and the prediction error rate in the high-skill group (mGOALS ≥ 12 points) was reduced by 42.3% compared with the traditional method.

[0071] For the bleeding feature data corresponding to each frame of image data, the bleeding volume feature data will be calculated with the bleeding pixel count as the core feature, thereby obtaining the total bleeding volume feature data, peak bleeding volume feature data, and average bleeding volume feature data corresponding to the frame-by-frame image data; then, a multidimensional bleeding volume feature vector will be constructed using the total bleeding volume feature data, peak bleeding volume feature data, and average bleeding volume feature data corresponding to the frame-by-frame image data, and the multidimensional bleeding volume feature data corresponding to the frame-by-frame image data will be obtained; finally, the multidimensional bleeding volume feature data will be scored and estimated using the mGOALS score prediction algorithm based on random forest regression to obtain the score estimation result; the mGOALS score prediction algorithm based on random forest regression realizes the automatic association of skill scores through feature extraction, model training, verification, and optimization.

[0072] In an embodiment of the present invention, the fully automatic calculation of the bleeding pixel count in the gallbladder triangle is realized, and its evaluation efficiency is verified to be significantly negatively correlated with the expert score; the problems of strong subjectivity and long time consumption of traditional scale evaluation are solved; by focusing on the key surgical stages, the interference of non-critical operations is avoided and the evaluation specificity is improved; "bleeding pixels per frame" rather than the total bleeding volume is used as an indicator, which enhances the stability and comparability of the results; and end-to-end automation of the entire process from video input to skill scoring is achieved, which significantly reduces the need for manual intervention.

[0073] Aiming at the complex scenario of high-density, low-flow-rate bleeding in the gallbladder triangle during laparoscopic cholecystectomy (LC), an innovative dual-stream feature extraction network (VGG-16 semantic features + ResNet-50 spatial features) was constructed, and the bleeding area segmentation was optimized through transfer learning (ImageNet pre-trained weights) and adaptive feature fusion (spatial pyramid pooling). In view of the fact that 89.6% of bleeding in LC surgery is concentrated in the critical safety field (CVS) stage, the model adopts a multi-scale feature extraction strategy, with a Dice coefficient of 0.803±0.015 (internal test set), which is significantly better than the traditional gastrointestinal surgery bleeding model. To solve the problem of misjudgment caused by the similar chromaticity between the liver parenchyma and the bleeding area, the HSV color enhancement algorithm was introduced to construct a contrastive learning framework, and the artifact interference was suppressed by the positive and negative sample collaborative training mechanism (artifact suppression rate 41.7%).

[0074] The proposed automated surgical skill assessment further incorporates the advantages of deep learning technology. The "average bleeding pixel count per frame" was proposed as a core quantitative metric. Using a 25-frame / second frame extraction strategy, the temporal and spatial distribution characteristics of bleeding events during the CVS phase were focused on. Spearman correlation analysis confirmed that the average bleeding pixel count was significantly negatively correlated with the modified GOALS score (ρ = -0.681, P < 0.001) and strongly associated with core skill dimensions such as bimanual coordination and tissue handling ability. A dual-center validation study (70 internal cases and 30 external cases) demonstrated excellent generalization of the model, with the average bleeding density in the high-skill group (mGOALS ≥ 12) reduced by 68.3% compared with the control group.

[0075] Small sample training of medical images is optimized through transfer learning combined with Xavier initialization. An innovative contrast learning framework based on the HSV color space enhancement algorithm is constructed, combining double-blind labeling with collaborative training on a library of negative examples of bleeding-free frames to address the problem of chromatic misjudgment between liver parenchyma and bleeding areas. An end-to-end, full-process automated evaluation system is developed (taking <2 minutes per case), which is compatible with public datasets and supports multimodal parameter expansion, providing the first objective evaluation paradigm based on biofluid dynamics characteristics for hepatobiliary surgical skills training.

[0076] For example 2, please refer to Figure 2 , Figure 2 Schematic diagram of the structure of an automatic evaluation device for laparoscopic cholecystectomy operating skills in an embodiment of the present invention.

