Enteroscope retreating quality real-time evaluation method and system based on effective mucous membrane coverage rate

Abstract

The invention provides an effective mucous membrane coverage rate-based colonoscope retreating quality real-time evaluation method and system, and relates to the technical field of medical image calculation and processing. The colonoscope retreating quality real-time evaluation method based on the effective mucous membrane coverage rate comprises the steps that a retreating video is obtained, and the retreating video comprises a real-time retreating video and a historical video; preprocessing the mirror retreating video to obtain an optimized data set; extracting effective frames based on the optimized data set; and inputting the effective frame into a preset image segmentation model to obtain an effective colonic mucosa coverage rate which is used for evaluating the despeculation quality. According to the colonoscope retreating quality real-time evaluation method based on the effective mucous membrane coverage rate, the technical problem that the clinical colonoscope retreating process lacks real-time supervision and feedback in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of medical image computation and processing technology, specifically to a method and system for real-time evaluation of colonoscopy withdrawal quality based on effective mucosal coverage. Background Technology

[0002] Colorectal cancer, a highly prevalent malignant tumor of the digestive system, faces an increasingly serious challenge in its prevention and treatment. Its incidence and mortality rates rank second and fourth among malignant tumors, respectively, posing a significant threat to public health. Colonoscopy, as the internationally recognized gold standard for colorectal cancer screening, directly determines its effectiveness in detecting early lesions through its quality control level.

[0003] Current domestic and international guidelines primarily rely on withdrawal time and adenoma detection rate (ADR) as important quality assessment indicators, but these have significant limitations in clinical application. The withdrawal time during colonoscopy is easily controlled subjectively by the operator, posing a risk of deviation through artificial extension or shortening. The adenoma detection rate cannot be obtained in real time and requires reviewing large amounts of data for calculation. Neither of these two indicators directly reflects the quality of mucosal observation during colonoscopy withdrawal, nor does it provide real-time feedback guidance for endoscopists. During colonoscopy withdrawal, the core of quality control lies in the amount of mucosa observed by the endoscopist through endoscopy manipulation. Numerous studies both domestically and internationally have confirmed a significant positive correlation between the quality of mucosal observation during withdrawal and the adenoma detection rate; the completeness of mucosal exposure directly determines the effectiveness of lesion detection. Therefore, effective measures are urgently needed clinically to assess the quality of mucosal observation during colonoscopy withdrawal to supervise the colonoscopy process, guide endoscopists' operations, and ensure the quality of colonoscopy.

[0004] However, current assessments of mucosal observation quality during colonoscopy withdrawal have the following shortcomings: First, they heavily rely on the subjective experience and immediate judgment of the operating physician, lacking objective and quantitative methods and tools for assessing mucosal observation quality. Second, there is a lack of monitoring during the procedure, particularly regarding fluctuations in withdrawal speed, field of view deviation angles, and mucosal observation conditions. Consequently, it is impossible to identify blind spots and provide corrective guidance in real time during the examination. More importantly, it is impossible to provide real-time supervision and timely feedback on the endoscopist's withdrawal procedure, which is particularly detrimental to junior endoscopists' ability to optimize their procedures, directly leading to a decline in examination quality. Third, the training feedback mechanism is lagging behind. Junior physicians cannot immediately recognize their own operational defects in specific intestinal segments (such as the splenic flexure and the rectosigmoid junction), making it difficult to develop an awareness of visual quality through traditional training, resulting in a prolonged learning curve. Fourth, the application of artificial intelligence technology has limitations. Although computer vision technology has made progress in the field of polyp detection (such as the application of architectures like CNN), it is still in its infancy in terms of quality assessment. Most of the AI-assisted monitoring systems for endoscopy withdrawal quality developed by teams at home and abroad are used to evaluate indirect indicators such as withdrawal time and effective withdrawal time. The basic approach is to split the endoscopic video stream into valid frames and invalid frames. Valid frames refer to images in which the colonic mucosa can be clearly seen, while invalid frames refer to blurry images. Then, the effective withdrawal time index is obtained by calculating the number of valid frames. Among these, the relevant research on the quality assessment of mucosal observation during the withdrawal process is not significant, and there are still key technological gaps, especially the lack of real-time objective assessment of mucosal observation quality and the lack of quantification of the dynamic changes in mucosal coverage with withdrawal speed.

