Electronic laryngoscope medical image detection method based on local fine-grained contour features
By constructing an image segmentation network model and extracting fine-grained features from laryngoscope images, the problem of reliance on experience in electronic laryngoscope detection methods is solved, achieving more efficient and accurate organ contour recognition and anomaly detection.
Patent Information
- Application Number
- CN202211042016.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-08-29
- Publication Date
- 2026-01-09
- Estimated Expiration
- 2042-08-29
AI Technical Summary
Existing electronic laryngoscope image detection methods rely on doctors' experience, are prone to missed and false detections, are inefficient, and struggle to accurately identify organ outlines and mark abnormalities in complex backgrounds.
By acquiring and preprocessing laryngoscope images, an image segmentation network model is constructed to extract fine-grained features of organ contours. ResNet network is then used for classification to generate detection results.
It improves the accuracy of organ abnormality detection in complex contexts, reduces missed and false detections, and increases detection efficiency.
Smart Images
Figure CN115393326B_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of electronic laryngoscope medical image detection, in particular to an electronic laryngoscope medical image detection method based on local fine-grained contour features. BACKGROUND
[0002] Electronic laryngoscope is a common examination method for diagnosing organ diseases. It is the most direct and effective means to observe the mucosa of hollow organs. In the laryngoscope image, the symptoms of organ diseases are concentrated in the internal contour of the organ, usually characterized by neoplasm, sausage-shaped mass and small protrusions. Electronic laryngoscope has the characteristics of light volume, long and convenient insertion, good insertion, so that the lens enters the laryngeal cavity closer to the lesion site, and can clearly see the subtle changes of the respiratory tract, thereby realizing faster diagnosis and treatment. The current electronic endoscope has some defects: first, the traditional laryngoscope image screening is labor-intensive and largely depends on the experience of medical experts; second, endoscopy requires concentrated attention, and doctors are prone to errors such as missed detection and misdiagnosis, and the nasopharyngeal part is delicate, so if missed detection and misdiagnosis occur, the patient will pay a high price; finally, doctors need to manually add the corresponding symptoms and descriptions to the report, which will increase the examination time and lead to low efficiency. Therefore, it is very necessary to design an electronic laryngoscope medical image detection method based on local fine-grained contour features. SUMMARY
[0003] The purpose of the present application is to provide an electronic laryngoscope medical image detection method based on local fine-grained contour features, which can identify and mark the organ contour in a complex background, generate a detection result, and improve the accuracy of organ abnormality detection.
[0004] To achieve the above purpose, the present application provides the following scheme:
[0005] An electronic laryngoscope medical image detection method based on local fine-grained contour features, comprising the following steps:
[0006] Step 1: Obtain an electronic laryngoscope image and pre-process it to obtain a training data set;
[0007] Step 2: Construct a network model for image segmentation and train it to obtain an organ segmentation model, input the training data set into the organ segmentation model, segment it, and generate an organ segmentation data set;
[0008] Step 3: Extract the contour edge based on the pixel point value of the organ segmentation data set, retain the organ edge saliency feature, and discard the remaining part of the image to generate an organ internal contour data set;
[0009] Step 4: Extract fine-grained features based on the tangent angle of each point of the organ edge in the organ inner contour data set, and combine all the features into final features;
[0010] Step 5: According to the final features, use the ResNet network to classify the laryngoscope image.
[0011] In step 1, the electronic laryngoscope image is acquired and preprocessed to obtain the training data set, which specifically includes:
[0012] Step 101: Collect images of correct classification of hospital electronic laryngoscope;
[0013] Step 102: Crop the collected images, identify the rectangular frame for displaying the laryngoscope image through image processing recognition method, crop the useless area outside the rectangular frame, and obtain the cropped image;
[0014] Step 103: Remove the blurred images in the cropped image, and name and save each image according to the disease type, number and serial number, to obtain the training data set.