[0077] like Figure 2 As shown, an automatic assessment device for laparoscopic cholecystectomy operating skills, the device comprising:

[0078] Framing module 21: used for acquiring video data in a laparoscopic cholecystectomy operation scene, and performing frame processing on the video data to form framed image data;

[0079] In the specific implementation process of the present invention, when performing a laparoscopic cholecystectomy operation, video capture is performed in the surgical operation scene through a preset camera device, thereby obtaining video data in the laparoscopic cholecystectomy operation scene, and then the video data is framed according to the acquisition exposure frequency of the camera device to form framed image data.

[0080] The frame extraction module 22 is configured to perform frame extraction processing on the framed image data based on a time sequence sampling strategy to obtain frame extracted image data;

[0081] In the specific implementation process of the present invention, the image frame processing is performed on the framed image data based on the time series sampling strategy to obtain the framed image data, including: focusing on the anatomical process of the gallbladder triangle in the key safety field of view stage based on the time series sampling strategy to perform image frame processing to obtain the framed image data.

[0082] Specifically, in order to reduce the subsequent calculation amount and improve the prediction accuracy, the framed image data needs to be processed accordingly. Therefore, it is necessary to perform image extraction processing on the framed image data according to the time sampling strategy of 25 frames / second and focus on the gallbladder triangle anatomy process of the gallbladder triangle anatomy stage of the critical safety field (CVS) to obtain the extracted image data; by constructing a CVS bleeding recognition model, it can be used to identify bleeding conditions during the core operation of LC (laparoscopic cholecystectomy), that is, image frame data with bleeding conditions can be extracted from the framed image data, so that each frame of image data in the extracted image data is image data with bleeding conditions.

[0083] Since the standardized establishment of the critical view of safety (CVS) is a core evaluation indicator in the laparoscopic cholecystectomy safety assessment system, and given that the risk of intraoperative bleeding is significantly negatively correlated with the surgeon's skill level, a structured video analysis method was used to focus on the gallbladder triangle dissection stage of the critical view of safety (CVS), which is the surgical process from serosal incision to complete exposure of the triangle.

[0084] Feature extraction module 23: used for inputting the frame image data into the convergent bleeding feature extraction model to extract bleeding features in the image, and obtaining bleeding feature data corresponding to each frame image data. The bleeding feature extraction model adopts PoolNet network architecture;

[0085] In the specific implementation process of the present invention, the process of training the bleeding feature extraction model to a convergent bleeding feature extraction model includes: obtaining a multi-source heterogeneous laparoscopic cholecystectomy video dataset, wherein the multi-source heterogeneous laparoscopic cholecystectomy video dataset includes N columns of laparoscopic cholecystectomy video datasets in an international standard library and M columns of real clinical laparoscopic cholecystectomy video datasets; performing frame extraction processing on the gallbladder triangle anatomy process of each column of laparoscopic cholecystectomy video data in the multi-source heterogeneous laparoscopic cholecystectomy video dataset in the focus on the key safety field stage using a time series sampling strategy to form a sample frame image dataset; labeling each sample frame image data in the sample frame image dataset based on a double-blind manual labeling method to obtain bleeding pixel point data corresponding to each sample frame image data, wherein the labeling includes three types of morphological features: acute spurting bleeding, tissue oozing and coagulation deposition; constructing a comparative learning framework for the bleeding pixel point data corresponding to each sample frame image data based on the HSV color space enhancement algorithm to obtain a comparative learning framework corresponding to each sample frame image data;

[0086] The bleeding pixel data corresponding to each sample frame image data and the comparative learning framework corresponding to each sample frame image data are mixed and randomly divided into a training data set, a verification data set, and a test data set in a ratio of 8:1:1; the training data set, the verification data set, and the test data set are sequentially input into the bleeding feature extraction model for training, verification, and testing until a converged bleeding feature extraction model is formed.

