Nasal cavity and throat image structure recognition method, device and equipment and medium

Abstract

The invention provides a nasal cavity and throat image structure recognition method and device, equipment and a medium, and the method comprises the steps: inputting a to-be-recognized nasal cavity and throat endoscope image into a first structure recognition model, and determining a main structure type to which the to-be-recognized nasal cavity and throat endoscope image belongs; if the main structure category is a laryngeal glottis region category, determining a vocal cord state subtype recognition result to which the to-be-recognized nasal pharyngoscope image belongs based on central region features of the to-be-recognized nasal pharyngoscope image; and if the main structure category is a non-glottis region category, inputting the to-be-identified nasal pharyngoscope image into the second structure identification model, determining an anatomical position subtype identification result under the main structure category to which the to-be-identified nasal pharyngoscope image belongs, and identifying the anatomical position subtype of the to-be-identified nasal pharyngoscope image by introducing a hierarchical identification strategy. The problems of feature confusion and class imbalance caused by task target conflicts of a traditional single-stage model are effectively solved, and the accuracy of vocal cord state judgment and sub-part positioning and the model generalization ability are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of medical image processing, and in particular to a method, apparatus, device, and medium for recognizing the structure of nasal and pharyngeal images. Background Technology

[0002] With the development of intelligent medical imaging diagnosis, automatic structural recognition technology for nasopharyngoscopy images has become an important tool to assist doctors in lesion localization and functional assessment. These images contain various anatomical structures, including the nasal cavity, nasopharynx, glottis, and hypopharynx, and these structures differ significantly in morphology, texture, and function, requiring accurate classification through computer vision and machine learning methods. However, because endoscopic images are easily affected by factors such as shooting angle, lighting conditions, and mucosal reflection, achieving high-precision, multi-level structural recognition has always been a technical challenge in this field.

[0003] Traditional nasal and pharyngeal image recognition methods typically employ end-to-end single-stage classification models (such as directly inputting images into a 14-classifier) ​​in an attempt to distinguish all anatomical structure categories and states simultaneously using a single model.

[0004] However, while the above methods simplify the process, they overlook the fundamental differences between the glottic and non-glottic structures: the glottic region requires further analysis of the dynamic functional state of the vocal cords (e.g., open / closed), while the non-glottic region requires detailed localization of specific anatomical locations (e.g., left / right nasal cavity). A single model struggles to simultaneously meet the dual needs of morphological classification and functional state discrimination, and is prone to insufficient generalization ability due to class imbalance and feature confusion. Therefore, existing nasopharyngeal image recognition methods suffer from technical problems: a single model cannot adequately address the differentiated needs of glottic functional state discrimination and non-glottic anatomical location subdivision, resulting in insufficient classification accuracy and limited generalization ability. Summary of the Invention

[0005] This application provides a method, device, equipment, and medium for recognizing nasal and pharyngeal image structures, in order to solve the problems of feature confusion and class imbalance caused by conflicting task objectives in traditional single-stage models, and improve the accuracy of vocal cord state discrimination and sub-site localization as well as the model's generalization ability.

[0006] In a first aspect, embodiments of this application provide a method for recognizing the structure of a nasal cavity and pharynx image, the method comprising:

[0007] The nasal cavity and laryngoscope image to be identified is input into a pre-trained first structure recognition model to determine the main structure category to which the nasal cavity and laryngoscope image to be identified belongs; wherein, the main structure category includes the glottic region category and at least one non-glottic region category;

[0008] If the main structural category is the laryngeal glottic region category, then based on the central region features of the nasopharyngeal endoscope image to be identified, the vocal cord state subtype identification result of the nasopharyngeal endoscope image to be identified is determined.

[0009] If the main structural category is at least one non-glottic region category, the nasal cavity laryngoscope image to be identified is input into a pre-trained second structural recognition model to determine the anatomical location subtype recognition result under the main structural category to which the nasal cavity laryngoscope image to be identified belongs.