[0005] Therefore, existing technologies need further development. Summary of the Invention

[0006] The purpose of this invention is to overcome the above-mentioned technical deficiencies and provide a method and system for real-time evaluation of colonoscopy withdrawal quality based on effective mucosal coverage, so as to solve the technical problem of lack of real-time monitoring and feedback in the withdrawal process of clinical colonoscopy in related technologies.

[0007] To achieve the above technical objectives, the present invention adopts the following technical solution: providing a real-time evaluation method for colonoscopy withdrawal quality based on effective mucosal coverage, including: acquiring withdrawal video in real time; preprocessing the withdrawal video to obtain an optimized dataset; extracting effective frames based on the optimized dataset; inputting the effective frames into a preset image segmentation model to obtain the effective colonic mucosal coverage, the effective colonic mucosal coverage being used to evaluate withdrawal quality.

[0008] Furthermore, the preprocessing method includes: splitting the video frame by frame to obtain continuous original colon images; cropping the non-target regions of the original colon images to obtain target region images; and enhancing the target region images using a preset method to obtain an optimized dataset.

[0009] Furthermore, a valid frame includes a fully visible colonic mucosa image; a valid frame also includes a partially visible colonic mucosa image. The definition of a valid frame was established by three senior endoscopy experts (who must simultaneously meet the following three conditions: an ADR ≥ 35% in the previous year, more than 5 years of endoscopic experience, and more than 10,000 endoscopic cases performed).

[0010] Furthermore, effective colonic mucosal coverage includes average effective mucosal coverage and real-time effective mucosal coverage. The method for obtaining average effective mucosal coverage is as follows: The method for obtaining real-time effective mucosal coverage is as follows:

[0011] Furthermore, the real-time assessment method for colonoscopy withdrawal quality based on effective mucosal coverage also includes: comparing the effective colonic mucosal coverage with a preset colonic mucosal coverage; if the effective colonic mucosal coverage is continuously less than the first preset colonic mucosal coverage within a first preset time period, a first preset warning is issued; if the effective colonic mucosal coverage is continuously less than the second preset colonic mucosal coverage within a second preset time period, a second preset warning is issued, and the withdrawal video within the second preset time period is automatically saved.

[0012] Furthermore, the real-time assessment method for colonoscopy withdrawal quality based on effective mucosal coverage also includes: analyzing whether the withdrawal video within a second preset time period belongs to a low-value region, wherein the low-value region is the intestinal segment where the real-time effective mucosal coverage is below 50% for 3 seconds and the average effective mucosal coverage shows a downward trend for 3 seconds; if so, the effective colonic mucosal coverage is recalculated.

[0013] Furthermore, the method for recalculating the real-time effective colonic mucosal coverage includes:

[0014] A real-time assessment system for colonoscopy withdrawal quality based on effective mucosal coverage includes: a real-time video acquisition and transmission module for acquiring withdrawal videos in real time; a video preprocessing module for preprocessing the withdrawal videos to obtain an optimized dataset; an intelligent image quality filtering module for extracting effective frames based on the optimized dataset; and an intelligent image mucosal coverage analysis module for inputting the effective frames into a preset image segmentation model to obtain the effective colonic mucosal coverage, which is used to assess the withdrawal quality.

[0015] Furthermore, the preset image segmentation model is specifically a semi-supervised colon image segmentation algorithm model, which includes: a student model, which is used to predict the effective colonic mucosal coverage rate through two effective frames with different strong perturbation processes; and a teacher model, which is used to process manually labeled effective frames and unlabeled effective frames with weak perturbation processes to supervise the prediction behavior of the student model.

[0016] A computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of any of the above methods for real-time assessment of colonoscopy withdrawal quality based on effective mucosal coverage.

[0017] Beneficial effects:

[0018] 1. The real-time assessment method for colonoscopy withdrawal quality based on effective mucosal coverage of the present invention calculates the percentage of effective mucosal area observed in the endoscopic field of view to the total endoscopic field of view using computer vision technology, and proposes a new indicator of effective mucosal coverage. By optimizing and analyzing real-time withdrawal video, the effective colonic mucosal coverage is obtained, which can objectively and quantitatively assess the real-time mucosal observation quality during withdrawal, and solves the technical problem of lack of real-time supervision and feedback in the withdrawal process of clinical colonoscopy in related technologies.