[0087] Furthermore, the bleeding feature extraction model integrates the VGG-16 network and the ResNet-50 network as the backbone network, and adopts the transfer learning strategy to initialize the training weight values ​​during the training process.

[0088] Furthermore, during the testing process, the bleeding feature extraction model evaluates the test results based on multi-dimensional quantitative indicators, including the Dice similarity coefficient to evaluate segmentation accuracy, the Precision rate to measure the positive predictive value, and the Recall rate to characterize the sensitivity, as follows:

[0089]

[0090] Among them, TP represents the number of correctly identified blood pixels, that is, true positives; FP represents the number of non-blood pixels incorrectly identified as blood pixels, that is, false positives; FN represents the number of unidentified bleeding pixels, that is, false negatives.

[0091] Specifically, the first step is to construct a bleeding feature extraction model, which adopts the PoolNet network architecture and integrates the VGG-16 network and ResNet-50 network as the backbone network. During the training process, the transfer learning strategy is used to initialize the training weight values; it is used to segment bleeding pixels in the gallbladder triangle (average DSC = 0.803 ± 0.015, Precision = 0.860 ± 0.022, Recall = 0.759 ± 0.018).

[0092] When training the bleeding feature extraction model, it is necessary to first build the corresponding data set. Therefore, it is necessary to combine the international standard library (cholec80) with real clinical data (30 cases from Zhujiang Hospital) to ensure that the data can cover a variety of surgical scenarios.

[0093] For these video data, a 25-frame-per-second temporal sampling strategy was adopted, focusing on the gallbladder triangle anatomy during the critical visual field of safety (CVS) phase. A dataset of bleeding pixels was constructed through double-blind manual annotation, with annotation categories including three morphological features: acute spurting bleeding, tissue hemorrhage, and coagulation deposition. To address the issue of color misclassification between liver parenchyma and bleeding areas, an HSV color space enhancement algorithm was introduced to construct a contrastive learning framework. A collaborative training strategy of positive and negative samples was used to suppress artifact interference, significantly improving the model's sensitivity to bleeding areas. The dataset was randomly divided into training, validation, and test sets in an 8:1:1 ratio.

[0094] Since the standardized establishment of the Critical View of Safety (CVS) is a core evaluation indicator in the laparoscopic cholecystectomy safety assessment system, and given that the risk of intraoperative bleeding is significantly negatively correlated with the surgeon's skill level, a structured video analysis method was used to focus on the gallbladder triangle dissection stage - the surgical process from serosal incision to complete exposure of the triangle. Video processing followed the following technical procedures: First, Wondershare Filmora software was used to export the video to a standard format (resolution 1920×1080, frame rate 25fps, bit rate 5999kbps). Subsequently, video frame extraction was performed in the PyCharm Community Edition development environment, extracting key frames every 25 frames (i.e., per second). Finally, the bleeding sites were manually labeled using a double-blind method using the LabelMe annotation system.

[0095] In image processing, based on the principles of digital image processing, the intraoperative bleeding area can be quantitatively identified through video annotation and pixel segmentation technology; the morphological standards of bleeding pixels are strictly defined, and according to the computer vision classification criteria, they mainly include the following three morphological features: (1) acute jet bleeding (sudden pulsating bleeding); (2) tissue oozing (low-flow exudative bleeding); (3) coagulation deposition (localized blood accumulation); the above three types of features are all included in the multimodal annotation system. To address the risk of misjudgment caused by the similar color of the liver parenchyma and the bleeding area in the laparoscopic surgical field, this embodiment will establish a double verification mechanism: by systematically collecting non-bleeding frame images as a counterexample training set, using the HSV color space enhancement algorithm to construct a contrast learning framework, effectively improving the sensitivity and specificity of bleeding feature recognition; this embodiment uses a positive and negative sample collaborative training strategy to significantly reduce the interference of tissue artifacts on the semantic segmentation model.