[0010] Secondly, embodiments of this application also provide a nasal cavity and pharyngeal image structure recognition device, the device comprising:

[0011] The main structure category recognition module is used to input the nasal cavity laryngoscope image to be recognized into a pre-trained first structure recognition model to determine the main structure category to which the nasal cavity laryngoscope image to be recognized belongs; wherein, the main structure category includes the laryngeal glottic region category and at least one non-glottic region category;

[0012] The first subtype structure category identification module is used to determine the vocal cord state subtype identification result of the nasal cavity laryngoscope image to be identified based on the central region features of the nasal cavity laryngoscope image to be identified if the main structure category is the laryngeal glottic region category.

[0013] The second subtype structure category recognition module is used to input the nasal cavity laryngoscope image to be identified into a pre-trained second structure recognition model if the main structure category is at least one non-glottic region category, and to determine the anatomical location subtype recognition result under the main structure category to which the nasal cavity laryngoscope image to be identified belongs.

[0014] Thirdly, embodiments of this application also provide an electronic device, which includes:

[0015] One or more processors;

[0016] Storage device for storing one or more programs.

[0017] When one or more programs are executed by one or more processors, the one or more processors implement the nasal cavity and pharyngeal image structure recognition method as described in any of the embodiments of this application.

[0018] Fourthly, embodiments of this application also provide a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a nasal cavity and pharyngeal image structure recognition method as described in any of the embodiments of this application.

[0019] This application provides a method, apparatus, device, and medium for nasal and laryngeal image structure recognition. The method involves inputting a nasal and laryngeal image to be recognized into a pre-trained first structure recognition model to determine the main structural category to which the image belongs. The main structural category includes a laryngeal glottic region category and at least one non-glottic region category. If the main structural category is the laryngeal glottic region category, the vocal cord state subtype of the image is determined based on the central region features of the image. If the main structural category is at least one non-glottic region category, the image is input into a pre-trained second structure recognition model to determine the anatomical location subtype under the main structural category to which the image belongs. The technical solution of this application introduces a hierarchical recognition strategy. First, it uses a first structural recognition model to distinguish the macroscopic categories of the glottic region and non-glottic region of the larynx. Then, it accurately analyzes the dynamic functional state of the vocal cords based on the central region features for the glottic region. For the non-glottic region, it refines the anatomical location through a second structural recognition model. This effectively solves the problems of feature confusion and category imbalance caused by conflicting task objectives in traditional single-stage models, and significantly improves the accuracy of vocal cord state discrimination and sub-site localization, as well as the model's generalization ability. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the exemplary embodiments of this application, the accompanying drawings used in describing the embodiments are briefly introduced below. Obviously, the accompanying drawings described are only a portion of the embodiments to be described in this application, and not all of them. For those skilled in the art, other drawings can be obtained from these drawings without any creative effort.

[0021] Figure 1 A flowchart illustrating a method for recognizing nasal and pharyngeal image structures provided in an embodiment of this application;

[0022] Figure 2 This is a schematic diagram of the anatomical structure of the vocal cords in the open and closed states as described in the embodiments of this application;

[0023] Figure 3 A schematic flowchart illustrating another method for recognizing nasal and pharyngeal image structures provided in this application embodiment;

[0024] Figure 4 This is a schematic diagram of the structure of a nasal cavity and pharyngeal image structure recognition device provided in an embodiment of this application;

[0025] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0026] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the application and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present application, not the entire structure.

[0027] Example 1

[0028] Figure 1 This is a flowchart illustrating a method for recognizing the structure of a nasal cavity and pharynx image provided in an embodiment of this application. This embodiment is applicable to any situation where automated structural recognition and functional classification of the nasal cavity and pharynx based on endoscopic images are required. This method can be executed by a nasal cavity and pharynx image structure recognition device, which can be implemented in the form of software and / or hardware. The hardware can be an electronic device, such as a mobile terminal, a PC, or a server.

[0029] like Figure 1 As shown, the method includes:

[0030] S110. Input the nasal cavity laryngoscope image to be identified into the pre-trained first structure recognition model to determine the main structure category to which the nasal cavity laryngoscope image to be identified belongs.