[0019] 2. The colonoscopy withdrawal quality real-time assessment method based on effective mucosal coverage of the present invention uses effective colonic mucosal coverage to monitor and provide timely feedback on the withdrawal operation of endoscopists in real time, enabling junior physicians to optimize their operations in a timely manner and improving the quality of colonoscopy examinations.

[0020] 3. The present invention proposes a semi-supervised colon image segmentation algorithm for the real-time evaluation method of colonoscopy withdrawal quality based on effective mucosal coverage. This solves the technical problems of traditional image segmentation algorithms based on convolutional neural networks (CNN) and U-Net architectures being limited by the scale of the receptive field and unable to obtain global information of the mucosal region image, and the previous fully supervised algorithms requiring a large number of manually labeled images, which is time-consuming and labor-intensive.

[0021] 4. The colonoscopy withdrawal quality real-time assessment method based on effective mucosal coverage of the present invention also proposes an automatic quality scoring and physician feedback module method, which generates an accurate scoring report on the colonoscopy physician's operation, examination quality, and operational level. The operating physician can know the shortcomings in the examination process based on the scoring report and make targeted improvements to the operation and improve the skill level. By establishing a spatial cognitive model of visual quality and generating scoring reports, the technical problem of lagging training feedback mechanism and junior physicians' inability to know their own operational defects in specific intestinal segments in a timely manner is solved. Attached Figure Description

[0022] Figure 1 This is a flowchart of the real-time evaluation method for colonoscopy withdrawal quality based on effective mucosal coverage adopted in the embodiments of the present invention;

[0023] Figure 2 This is a schematic diagram of the structure of the real-time evaluation system for lens removal quality used in an embodiment of the present invention;

[0024] Figure 3 This is a schematic diagram of the image mucosal coverage intelligent analysis module used in an embodiment of the present invention;

[0025] Figure 4 This is a flowchart of the training process for the preset image segmentation model used in the embodiments of the present invention;

[0026] Figure 5 This is a schematic diagram of the network structure of the teacher model and student model used in an embodiment of the present invention;

[0027] Figure 6 This is a schematic diagram of the valid frames extracted according to an embodiment of the present invention;

[0028] Figure 7 This is a schematic diagram of the effective colonic mucosal coverage obtained in an embodiment of the present invention;

[0029] Figure 8 This is a schematic diagram of the image display and real-time early warning module used in an embodiment of the present invention;

[0030] Figure 9 This is a schematic diagram illustrating the correlation analysis results between the average effective mucosal coverage rate and the endoscopist's ADRs over the past 12 months in Embodiment 3 of the present invention. Detailed Implementation

[0031] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0032] According to embodiments of the present invention, a method for real-time assessment of colonoscopy withdrawal quality based on effective mucosal coverage is provided. Please refer to [link to relevant documentation]. Figures 1 to 9 ,include:

[0033] S100 acquires exit shots in real time;

[0034] In this embodiment, a colonoscopy video real-time acquisition and transmission module is provided that is connected to the endoscope host in a standardized manner, which can acquire the original colonoscopy withdrawal video output by the endoscope host in real time during the colonoscopy withdrawal process.

[0035] S200 preprocesses the exit shots of the video to obtain an optimized dataset;

[0036] In this embodiment, the method for preprocessing the exit shot video includes:

[0037] S210 performs frame splitting of the out-of-camera video to obtain continuous raw colon images;

[0038] S220 cropes the non-target region of the original colon image to obtain the target region image;

[0039] S230 uses a preset method to enhance the target region image, resulting in an optimized dataset.

[0040] In practice, the colonoscopy withdrawal video is optimized using a pre-set optimization method to obtain an optimized RGB image.

[0041] Preferably, the colonoscopy withdrawal video is first split into frames at 30 frames per second to obtain continuous raw colon images. Then, the non-target regions of the raw colon images are cropped to obtain RGB colon images of the target regions (i.e., target region images). Finally, the RGB colon image format is unified and the image features are enhanced to obtain optimized RGB colon images (i.e., optimized datasets).

[0042] It should be noted that an RGB image is a digital image composed of three color channels: red, green, and blue. The optimized dataset in this embodiment consists of multiple optimized RGB colon images.

[0043] S300 extracts valid frames based on an optimized dataset;

[0044] See Figure 6 In this embodiment, the effective frame includes a fully visible colonic mucosa image; the effective frame includes a partially visible colonic mucosa image.