[0096] The bleeding feature extraction model adopts a dual-stream feature extraction mechanism: VGG-16 and ResNet-50 are integrated as the backbone network, the pre-trained weights (based on the ImageNet dataset) are initialized through a transfer learning strategy, and the newly constructed layers are initialized using Xavier.

[0097] During the training process, the input training data set is obtained through structured video sampling, and a static image data set is generated following the time-series sampling rule of 25 frames per second. The data annotation and processing process strictly follows the evidence-based medicine standards: according to the predefined hemorrhage morphology classification system, the annotation coordinates are mapped to the HSV color space for standardized conversion, and the pixel density of the hemorrhage area is automatically calculated by the convolutional neural network. The data set is randomly divided into training set, validation set and test set in a stratified ratio of 8:1:1 to ensure the statistical validity of the model generalization ability evaluation; the model performance evaluation uses multi-dimensional quantitative indicators: including the Dice similarity coefficient (DSC) to evaluate segmentation accuracy, the precision rate (Precision) to measure the positive predictive value, and the recall rate (Recall) to characterize the sensitivity. A special adaptive feature fusion mechanism is constructed to optimize the multi-scale hemorrhage feature extraction through spatial pyramid pooling. The mathematical definitions of these three indicators are as follows:

[0098]

[0099] Among them, TP represents the number of correctly identified blood pixels, that is, true positives; FP represents the number of non-blood pixels incorrectly identified as blood pixels, that is, false positives; FN represents the number of unidentified bleeding pixels, that is, false negatives.

[0100] The score estimation module 24 is used to perform score estimation processing on the bleeding feature data corresponding to each frame of image data using the mGOALS score prediction algorithm based on random forest regression to obtain a score estimation result.

[0101] In the specific implementation process of the present invention, the mGOALS score prediction algorithm based on random forest regression is used to perform score estimation processing on the bleeding feature data corresponding to each frame of image data to obtain a score estimation result, including: using the bleeding feature data corresponding to each frame of image data to construct a multidimensional bleeding volume feature vector to obtain multidimensional bleeding volume feature data corresponding to the frame image data, wherein the multidimensional bleeding volume feature data is composed of total bleeding volume feature data, peak bleeding volume feature data and average bleeding volume feature data; using the mGOALS score prediction algorithm based on random forest regression to perform score estimation processing on the multidimensional bleeding volume feature data to obtain a score estimation result; wherein the mGOALS score prediction algorithm based on random forest regression realizes automatic association of skill scores through feature extraction, model training, verification and optimization.

[0102] Furthermore, the method of using the bleeding characteristic data corresponding to each frame of image data to construct a multidimensional bleeding volume feature vector to obtain the multidimensional bleeding feature data corresponding to the frame-by-frame image data includes: calculating the bleeding volume feature data using the bleeding pixel count as the core feature of the bleeding characteristic data corresponding to each frame of image data to obtain the total bleeding volume feature data, peak bleeding volume feature data and average bleeding volume feature data corresponding to the frame-by-frame image data; and constructing a multidimensional bleeding volume feature vector based on the total bleeding volume feature data, peak bleeding volume feature data and average bleeding volume feature data corresponding to the frame-by-frame image data to obtain the multidimensional bleeding volume feature data corresponding to the frame-by-frame image data.

[0103] Specifically, an mGOALS score prediction algorithm based on random forest regression was established, which achieved automatic association of skill scores through feature extraction, model training, verification and optimization.

[0104] Specifically, the average bleeding pixel count per frame was used as the core feature, combined with auxiliary indicators such as total bleeding volume and peak bleeding volume to construct a multidimensional bleeding behavior feature vector. A video analysis framework was constructed using 70 internal datasets, and each video was independently encoded and input into the deep learning system. The model executed a fully automated processing flow: video preprocessing - bleeding area segmentation - quantitative indicator output, ultimately generating three sets of key parameters: (1) total bleeding volume (total number of pixels); (2) peak bleeding volume (maximum number of pixels per frame); and (3) average number of bleeding pixels per frame (number of pixels / frame).