[0015] In step 2, the network model of image segmentation is constructed and trained to obtain the organ segmentation model, the training data set is input into the organ segmentation model for segmentation to generate the organ segmentation data set, which specifically includes the following steps:
[0016] Step 201: Collect the training data set, mark the organ edge in the image of the training data set through irregular polygon, and mark the organ category;
[0017] Step 202: Construct the network model of image segmentation, take the marked image data as training sample, train the network model, and obtain the organ segmentation model after training;
[0018] Step 203: Input the training data set into the organ segmentation model for segmentation, output the organ category and contour image, and generate the organ segmentation data set according to the organ category and contour image.
[0019] In step 3, the contour edge extraction based on pixel point value is performed on the organ segmentation data set, the organ edge saliency features are retained, and the remaining part of the image is discarded to generate the organ inner contour data set, which specifically includes:
[0020] Step 301: Perform contour edge extraction based on pixel point value on the organ segmentation data set, retain the organ edge saliency features, and discard the remaining part of the image to generate the organ inner contour data set, and obtain the internal contour data of the organ;
[0021] Step 302: constructing a network model for image segmentation, taking the inner contour curve image data in the organ inner contour data as a training sample to train the network model, and obtaining an organ inner contour segmentation model;
[0022] Step 303: inputting the organ segmentation data set into the organ inner contour segmentation model to obtain an organ inner contour data set.
[0023] In step 4, fine-grained features are extracted based on the tangent angle of each point of the organ edge in the organ inner contour data set, and all features are combined into final features, including the following steps:
[0024] Step 401: collecting a set of edge points of the organ in the organ inner contour data set, C={cp1, cp2, cp3,..., cp n}, calculating the tangent angle a(p n ) of each point p n , and the feature vector of each point cp n (i,j) is represented as A i,j,a =[i,j,a i,j ];
[0025] Step 402: extracting fine-grained features based on the tangent angle a(p n ) of each point of the organ edge in the organ inner contour data set, and combining all features into final features.
[0026] According to the specific embodiments provided by the present application, the following technical effects are disclosed: the electronic laryngoscope medical image detection method based on local fine-grained contour features provided by the present application, the method includes obtaining an electronic laryngoscope image, and pre-processing the same to obtain a training data set, constructing a network model for image segmentation, and training the same to obtain an organ segmentation model, inputting the training data set into the organ segmentation model for segmentation to generate an organ segmentation data set, extracting the contour edge based on the pixel point value of the organ segmentation data set, retaining the organ edge saliency feature, and discarding the remaining part of the image to generate an organ inner contour data set, extracting fine-grained features based on the tangent angle of each point of the organ edge in the organ inner contour data set, and combining all features into final features, and predicting the laryngoscope image category according to the final features; the method can recognize and mark the organ contour under the condition of complex background to generate a detection result, and improve the accuracy of organ abnormality detection. BRIEF DESCRIPTION OF DRAWINGS
[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments. Obviously, the drawings described below only illustrate some of the embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.
[0028] Figure 1 A flowchart of a local fine-grained contour feature-based electronic laryngoscope medical image detection method according to an embodiment of the present application is shown in FIG. 1. DETAILED DESCRIPTION
[0029] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the present application.
[0030] The purpose of the present application is to provide a local fine-grained contour feature-based electronic laryngoscope medical image detection method, which can identify and mark organ contours in complex background conditions, generate detection results, and improve the accuracy of organ anomaly detection.
[0031] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the present application will be further described in detail below with reference to the drawings and specific embodiments.
[0032] As shown in FIG. 1, the local fine-grained contour feature-based electronic laryngoscope medical image detection method according to an embodiment of the present application includes the following steps: Figure 1
[0033] Step 1: Obtain an electronic laryngoscope image and pre-process it to obtain a training data set;
[0034] Step 2: Construct a network model for image segmentation and train it to obtain an organ segmentation model. Input the training data set into the organ segmentation model to perform segmentation and generate an organ segmentation data set;
[0035] Step 3: Extract the contour edges based on the pixel point values of the organ segmentation data set, retain the organ edge saliency features, and discard the remaining part of the image to generate an intra-organ contour data set;
[0036] Step 4: Extract fine-grained features based on the tangent angle of each point of the organ edge in the intra-organ contour data set, and combine all the features into final features;
[0037] Step 5: According to the final features, use the ResNet network to classify the laryngoscope image, and report the generation, and record the accuracy.