[0045] It should be noted that, unlike the valid frames in the prior art, the valid frames here refer to fully or partially clearly visible images of the colonic mucosa, while invalid frames are blurry images. In the prior art, valid frames and invalid frames are also called information frames and non-information frames. In the prior art, valid frames refer to the lumen and partial views of the colonic mucosa that can be clearly seen; invalid frames are images of the colonic mucosa that cannot be clearly seen.

[0046] This embodiment redefines the effective frame, extracting all or part of the clearly visible colonic mucosa image as the effective frame during the video preprocessing stage. This is more conducive to obtaining the effective colonic mucosa coverage rate and quantitatively evaluating the quality of the endoscopic retraction.

[0047] The S400 inputs the effective frames into a preset image segmentation model to obtain the effective colonic mucosal coverage, which is used to evaluate the quality of the endoscopy withdrawal.

[0048] In this embodiment, the effective colonic mucosal coverage rate includes the average effective mucosal coverage rate and the real-time effective mucosal coverage rate. The method for obtaining the average effective mucosal coverage rate is as follows:

[0049]

[0050] The method for obtaining real-time effective mucosal coverage is as follows:

[0051]

[0052] It should be noted that, in this embodiment, the effective frame refers to a colonic mucosa image that is fully or partially visible, and the effective mucosal region refers to the colonic mucosa region that is fully or partially visible within the effective frame. This can be used to determine whether a lesion has occurred. See [link to relevant documentation]. Figure 7 .

[0053] In practice, it is necessary to train a preset image segmentation model. In this embodiment, a semi-supervised method is selected to extract the effective colonic mucosal region in the effective frame. This avoids the time-consuming and laborious process of manually annotating images as required by the fully supervised algorithm, and can extract the effective mucosal coverage in the withdrawal video in real time, quantify the mucosal observation quality in real time, and solve the technical problem of lack of real-time supervision and feedback in the withdrawal process of clinical colonoscopy in related technologies.

[0054] Specifically, see Figure 4 The preset image segmentation models include:

[0055] (1) Image preprocessing unit

[0056] In this embodiment, the input dataset of the image preprocessing unit consists of a small number of manually labeled valid frames and a large number of unlabeled valid frames. The size of the manually labeled and unlabeled valid frames in the input dataset is normalized and subjected to strong and weak perturbation processing to generate standardized input valid frames. The unlabeled valid frames in the input dataset are subjected to weak perturbation processing (rotation ±10°, translation ±5%, brightness adjustment ±15%) and strong perturbation processing (random occlusion (CutOut), random fusion (CutMix), etc.).

[0057] Unlabeled valid frames processed with weak perturbation are input into the teacher model, while unlabeled valid frames processed with strong perturbation are input into the student model. Local features of the input valid frames are extracted using conv2 (two-dimensional convolution). Specifically, Conv2D (two-dimensional convolution) is a core operation in convolutional neural networks (CNNs) used to process two-dimensional data (such as images), and its function is to extract local features of the input data by sliding the convolution kernel.

[0058] The embedding layer converts the 3D feature map (width × height × channels) output by convolution into a serialized vector and superimposes learnable positional encodings, compensating for the loss of spatial relationships in the convolutional layer during serialization. This allows the Transformer module to model global spatial dependencies, essentially acting as a bridge connecting local feature extraction and global relationship modeling. Specifically, the embedding layer converts the 2D feature map after convolution into a serialized high-dimensional vector through block division, flattening, and positional encoding superposition.

[0059] (2) Teacher-student collaborative training unit

[0060] The teacher-student collaborative training unit in this embodiment includes a teacher model and a student model. The teacher model processes manually labeled valid frames and unlabeled valid frames with weak perturbation, generating pseudo-labels to supervise the prediction behavior of the student model. The student model learns robust features through two different strongly perturbation-processed data sets. The cross-entropy loss between unenhanced and strongly enhanced predictions gradually makes the strongly enhanced prediction results approach the unenhanced results during training, improving model consistency. Furthermore, for unlabeled valid frames, a cross-channel consistency loss function is designed to force the model to maintain stable prediction results under multiple perturbation conditions.

[0061] It should be noted that the teacher model and student model in this embodiment have the same network structure, which consists of two modules. See [link to documentation]. Figure 5 Specifically, the network structure includes:

[0062] 1) Encoder module: The encoder module is used to obtain the output multi-scale colonic mucosal feature map from manually labeled valid frames and unlabeled valid frames after weak perturbation processing.