[0105] A random forest regression model was used to capture the nonlinear relationship between bleeding characteristics and mGOALS scores through decision tree ensemble learning. Total video bleeding volume was normalized (total bleeding pixel count / total number of frames) to establish a Spearman rank correlation model with the total mGOALS score and each dimension. The Mann-Whitney U test was used to compare the statistical differences in bleeding indicators between the high- and low-skill groups. End-to-end automation was achieved throughout the entire process, from video input to skill scoring, significantly reducing the need for manual intervention. Validation and optimization: The model was validated in an internal test set (70 cases) and an external test set (30 cases). The Kruskal-Wallis test was used to compare the two data sets, and the Fleiss kappa test was used to evaluate the consistency of the scores of three expert surgeons. The results showed that the correlation coefficient (ρ) between the predicted score and the expert score was -0.681 (P < 0.001), and the prediction error rate in the high-skill group (mGOALS ≥ 12 points) was reduced by 42.3% compared with the traditional method.

[0106] For the bleeding feature data corresponding to each frame of image data, the bleeding volume feature data will be calculated with the bleeding pixel count as the core feature, thereby obtaining the total bleeding volume feature data, peak bleeding volume feature data, and average bleeding volume feature data corresponding to the frame-by-frame image data; then, a multidimensional bleeding volume feature vector will be constructed using the total bleeding volume feature data, peak bleeding volume feature data, and average bleeding volume feature data corresponding to the frame-by-frame image data, and the multidimensional bleeding volume feature data corresponding to the frame-by-frame image data will be obtained; finally, the multidimensional bleeding volume feature data will be scored and estimated using the mGOALS score prediction algorithm based on random forest regression to obtain the score estimation result; the mGOALS score prediction algorithm based on random forest regression realizes the automatic association of skill scores through feature extraction, model training, verification, and optimization.

[0107] In an embodiment of the present invention, the fully automatic calculation of the bleeding pixel count in the gallbladder triangle is realized, and its evaluation efficiency is verified to be significantly negatively correlated with the expert score; the problems of strong subjectivity and long time consumption of traditional scale evaluation are solved; by focusing on the key surgical stages, the interference of non-critical operations is avoided and the evaluation specificity is improved; "bleeding pixels per frame" rather than the total bleeding volume is used as an indicator, which enhances the stability and comparability of the results; and end-to-end automation of the entire process from video input to skill scoring is achieved, which significantly reduces the need for manual intervention.

[0108] An embodiment of the present invention provides a computer-readable storage medium, on which an application is stored, and when the program is executed by a processor, the automatic evaluation method of any one of the above embodiments is implemented. The computer-readable storage medium includes, but is not limited to, any type of disk (including floppy disks, hard disks, optical disks, CD-ROMs, and magneto-optical disks), ROM (Read-Only Memory), RAM (Random Access Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, magnetic cards, or optical cards. In other words, the storage device includes any medium that can be used by a device (e.g., a computer, a mobile phone) to store or transmit information in a readable form, which can be a read-only memory, a disk, or an optical disk, etc.

[0109] An embodiment of the present invention further provides a computer application program that runs on a computer and is used to execute the automatic evaluation method of any one of the above embodiments.

[0110] also, Figure 3 It is a schematic diagram of the structure of an electronic device in an embodiment of the present invention.

[0111] The embodiment of the present invention further provides an electronic device, such as Figure 3 The electronic device includes a processor 302, a memory 303, an input unit 304, a display unit 305 and other components. Those skilled in the art will understand that Figure 3 The device structure components shown do not constitute a limitation on all devices, and may include more or fewer components than shown, or combine certain components. The memory 303 can be used to store the application 301 and various functional modules, and the processor 302 runs the application 301 stored in the memory 303, thereby executing various functional applications and data processing of the device. The memory can be an internal memory or an external memory, or include both internal and external memories. The internal memory may include a read-only memory (ROM), a programmable ROM (PROM), an electrically programmable ROM (EPROM), an electrically erasable programmable ROM (EEPROM), a flash memory, or a random access memory. The external memory may include a hard disk, a floppy disk, a ZIP disk, a USB flash drive, a magnetic tape, etc. The memory disclosed in the present invention includes but is not limited to these types of memories. The memory disclosed in the present invention is only an example and not a limitation.