[0031] Among them, the nasopharyngeal images to be identified refer to digital images of the nasal cavity, pharynx or laryngeal region acquired by endoscopic equipment and that need to be analyzed and classified by image recognition methods. The content usually includes the anatomical structure and morphological features of the above-mentioned areas, and is the original input data for subsequent models to distinguish structures and determine states.

[0032] The first structural recognition model refers to a pre-trained classification model or algorithm module capable of comprehensively analyzing input nasopharyngoscopy images and outputting their main anatomical structure categories (specifically including the glottic and non-glottic regions of the larynx). Optionally, in this embodiment, the first structural recognition model can be a coupled scaling network based on structural region localization (referred to as the Efficientnetb0 network). This network simultaneously adjusts the network's depth, width, and resolution through composite coefficients to adapt to the identification of structural images in nasopharyngoscopy images at different resolutions, distances, and lighting conditions. Furthermore, it incorporates inherent human structural examination logic to further optimize accuracy.

[0033] The primary structural category refers to the macroscopic category required for high-level anatomical partitioning of nasopharyngeal endoscopy images. Its core function is to distinguish whether the image content includes the specific functional region of the laryngeal glottis, thus providing a basis for subsequent detailed identification. The primary structural category includes the laryngeal glottis category and at least one non-glottis category. The laryngeal glottis category refers to the anatomical region category determined by the image recognition model to which the image content belongs to the key structures of the human larynx's core vocal and respiratory channels (mainly the vocal cords and their adjacent tissues). This category is used to indicate the need for subsequent specific analysis of the vocal cord morphology and functional status. At least one non-glottis category refers to the category determined by the image recognition model to which the image content does not contain the laryngeal glottis structure but belongs to other upper respiratory tract anatomical regions within the nasal cavity to hypopharynx range. This category is used to indicate the need for further anatomical location subdivision identification based on specific sites. Specifically, at least one non-glottis category can include nasal cavity structure category, nasopharyngeal structure category, hypopharyngeal structure category, or pharyngeal structure category.

[0034] In this embodiment, the nasal cavity and pharyngeal endoscopy images to be identified, which have not yet been manually interpreted and originate from nasal cavity or pharyngeal endoscopy examinations, can be sent in a predetermined format to the first structure recognition model, which has completed parameter learning and verification through a large number of labeled samples. The first structure recognition model uses its internal multi-layer feature extraction and classification mechanism to calculate the pixel distribution, texture, spatial relationship and semantic information of the overall nasal cavity and pharyngeal endoscopy images to be identified, and finally outputs the category label with the highest probability, thereby clearly classifying the image into one of the two mutually exclusive and all possible main structure categories: "laryngeal glottic region category" or "at least one non-glottic region category".

[0035] For example, a newly acquired nasopharyngoscopy image is directly fed into the first structure recognition model, which has been trained with tens of thousands of labeled images. The first structure recognition model then outputs a probability table: laryngeal glottis 0.87, nasal cavity structure 0.03, nasopharyngeal structure 0.05, hypopharyngeal structure 0.04, and pharyngeal structure 0.01. The image is then labeled as the laryngeal glottis category, thus completing the automatic determination of its main structural categories.

[0036] S120. If the main structural category is the laryngeal glottic region category, then based on the central region features of the nasopharyngeal endoscope image to be identified, determine the vocal cord state subtype identification result to which the nasopharyngeal endoscope image to be identified belongs.

[0037] The central region features refer to the set of pixel statistics and morphological quantification attributes extracted from the nasopharyngoscopy image to be identified, located in the central region of the image. For example, the central region features may specifically include brightness and contrast information represented by the mean and standard deviation of pixel values ​​in the region, as well as the shape compactness index defined by the ratio of perimeter to area obtained through contour calculation, which are used together to describe the morphology and functional state of the vocal cord region.

[0038] Among them, the vocal cord state subtype identification result refers to the classification and judgment conclusion of the current physiological function state of the vocal cords obtained based on the feature analysis of the central region of the image. Specifically, it is manifested as the functional state category of the dynamic form of the vocal cords, such as open or closed, mapped by quantitative features (such as pixel mean, standard deviation and shape compactness).