[0063] 2) Decoder module: The decoder module processes the multi-scale colonic mucosal feature map and outputs a pixel-level segmentation probability map. Based on the segmentation mask and visualization processing, the percentage of the effective colonic mucosal region in the endoscopic field of view is calculated to obtain the effective colonic mucosal coverage.

[0064] The encoder module uses ViT-B (or S,L) pre-trained with DINOv2 as the base encoder, and adjusts the input resolution of the effective frames output by the image preprocessing unit to 224*224, outputting multi-scale feature maps (stage-1 to stage-4, with sizes of 1 / 28, 1 / 14, 1 / 7, and 1 / 2 of the original). This adjustment of the effective frame resolution greatly improves the computational efficiency of the encoder.

[0065] Specifically, DINOv2 stands for Depth-Induced Noise for Online Video Self-Supervision V2; ViT-B stands for Vision Transformer-Base; ViT-S stands for Vision Transformer-Small; and ViT-L stands for Vision Transformer-Large.

[0066] Preferably, by inserting LoRA layer by layer into ViT, lightweight adjustments to feature projections are made through low-rank matrices, achieving efficient feature space adaptation and balancing parameter efficiency and model performance. Specifically, Low-Rank Adaptation (LoRA) is a technique for efficiently fine-tuning large pre-trained models. By introducing low-rank matrices to simulate parameter updates, it significantly reduces the number of training parameters while maintaining model performance. This reduces the number of parameters and is suitable for medical image scenarios with limited data and computational resources.

[0067] Specifically, ViT stands for Vision Transformer.

[0068] Preferably, the network structure also includes a feature fusion module, which is an improvement on the DPT architecture. The hierarchical features of ViT are upsampled to the original image resolution through transposed convolution, and skip connections are introduced. After passing through a multi-layer, multi-scale fusion module (including 4 layers of 3x3 convolution + upsampling), and finally through an output convolution module, a pixel-level segmentation probability map is output.

[0069] Specifically, DPT stands for Dense Prediction Transformer.

[0070] In practice, network architecture also includes:

[0071] Deconvolution (also known as Transposed Convolution) is the inverse process of convolution. It enlarges the feature map size through mathematical transpose convolution. Essentially, it reconstructs high-resolution data by padding and sliding the convolution kernel.

[0072] The output convolutional layer (OutConv) is the core module at the end of the decoder. Its function is to convert the high-dimensional feature maps processed by the decoder into the final pixel-level classification results.

[0073] Transposed convolution (also known as deconvolution or transposed 2D convolution) is used to upsample feature maps, restoring a high-resolution output from a low-resolution input through parameterization.

[0074] Preferably, the teacher-student model training process simultaneously utilizes strong supervision of labeled data and consistency constraints of unlabeled data.

[0075] In practice, the loss function in this embodiment is designed as follows:

[0076] 1) There is a supervisory loss:

[0077] For manually labeled valid frame data, after weak perturbation, all data is input into the teacher model for fully supervised training. In this embodiment, the weak perturbation specifically includes rotation ±10°, translation ±5%, and brightness adjustment ±15%.

[0078] The mask segmentation loss is calculated by weighted summation of Focal Loss and Dice Loss to alleviate class imbalance. Specifically, the method for calculating the mask segmentation loss is as follows:

[0079]

[0080] Among them, L mask For mask segmentation loss, M is the predicted label result output by the teacher's model after the manually annotated valid frame data is processed.

[0081] The colonic mucosa and background classification loss is based on the label smoothing KL divergence loss, which constrains the consistency between the predicted attribute labels and the manually labeled labels. Specifically, the method for calculating the colonic mucosa and background classification loss is as follows:

[0082]

[0083] Among them, L classify For the classification loss of colonic mucosa and background, It is a true classification, P c (i) is the predicted category.

[0084] Specifically, referring to ImageNet training settings, ∈ = 0.1 is a commonly used value for label smoothing, which balances the label correction strength and model confidence.

[0085] For supervised losses, the total loss is calculated as L. sup =L mask +L classify ;

[0086] Among them, L sup For supervised total loss, L mask For mask segmentation loss, L classify The loss is attributed to the classification of colonic mucosa against the background.

[0087] 2) Semi-supervised loss:

[0088] For the unlabeled valid frame data, apply one weak perturbation w and two different strong perturbations S1 and S2 respectively. The valid frame data after one weak perturbation w is input into the teacher model, and the valid frame data after two different strong perturbations S1 and S2 is input into the student model.