[0112] The input unit 304 is used to receive input signals and keywords entered by the user. The input unit 304 may include a touch panel and other input devices. The touch panel can collect user touch operations on or near it (for example, operations performed by the user using a finger, stylus, or any other suitable object or accessory on or near the touch panel) and drive the corresponding connected device according to a pre-set program. Other input devices may include, but are not limited to, one or more of a physical keyboard, function keys (such as playback control keys, on / off keys, etc.), a trackball, a mouse, a joystick, etc. The display unit 305 can be used to display information entered by the user or information provided to the user, as well as various menus of the terminal device. The display unit 305 can be in the form of a liquid crystal display, an organic light-emitting diode, etc. The processor 302 is the control center of the terminal device, connecting the various parts of the entire device using various interfaces and circuits. It performs various functions and processes data by running or executing software programs and / or modules stored in the memory 303 and accessing data stored in the memory.

[0113] As an embodiment, the electronic device includes: one or more processors 302, a memory 303, and one or more applications 301, wherein the one or more applications 301 are stored in the memory 303 and are configured to be executed by the one or more processors 302, and the one or more applications 301 are configured to execute the automatic evaluation method in any one of the above embodiments.

[0114] In an embodiment of the present invention, the fully automatic calculation of the bleeding pixel count in the gallbladder triangle is realized, and its evaluation efficiency is verified to be significantly negatively correlated with the expert score; the problems of strong subjectivity and long time consumption of traditional scale evaluation are solved; by focusing on the key surgical stages, the interference of non-critical operations is avoided and the evaluation specificity is improved; "bleeding pixels per frame" rather than the total bleeding volume is used as an indicator, which enhances the stability and comparability of the results; and end-to-end automation of the entire process from video input to skill scoring is achieved, which significantly reduces the need for manual intervention.

[0115] In addition, the above is a detailed introduction to the automatic evaluation method and related devices for laparoscopic cholecystectomy operation skills provided by the embodiment of the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for automatically evaluating laparoscopic cholecystectomy operating skills, characterized in that: The method comprises: Acquiring video data of a laparoscopic cholecystectomy operation scene, and performing frame processing on the video data to form framed image data; Performing image extraction processing on the framed image data based on a time sequence sampling strategy to obtain extracted frame image data; Inputting the extracted frame image data into a converged bleeding feature extraction model to perform bleeding feature extraction processing in the image, and obtaining bleeding feature data corresponding to each frame of image data, wherein the bleeding feature extraction model adopts a PoolNet network architecture; The mGOALS score prediction algorithm based on random forest regression was used to perform score estimation processing on the bleeding feature data corresponding to each frame of image data to obtain the score estimation result.

2. The automatic evaluation method according to claim 1, characterized in that: The performing frame extraction processing on the framed image data based on the time sequence sampling strategy to obtain the frame extracted image data includes: Based on the time series sampling strategy, the anatomical process of the gallbladder triangle in the key safety field of view is focused on to perform image frame processing and obtain frame image data.

3. The automatic evaluation method according to claim 1, characterized in that: The process of training the bleeding feature extraction model to a converged bleeding feature extraction model includes: Obtain a multi-source heterogeneous laparoscopic cholecystectomy video dataset, wherein the multi-source heterogeneous laparoscopic cholecystectomy video dataset includes N columns of laparoscopic cholecystectomy video datasets in an international standard library and M columns of real clinical laparoscopic cholecystectomy video datasets; For each column of laparoscopic cholecystectomy video data in the multi-source heterogeneous laparoscopic cholecystectomy video dataset, a time-series sampling strategy is used to perform frame extraction processing on the gallbladder triangle dissection process during the focusing on the key safety field of view to form a sample frame-extracted image dataset; Annotating each sample frame image data in the sample frame image data set based on a double-blind manual annotation method to obtain bleeding pixel point data corresponding to each sample frame image data, wherein the annotation includes three types of morphological features: acute ejective bleeding, tissue oozing, and coagulation deposition; Based on the HSV color space enhancement algorithm, a comparative learning framework is constructed for the bleeding pixel data corresponding to each sample frame image data to obtain a comparative learning framework corresponding to each sample frame image data; The bleeding pixel data corresponding to each sample frame image data and the contrast learning framework corresponding to each sample frame image data are mixed and randomly divided into training data set, verification data set and test data set in a ratio of 8:1:1; The training data set, the validation data set, and the test data set are sequentially input into a bleeding feature extraction model for training, validation, and testing until a converged bleeding feature extraction model is formed.