[0039] See the schematic diagrams of the anatomical structure of the vocal cords in open and closed states. Figure 2 ,like Figure 2 As shown, Figure 2 The left side shows a diagram of the vocal cords in an open state, with the central area of ​​the airway appearing darker. Figure 2 The right side shows a schematic diagram of the vocal cords, which appear as a relatively uniform, grayish-white tissue. It is evident that there is a significant difference in the central region between the two states: in the open state, the vocal cords are separated, exposing the airway, and the central region exhibits a dark, low-grayscale characteristic; in the closed state, the vocal cords are joined together to form continuous tissue, and the central region exhibits a light-colored, high-grayscale characteristic with uniform texture. This difference allows for the determination of the vocal cord state subtype of the nasopharyngoscopy image to be identified based on the central region features (such as pixel mean, dispersion, and morphological compactness).

[0040] In this embodiment, when the first structural recognition model determines that the nasopharyngoscope image to be identified belongs to the glottic region of the larynx, the pixel mean, pixel standard deviation, perimeter and area can be extracted from the central preset rectangular area of ​​the image, and the compactness is calculated by the ratio of area to the square of perimeter. Thus, these three types of central region features constitute a low-dimensional vector input to the vocal cord state discriminator that has been trained offline. The discriminator divides the vector space by internal threshold or classification boundary and directly outputs "open state" or "closed state" as the unique recognition result of the image in the vocal cord state subtype.

[0041] S130. If the main structural category is at least one non-glottic region category, the nasal cavity laryngoscope image to be identified is input into the pre-trained second structural recognition model to determine the anatomical location subtype recognition result under the main structural category to which the nasal cavity laryngoscope image to be identified belongs.

[0042] The second structure recognition model refers to a pre-trained classification model or algorithm module specifically designed for fine-grained anatomical location analysis of nasopharyngoscopy images that have been determined to belong to the non-glottic region category. Its function is to further distinguish and output the specific subtype category of the image within the non-glottic region by extracting image features. Optionally, the second structure recognition model can be a ResNet50 image location classification neural network model, using the deep semantic features extracted by this network model to distinguish sub-level structures that differ in left-right or distance.

[0043] Among them, the anatomical location subtype identification result refers to the specific anatomical location classification conclusion output after fine-grained analysis of non-glottic region images through the second structure identification model. Its function is to further distinguish the specific sub-region location information under macro categories such as nasal cavity, nasopharynx, hypopharynx or pharynx.

[0044] Specifically, when the first structure recognition model determines that the main structure category of the nasopharyngoscope image to be recognized is any non-glottic region category, the entire image can be automatically fed into the second structure recognition model, which has been trained and converged with a large number of finely labeled samples of the corresponding category. Based on the local texture, edge morphology, spatial distribution and depth semantic features of the image, the model performs forward reasoning within it and directly outputs the unique and most reliable subtype classification result according to the pre-set mutually exclusive anatomical location subtype label set of the main structure category, thereby completing a more refined anatomical location subtype determination within the non-glottic region category.

[0045] In this embodiment, at least one non-glottic region category includes: nasal cavity structure category, nasopharyngeal structure category, hypopharyngeal structure category, or pharyngeal structure category. Optionally, the anatomical location subtype identification results under the nasal cavity structure category include: left nasal cavity structure category or right nasal cavity structure category; the anatomical location subtype identification results under the nasopharyngeal structure category include: left nasopharyngeal structure category or right nasopharyngeal structure category; the anatomical location subtype identification results under the hypopharyngeal structure category include: left pyriform fossa structure category, right pyriform fossa structure category, or midline hypopharyngeal structure category; the anatomical location subtype identification results under the pharyngeal structure category include: oropharynx and proximal and distal hypopharynx or root of tongue and vallecula. In other words, when an image is classified as a nasal cavity structure, the model can only further classify it as either a left or right nasal cavity structure; when an image is classified as a nasopharyngeal structure, it can only further classify it as either a left or right nasopharyngeal structure; and when an image is classified as a pharyngeal structure, it is necessary to further determine a unique subtype from the two mutually exclusive options of the proximal and distal oropharyngeal regions and the root of the tongue and the valley of the epiglottis, thereby completing the precise localization of the non-glottic region image in terms of horizontal left and right, vertical depth and specific anatomical regions.