[0038] In step 1, the electronic laryngoscope image is acquired and preprocessed to obtain a training data set, which specifically includes:
[0039] Step 101: Collect and organize the images of the hospital electronic laryngoscope classification, and the corresponding disease and number need to be collected;
[0040] Step 102: Crop the collected images, identify the rectangular frame for displaying the laryngoscope image through image processing recognition method, crop the useless area outside the rectangular frame, get the cropped image, keep the image part inside the frame, the size is 224x224;
[0041] Step 103: Remove the blurred images in the cropped images, and name and save each image according to the disease, number and sequence number, get the training data set, use this naming method to realize image traceability.
[0042] In step 2, the network model of image segmentation is constructed and trained to obtain the organ segmentation model. The training data set is input into the organ segmentation model for segmentation to generate the organ segmentation data set, which includes the following steps:
[0043] Step 201: Collect the training data set, mark the organ edge in the image of the training data set through irregular polygon, and mark the organ category;
[0044] Step 202: Construct the network model of image segmentation, use the marked image data as training sample, train the network model, after training, get the organ segmentation model, which is used to identify and segment the organ in the detection image;
[0045] Step 203: Input the training data set into the organ segmentation model for segmentation, output the organ category and contour image, and generate the organ segmentation data set according to the organ category and contour image.
[0046] In step 3, the organ segmentation data set is extracted based on the pixel point value of the contour edge, the organ edge saliency feature is retained, and the remaining part of the image is discarded to generate the organ inner contour data set, which is specifically:
[0047] Step 301: Extract the organ segmentation data set based on the pixel point value of the contour edge, retain the organ edge saliency feature, and discard the remaining part of the image to generate the organ inner contour data set, get the internal contour data of the organ;
[0048] Wherein, specifically: the sum of the pixel values of adjacent points is recorded as When, it indicates that the pixel value at point (i, j) has changed, and the point is recorded as M i,K , where K represents that the point is the Kth pixel value mutation in the ith row. From the overall organ contour, it can be seen that each row can contain 0-6 pixel value mutation points, and the following processing will be performed for the value:
[0049] 1) When the total number of mutation points of each row of pixel values is 4 (K=4), the second and third mutation points recorded are the points on the inner contour of the organ;
[0050] 2) When the total number of pixel mutation points of each row of pixels is less than 4 (K<4), the mutation points that appear are points on the outer contour of the organ or points at the top and bottom of the organ, and are not taken as points on the inner contour;
[0051] 3) When the total number of pixel mutation points of each row of pixels is greater than 4 (K>4), these points often contain abnormal protrusions on the inner contour, so the points other than the first point and the last point in these rows are taken as points on the inner contour
[0052] Step 302: Construct a network model for image segmentation, take the inner contour curve image data in the organ inner contour data as a training sample to train the network model, and obtain an organ inner contour segmentation model;
[0053] Step 303: Input the organ segmentation data set into the organ inner contour segmentation model to obtain an organ inner contour data set.
[0054] In step 4, fine-grained features are extracted based on the tangent angle of each point on the edge of the organ in the organ inner contour data set, and all features are combined into final features, including the following steps:
[0055] Step 401: Collect the edge point set of the organ in the organ inner contour data set, which is C={cp1, cp2, cp3,..., cp n}, calculate the tangent angle a(p n ) of each point p n , and the feature vector of each point cp n (i, j) is represented as A i,j,a =[i, j, a i,j ];
[0056] Step 402: Extract fine-grained features based on the tangent angle a(p n ) of each point on the edge of the organ in the organ inner contour data set, and combine all features into final features.