4. The automatic evaluation method according to claim 3, characterized in that: The PoolNet network architecture of the bleeding feature extraction model integrates the VGG-16 network and the ResNet-50 network as the backbone network, and uses a transfer learning strategy to initialize the training weight values ​​during the training process.

5. The automatic evaluation method according to claim 3, characterized in that: During the testing process, the bleeding feature extraction model evaluates the test results based on multi-dimensional quantitative indicators, including the Dice similarity coefficient to evaluate segmentation accuracy, the Precision rate to measure the positive predictive value, and the Recall rate to characterize the sensitivity, as follows: Among them, TP represents the number of correctly identified blood pixels, that is, true positives; FP represents the number of non-blood pixels incorrectly identified as blood pixels, that is, false positives; FN represents the number of unidentified bleeding pixels, that is, false negatives.

6. The automatic evaluation method according to claim 1, characterized in that: The mGOALS score prediction algorithm based on random forest regression is used to perform score estimation processing on the bleeding feature data corresponding to each frame of image data to obtain a score estimation result, including: Performing multidimensional bleeding volume feature vector construction processing using bleeding feature data corresponding to each frame of image data to obtain multidimensional bleeding volume feature data corresponding to the frame of image data, wherein the multidimensional bleeding volume feature data is composed of total bleeding volume feature data, peak bleeding volume feature data, and average bleeding volume feature data; The multidimensional bleeding volume feature data is scored and estimated using the mGOALS score prediction algorithm based on random forest regression to obtain a score estimation result; wherein the mGOALS score prediction algorithm based on random forest regression realizes automatic association of skill scores through feature extraction, model training, verification and optimization.

7. The automatic evaluation method according to claim 6, characterized in that: The method of constructing a multidimensional bleeding volume feature vector using the bleeding feature data corresponding to each frame of image data to obtain the multidimensional bleeding feature data corresponding to the frame of image data includes: The bleeding volume characteristic data is calculated and processed by taking the bleeding pixel count as the core feature of the bleeding characteristic data corresponding to each frame of image data to obtain the total bleeding volume characteristic data, the peak bleeding volume characteristic data and the average bleeding volume characteristic data corresponding to the frame of image data; A multidimensional bleeding volume feature vector is constructed based on the total bleeding volume feature data, peak bleeding volume feature data and average bleeding volume feature data corresponding to the frame-extracted image data to obtain the multidimensional bleeding volume feature data corresponding to the frame-extracted image data.

8. An automatic evaluation device for laparoscopic cholecystectomy operating skills, characterized in that: The device comprises: Framing module: used to obtain video data in the laparoscopic cholecystectomy operation scene, and perform frame processing on the video data to form framed image data; Frame extraction module: used for performing frame extraction processing on the framed image data based on a time sequence sampling strategy to obtain frame extracted image data; Feature extraction module: used to input the sampled frame image data into the converged bleeding feature extraction model to extract the bleeding features in the image and obtain the bleeding feature data corresponding to each frame of image data. The bleeding feature extraction model adopts the PoolNet network architecture; Score estimation module: used to use the mGOALS score prediction algorithm based on random forest regression to perform score estimation processing on the bleeding feature data corresponding to each frame of image data to obtain the score estimation result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images