[0089] Specifically, the teacher model generates pseudo-labels. As a supervisory signal, the student model generates two predictions. The mean squared error of the student model's two predicted labels and the pseudo-labels generated by the teacher model is calculated using the following method:

[0090]

[0091] In order to suppress noise propagation, the loss is calculated only for pixels with confidence thresholds in the pseudo-labels.

[0092] Finally, the total loss is the sum of the supervised loss and the semi-supervised loss:

[0093] L total =L sup +β(t)L consist ;

[0094] Wherein, β(t) is a variable weight coefficient, which increases from 0.1 to 0.5 with each round t.

[0095] Preferably, β(t) is manually set based on experience during training. In the early stage, β = 0.1 allows the model to prioritize learning supervised signals (labeled data) and avoid interference from unsupervised terms. In the later stage, β = 0.5 gradually strengthens the consistency constraint and improves the generalization ability.

[0096] Preferably, training is stopped when the loss function stops decreasing for multiple consecutive training rounds, thus completing the training process.

[0097] In some embodiments, the number of training rounds for the teacher-student model is fixed.

[0098] Preferably, see Figure 8 This embodiment is equipped with an image display and real-time warning module, which displays the current effective mucosal coverage rate as a percentage value on the information panel of the monitor in real time.

[0099] In this embodiment, the real-time assessment method for colonoscopy withdrawal quality based on effective mucosal coverage further includes:

[0100] The real-time effective colonic mucosal coverage rate was compared with the preset colonic mucosal coverage rate.

[0101] If the real-time effective colonic mucosal coverage rate is continuously less than the first preset colonic mucosal coverage rate within the first preset time period, a first preset warning will be issued.

[0102] If the real-time effective colonic mucosal coverage rate is consistently lower than the second preset colonic mucosal coverage rate within the second preset time period, a second preset warning will be issued, and the withdrawal video within the second preset time period will be automatically saved.

[0103] Example 1:

[0104] In this embodiment, a multi-level alarm mechanism is triggered based on a pre-set mucosal coverage threshold. If the real-time effective mucosal coverage of the current intestinal segment is less than 60% for 3 consecutive seconds, an alarm will be triggered. If the real-time effective mucosal coverage of the current intestinal segment is less than 50% for 3 consecutive seconds and the average effective mucosal coverage shows a downward trend for 3 consecutive seconds, the endoscopist will be prompted to perform treatment procedures via voice prompts.

[0105] In practice, real-time assessment methods for colonoscopy withdrawal quality based on effective mucosal coverage also include:

[0106] Analyze whether the videos of characters leaving the camera within the second preset time period belong to the low-value area;

[0107] If so, then recalculate the real-time effective colonic mucosal coverage rate.

[0108] It should be noted that if the real-time effective mucosal coverage of the current intestinal segment is detected to be less than 50% for 3 consecutive seconds, the endoscopist will be prompted by voice to perform the operation. At the same time, the current mucosal coverage will not be calculated, and the video of the intestinal segment with the low mucosal coverage value and the randomly captured image will be automatically saved. After the subsequent examination, the physician can use the video record to confirm the actual situation of the area with the low effective mucosal coverage value. The subsequent video will be input into the next module for detection.

[0109] This embodiment also includes a low-value area detection module, which is used to confirm whether the automatically saved intestinal segment video and randomly captured image with low mucosal coverage value belong to the low-value area. If so, the mucosal coverage of the current operating physician will be recalculated.

[0110] Example 2:

[0111] In this embodiment, the method for recalculating the real-time effective colonic mucosal coverage includes:

[0112]

[0113] The real-time evaluation method for colonoscopy withdrawal quality based on effective mucosal coverage in this embodiment further includes: after withdrawal, generating a mucosal coverage score report based on the withdrawal video.

[0114] Specifically, this embodiment includes an automatic quality scoring and physician feedback module, which generates an accurate scoring report on the physician's operation, examination quality, and skill level during colonoscopy. Based on the scoring report, the operating physician can identify the shortcomings in the examination process and make targeted improvements to enhance their skills.

[0115] Preferably, this embodiment also includes a system extension and integration module for continuous improvement and optimization of the algorithm, while providing a modular extension interface to facilitate the integration of other algorithm modules.