[0046] For example, when the second structure recognition model determines that the image belongs to a non-glottic region category (such as nasal cavity, nasopharynx, etc.), the second structure recognition model further analyzes the nasal cavity laryngoscope image to be identified and outputs its specific left or right side or partition location result under the macro category. For example, if the first model outputs "nasal cavity structure category", the second model can further identify whether the image belongs to "left nasal cavity" or "right nasal cavity".

[0047] This application provides a method, apparatus, device, and medium for nasal and laryngeal image structure recognition. The method involves inputting a nasal and laryngeal image to be recognized into a pre-trained first structure recognition model to determine the main structural category to which the image belongs. The main structural category includes a laryngeal glottic region category and at least one non-glottic region category. If the main structural category is the laryngeal glottic region category, the vocal cord state subtype of the image is determined based on the central region features of the image. If the main structural category is at least one non-glottic region category, the image is input into a pre-trained second structure recognition model to determine the anatomical location subtype under the main structural category to which the image belongs. The technical solution of this application introduces a hierarchical recognition strategy. First, it uses a first structural recognition model to distinguish the macroscopic categories of the glottic region and non-glottic region of the larynx. Then, it accurately analyzes the dynamic functional state of the vocal cords based on the central region features for the glottic region. For the non-glottic region, it refines the anatomical location through a second structural recognition model. This effectively solves the problems of feature confusion and category imbalance caused by conflicting task objectives in traditional single-stage models, and significantly improves the accuracy of vocal cord state discrimination and sub-site localization, as well as the model's generalization ability.

[0048] Example 2

[0049] Figure 3 This is a schematic diagram of a nasal cavity and pharyngeal image structure recognition method provided in an embodiment of this application. Based on the foregoing embodiments, S120 is described in detail, and the specific implementation method can be found in the technical solution of this embodiment. Technical terms that are the same as or corresponding to those in the above embodiments will not be repeated here.

[0050] like Figure 3 As shown, the method specifically includes the following steps:

[0051] S210. Input the nasal cavity laryngoscope image to be identified into the pre-trained first structure recognition model to determine the main structure category to which the nasal cavity laryngoscope image to be identified belongs.

[0052] The main structural categories include the laryngeal glottic region category and at least one non-glottic region category.

[0053] S220. Perform edge detection on the nasopharyngoscope image to be identified, and identify and locate the central region of the image.

[0054] The image center region refers to the focused analysis area located in the nasal cavity and pharyngoscopy image to be identified by the edge detection model, with the suspected location of the vocal cords as the core. For example, when the endoscopic image contains the glottis of the larynx, this region usually corresponds to the set of pixels within the contour range of the vocal cord vibration part in the middle of the image. Its range is automatically defined by the edge detection algorithm and is used to extract pixel statistical features to determine the state of the vocal cords.

[0055] In this embodiment, after performing Canny edge detection on the nasal cavity laryngoscope image to be identified, a binary edge map is obtained. Then, Hough circle transform is used to detect an approximately circular region with a radius of about 180 pixels and a center coordinate of (256, 256) in the edge map. This circle is then locked by the system as the "image center region".

[0056] Optionally, the specific implementation methods for edge detection, identification, and localization of the central region of the nasopharyngoscope image to be identified may include:

[0057] S1. Input the nasal cavity laryngoscope image to be identified into the edge detection model and extract the edge contour points of the central region of the image.

[0058] Among them, the edge detection model is a learnable function that is pre-trained in a supervised or self-supervised manner. It takes nasal laryngoscope images as input and uses an encoder-decoder network structure composed of convolution, nonlinear activation and upsampling to estimate the probability that each position belongs to the anatomical edge in the pixel-level output space. Then, high-probability positions are marked as edge pixels, thereby generating continuous, closed edge contours that are aligned with the real anatomical boundaries for subsequent central region localization and feature extraction.

[0059] Among them, edge contour points refer to the set of points obtained by arranging all the pixel positions determined as "edges" in the binary edge map output by the edge detection model after edge detection is completed in topological order. They form one or more closed or open discrete curves in the form of two-dimensional image coordinates, which are used to accurately describe the boundary shape of the target anatomical structure and serve as the spatial basis for subsequent sparse homogenization processing and central region extraction.