[0057] The application provides an electronic laryngoscope medical image detection method based on local fine-grained contour features, which comprises the following steps: acquiring an electronic laryngoscope image, pre-processing the electronic laryngoscope image to obtain a training data set, constructing a network model for image segmentation, training the network model to obtain an organ segmentation model, inputting the training data set into the organ segmentation model, segmenting, generating an organ segmentation data set, extracting a contour edge based on pixel point values of the organ segmentation data set, retaining organ edge saliency features, discarding the remaining part of the image, generating an intra-organ contour data set, extracting fine-grained features based on the tangent angle of each point of the organ edge in the intra-organ contour data set, combining all the features into final features, and predicting the laryngoscope image category according to the final features; the method can recognize and mark the organ contour under the condition of complex background, generate a detection result, and improve the accuracy of organ abnormality detection.
[0058] The principles and implementation manners of the present application are described by applying specific examples in the present text, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for the general technical personnel in the field, the specific implementation manners and application ranges will be changed according to the idea of the present application. In conclusion, the content of the present specification should not be understood as the limitation of the present application.
Claims
1. A method for detecting electronic laryngoscope medical images based on local fine-grained contour features, characterized in that, The method comprises the following steps: Step 1: obtaining electronic laryngoscope images and pre-processing the same to obtain a training data set; Step 2: constructing a network model for image segmentation and training the same to obtain an organ segmentation model, inputting the training data set into the organ segmentation model to perform segmentation and generating an organ segmentation data set; Step 3: performing contour edge extraction based on pixel point values on the organ segmentation data set, retaining organ edge saliency features and discarding the remaining part of the image to generate an organ inner contour data set; Step 4: extracting fine-grained features based on the tangent angle of each point of the organ edge in the organ inner contour data set and combining all the features into final features; Step 5: using a ResNet network to classify laryngoscope images according to the final features; In step 4, fine-grained features are extracted based on the tangent angle of each point of the organ edge in the organ inner contour data set, and all the features are combined into final features, which specifically comprises the following steps: Step 401: Collect the edge point set of the organ in the organ inner contour data set, as , the tangent inclination angle of each point is calculated , and the feature vector of each point is represented as ; Step 402: based on the tangent angle of each point of the edge of the organ in the organ inner contour data set Extract fine-grained features and combine all features into final features. 2.The electronic laryngoscope medical image detection method based on local fine-grained contour features according to claim 1, wherein, In step 1, electronic laryngoscope images are obtained and pre-processed to obtain a training data set, which specifically comprises: Step 101: collecting images of electronic laryngoscopes classified correctly in hospitals; Step 102: cropping the collected images, identifying a rectangular frame for displaying laryngoscope images through image processing recognition methods, cropping useless areas outside the rectangular frame to obtain cropped images; Step 103: removing blurred images in the cropped images and naming and saving each image according to disease type, number and serial number to obtain a training data set. 3.The electronic laryngoscope medical image detection method based on local fine-grained contour features according to claim 2, characterized in that, In step 2, a network model for image segmentation is constructed and trained to obtain an organ segmentation model, the training data set is input into the organ segmentation model to perform segmentation and generate an organ segmentation data set, which specifically comprises the following steps: Step 201: collecting the training data set, marking the organ edge in the image of the training data set through an irregular polygon, and marking the organ category; Step 202: constructing a network model for image segmentation, taking the marked image data as a training sample to train the network model, and obtaining an organ segmentation model after the training is completed; Step 203: inputting the training data set into the organ segmentation model to perform segmentation, outputting the organ category and contour image, and generating an organ segmentation data set according to the organ category and contour image.
4. The electronic laryngoscopy medical image detection method based on local fine-grained contour features according to claim 3, characterized in that, In step 3, the organ segmentation data set is subjected to contour edge extraction based on pixel point values, the organ edge saliency features are retained, and the remaining part of the image is discarded to generate an organ inner contour data set, which specifically comprises: Step 301: performing contour edge extraction based on pixel point values on the organ segmentation data set, retaining organ edge saliency features and discarding the remaining part of the image to generate an organ inner contour data set, obtaining the internal contour data of the organ; Step 302: constructing a network model for image segmentation, taking the internal contour curve image data in the internal contour data of the organ as a training sample to train the network model, and obtaining an organ inner contour segmentation model; Step 303: inputting the organ segmentation data set into the organ inner contour segmentation model to obtain an organ inner contour data set.