[0116] Example 3:

[0117] See Figure 9 In this embodiment, 209 real colonoscopy withdrawal videos from 16 endoscopists were collected. Then, clinical pathology data were obtained and the adenoma detection rate of each endoscopist over the past 12 months was calculated (see Table 1). The average effective mucosal coverage rate of all videos for each endoscopist was obtained using the real-time assessment of colonoscopy withdrawal quality based on effective mucosal coverage rate in this embodiment. The average effective mucosal coverage rate of all videos for each endoscopist was analyzed with the ADR of the endoscopist over the past 12 months. The results showed a significant correlation between the two.

[0118] Specifically, see Figure 9 In this embodiment, r = 0.609, P = 0.012, and the 95% confidence interval is 0.163-0.849, where r is the correlation coefficient and P is the significance level.

[0119] It should be noted that in correlation analysis, r (correlation coefficient) and P (significance level) are two core indicators, used to measure the strength and direction of the correlation between variables, and the statistical significance of this correlation, respectively. r is a quantitative indicator that measures the degree of linear correlation between two variables, with a value range of [-1, 1]. The larger the absolute value, the stronger the correlation. P is used to judge the statistical significance of r.

[0120] Specifically, P < α, where α is the significance level, usually set to 0.05 or 0.01, which is an artificially set threshold.

[0121] It's important to note that ADR (Adenoma Detection Rate) is a core performance indicator for colonoscopy quality. ADR is defined as the percentage of patients found to have at least one adenoma in the average-risk population undergoing colonoscopy. For example, if a physician performs colonoscopies on 1000 patients in a year, and 300 are pathologically diagnosed with adenomas, then that physician's ADR is 30%.

[0122] Table 1. All video recordings, previous ADRs, and average effective mucosal coverage data for each endoscopist.

[0123]

[0124]

[0125] This embodiment provides a real-time assessment system for colonoscopy withdrawal quality based on effective mucosal coverage. The real-time withdrawal quality assessment system includes:

[0126] The real-time video acquisition and transmission module is used to acquire back-to-back video footage in real time.

[0127] The video preprocessing module is used to preprocess the exit shot video to obtain an optimized dataset;

[0128] The image quality intelligent filtering module is used to extract valid frames based on an optimized dataset.

[0129] The image mucosal coverage intelligent analysis module is used to input effective frames into a preset image segmentation model to obtain the effective colonic mucosal coverage, which is used to evaluate the quality of endoscopic withdrawal.

[0130] Specifically, the real-time assessment system for lens removal quality also includes:

[0131] The image display and real-time early warning module is used to display and monitor the effective mucosal coverage in real time and trigger a multi-level alarm mechanism based on a preset mucosal coverage threshold.

[0132] The automatic quality scoring and physician feedback module is used to automatically generate a mucosal coverage score report for the physician after the examination, and to collect the physician's evaluation feedback on the use of the system for auxiliary examinations.

[0133] The system has an extension and integration module for continuous improvement and optimization of the algorithm, and also provides a modular extension interface to facilitate the integration of other algorithm modules.

[0134] The preset image segmentation model in this embodiment is specifically a semi-supervised colon image segmentation algorithm model, which includes: a student model, which is used to predict the effective colonic mucosal coverage rate through effective frames processed by two different strong perturbations; and a teacher model, which is used to process manually labeled effective frames and unlabeled effective frames processed by weak perturbation to supervise the prediction behavior of the student model.

[0135] This embodiment provides a computer-readable storage medium storing computer-readable instructions. When executed by a processor, the computer-readable instructions implement the steps of any of the above methods for real-time evaluation of colonoscopy withdrawal quality based on effective mucosal coverage.

[0136] This invention can take the form of a computer program product implemented on one or more storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing program code. Computer-readable storage media include permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer-readable storage media include, but are not limited to: novel memories such as phase-change memory / resistive random access memory / magnetic memory / ferroelectric memory (PRAM / RRAM / MRAM / FeRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, read-only optical disc read-only memory (CD-ROM), digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic magnetic disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device.

[0137] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0138] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0139] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0140] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0141] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A real-time evaluation method of colonoscopy withdrawal quality based on effective mucosal coverage, characterized in that, The method comprises the following steps: acquiring a video of a withdrawal of a lens, wherein the video of the withdrawal of the lens comprises a real-time video of the withdrawal of the lens and a historical video of the withdrawal of the lens; preprocessing the video of the withdrawal of the lens to obtain an optimized data set; extracting effective frames based on the optimized data set; inputting the effective frames into a preset image segmentation model to obtain effective colonic mucosa coverage, wherein the effective colonic mucosa coverage is used for evaluating the quality of the withdrawal of the lens.