[0060] In this embodiment, the entire nasal cavity and laryngoscope image to be identified can be fed into a pre-trained edge detection model. The model predicts and marks all high-confidence pixels belonging to the anatomical boundary at the pixel level through internal convolution, nonlinear activation and upsampling operations. Then, the positions of these pixels in the central region of the image are extracted in topological order to form a set of edge contour points, which are used for subsequent sparse homogenization and central region localization.

[0061] S2. Traverse each edge contour point in sequence, discard edge contour points whose distance between two adjacent points is less than a preset number of pixels, and supplement edge contour points whose distance between two adjacent points exceeds a preset number of pixels by interpolation points with a preset number of pixels as the interval, to obtain a sparse and uniform contour point sequence.

[0062] In this embodiment, the Euclidean distance between each pair of adjacent points can be examined one by one in a predetermined order along the extracted edge contour point sequence. If the distance is less than a preset pixel threshold, the two points are considered too dense and the latter point is removed directly to achieve redundancy suppression. If the distance is greater than the same threshold, several new points are inserted between the two points at equal intervals according to the threshold length, so that the spacing is uniformly aligned to the preset value, thereby eliminating both over-concentration and over-sparseness, and finally obtaining a sparse and uniform contour point sequence with constant point spacing and moderate density for subsequent region definition and feature calculation.

[0063] S3. Determine the image center region as the image area enclosed by the contour point sequence.

[0064] In this embodiment, after obtaining the sparse and uniform contour point sequence, all points in the sequence are connected end to end in the order of their positions in the image coordinate system to form a closed boundary. Then, the positions of all pixels inside the boundary are identified and marked by polygon filling or scan line algorithms. Thus, this connected region strictly enclosed by contour points is defined as the "image center region" required for subsequent calculations. Pixels outside this region no longer participate in the feature extraction of the center region.

[0065] S230. Based on each pixel within the central region of the image, determine the central region features of the nasopharyngoscope image to be identified.

[0066] In this embodiment, statistical calculations can be performed on all pixel grayscale values ​​within the locked image center region: first, the arithmetic mean is calculated to obtain the pixel mean, and then the root mean square of the sum of squares of the differences between each pixel and the mean is calculated to obtain the pixel standard deviation; at the same time, the closed perimeter and pixel area of ​​the center region are obtained from the sparsely homogenized contour point sequence according to the planar geometric relationship, and a compactness factor is constructed by the ratio of area to the square of perimeter; thus, at least one of the mean, standard deviation, and compactness factor is used as the central region feature of the nasopharyngoscope image to be identified.

[0067] Optionally, the specific steps for determining the features of the central region of the nasopharyngoscope image to be identified based on the grayscale values ​​of each pixel within the central region of the image may include:

[0068] S1. Calculate the mean value of the gray values ​​of all pixels in the central region to obtain the first feature parameter.

[0069] In this embodiment, the grayscale values ​​of each pixel can be read sequentially in the raster scanning order within the determined image center area. The values ​​are then summed and divided by the total number of pixels to obtain the arithmetic mean of the grayscale distribution within the area. This mean is then directly defined as the first feature parameter to quantify the overall brightness level of the center area.

[0070] Specifically, the formula for determining the first characteristic parameter can be expressed as:

[0071]

[0072] In the formula, Represents the first characteristic parameter. This represents the total number of pixels within the central region. This represents the grayscale value of the p-th pixel.

[0073] S2. Calculate the standard deviation of the gray values ​​of all pixels in the central region to obtain the second feature parameter.

[0074] In this embodiment, the gray values ​​of all pixels in the central region can be used as samples. First, their arithmetic mean is calculated. Then, the sum of the squares of the differences between each gray value and the mean is calculated. The square root of the sum is divided by the total number of pixels. The result is the standard deviation of the gray value distribution in the central region. This standard deviation is used as the second feature parameter to quantify the degree of gray value dispersion in the region.