2. The real-time evaluation method of withdrawal quality of colonoscopy based on effective mucosal coverage according to claim 1, characterized in that, The preprocessing method comprises the following steps: frame processing of the video of the withdrawal of the lens to obtain continuous original colonic images; target region cropping of the original colonic images to obtain target region images; enhancing the target region images by using a preset method to obtain the optimized data set.

3. The real-time evaluation method of withdrawal quality of colonoscopy based on effective mucosal coverage according to claim 1, characterized in that, The effective frames comprise all clearly visible colonic mucosa images; and / or The effective frames comprise part of the clearly visible colonic mucosa images.

4. The real-time evaluation method of withdrawal quality of colonoscopy based on effective mucosal coverage according to claim 1, characterized in that, The effective colonic mucosa coverage comprises an average effective mucosa coverage and a real-time effective mucosa coverage, wherein the method for acquiring the average effective mucosa coverage is as follows: The method for acquiring the real-time effective mucosa coverage is as follows: wherein the effective mucosa region is part or all of the clearly visible colonic mucosa regions in the effective frames.

5. The real-time evaluation method of withdrawal quality of colonoscopy based on effective mucosal coverage according to claim 4, characterized in that, The real-time evaluation method of the quality of the withdrawal of the lens based on the effective mucosa coverage further comprises the following steps: comparing the real-time effective colonic mucosa coverage with a preset colonic mucosa coverage; if the real-time effective colonic mucosa coverage is continuously less than a first preset colonic mucosa coverage within a first preset time period, a first preset warning is given; if the real-time effective colonic mucosa coverage is continuously less than a second preset colonic mucosa coverage within a second preset time period, a second preset warning is given, and the video of the withdrawal of the lens within the second preset time period is automatically saved.

6. The real-time evaluation method of withdrawal quality of colonoscopy based on effective mucosal coverage according to claim 5, characterized in that, The real-time evaluation method of the quality of the withdrawal of the lens based on the effective mucosa coverage further comprises the following steps: analyzing whether the video of the withdrawal of the lens within the second preset time period belongs to a low-value region, wherein the low-value region is an intestinal segment in which the real-time effective mucosa coverage is continuously less than 50% for 3 seconds and the average effective mucosa coverage shows a downward trend for 3 seconds; if yes, the real-time effective colonic mucosa coverage is recalculated.

7. The real-time evaluation method of withdrawal quality of colonoscopy based on effective mucosal coverage according to claim 6, characterized in that, The method for recalculating the real-time effective colonic mucosa coverage comprises the following steps:

8. A real-time evaluation system for colonoscopy withdrawal quality based on effective mucosal coverage, characterized in that, The real-time evaluation system of the quality of the withdrawal of the lens based on the effective mucosa coverage comprises the following steps: a video real-time acquisition and transmission module, which is used for acquiring a video of a withdrawal of a lens in real time; a video preprocessing module, which is used for preprocessing the video of the withdrawal of the lens to obtain an optimized data set; an image quality intelligent screening module, which is used for extracting effective frames based on the optimized data set; an image mucosa coverage intelligent analysis module, which is used for inputting the effective frames into a preset image segmentation model to obtain effective colonic mucosa coverage, wherein the effective colonic mucosa coverage is used for evaluating the quality of the withdrawal of the lens.

9. The real-time assessment system of colonoscopy withdrawal quality based on effective mucosal coverage according to claim 8, characterized in that, The preset image segmentation model is specifically a semi-supervised colonic image segmentation algorithm model, wherein the semi-supervised colonic image segmentation algorithm model comprises the following steps: a student model for predicting the effective colonic mucosa coverage from twice different strong perturbation processed valid frames; a teacher model for processing artificially labeled valid frames and weak perturbation processed unlabeled valid frames to supervise the prediction behavior of the student model.

10. A computer-readable storage medium having stored thereon computer-readable instructions, wherein, the computer readable instructions, when executed by the processor, implement the steps of the method for real-time evaluation of colonoscopy withdrawal quality based on effective mucosa coverage according to any one of claims 1-8.

Images