[0075] Specifically, the formula for determining the second characteristic parameter can be expressed as:

[0076]

[0077] In the formula, Indicates the second characteristic parameter. This represents the total number of pixels within the central region. This represents the grayscale value of the p-th pixel. This represents the average grayscale value of all pixels within the central region.

[0078] S3. Calculate the perimeter and area of ​​the central region's contour information, and perform a compactness factor operation based on the ratio of perimeter to area to obtain the third feature parameter.

[0079] In this embodiment, the Euclidean distance between adjacent points can be calculated sequentially along the sparsely uniform contour point sequence and accumulated to obtain the closed perimeter of the central region; then the pixel area of ​​the region enclosed by the contour can be calculated according to the planar polygon area formula; then the ratio of the area to the square of the perimeter is used as the compactness factor, the value of which is between 0 and 1 / (4π). The closer it is to the upper limit, the closer the shape is to a circle and the more compact it is, and vice versa, the narrower or more irregular it is; this factor is directly defined as the third feature parameter to quantify the geometric compactness of the central region.

[0080] Specifically, the formula for determining the third characteristic parameter can be expressed as:

[0081]

[0082] In the formula, Represents the third characteristic parameter. Indicates the area of ​​the central region. This indicates the perimeter of the central area.

[0083] S4. Use at least one of the first feature parameter, the second feature parameter, and the third feature parameter as the central region feature of the nasopharyngoscope image to be identified.

[0084] In this embodiment, after the independent calculation of the first, second, and third feature parameters is completed, one, two, or all three values ​​can be selected from these three values ​​according to the preset strategy or subsequent model requirements, and directly combined or weighted to form the final "central region feature" vector for vocal cord state subtype identification; in other words, the central region feature only needs to include at least one of these three, without the need to force the adoption of all of them, thereby ensuring the sufficiency of information while retaining the flexibility of feature selection.

[0085] S240. Based on the central region features and vocal cord state determination mapping conditions, determine the vocal cord state subtype identification result of the nasopharyngoscope image to be identified.

[0086] Among them, the vocal cord state determination mapping condition refers to a set of pre-set mathematical or logical rules that establish a correspondence between central region features (such as pixel mean, pixel standard deviation and compactness factor and other quantitative indicators) and specific vocal cord functional states (such as open or closed). Its essence is to map numerical features into clinically meaningful vocal cord dynamic morphology classification results through threshold judgment or classifier decision.

[0087] In this embodiment, the extracted numerical features of the central region (such as pixel mean, standard deviation, and compactness factor) can be matched and calculated with the vocal cord state determination mapping conditions to output an automated decision-making process for the specific functional state category (such as open or closed) of the vocal cords in the image.

[0088] Specifically, optional vocal cord state determination mapping conditions include: pixel grayscale mean determination mapping conditions, pixel dispersion determination mapping conditions, and compactness determination mapping conditions. Based on these, the specific implementation method for determining the vocal cord state subtype of the nasopharyngoscopy image to be identified, based on central region features and preset vocal cord state determination conditions, may include:

[0089] (1) Based on the pixel grayscale mean determination mapping condition, the first feature parameter is subjected to vocal cord state subtype determination mapping to determine the first confidence level determination result.

[0090] The pixel gray-scale mean determination mapping condition refers to a predefined set of rules used to establish a mathematical relationship between the average pixel gray-scale value in the central region of an image and the confidence level of the vocal cord state. Specifically, it positively maps the gray-scale mean value to the typical pixel values ​​of preset vocal cord closure and open states, respectively, into a quantitative index of the confidence level of the corresponding state determination result. The pixel gray-scale mean determination mapping condition includes: the closeness of the first feature parameter value to the first preset pixel value is positively correlated with the first confidence level of the recognition result of the vocal cord closure state subtype; the closeness of the first feature parameter value to the second preset pixel value is positively correlated with the second confidence level of the recognition result of the vocal cord open state subtype. In other words, the closer the average gray value of the pixels in the central region of the image (the first feature parameter) is to the typical gray value of the vocal cords in a closed state (the first preset pixel value, for example, 211 grayish-white), the higher the confidence level of determining that it is in a closed state; at the same time, the closer the average value is to the typical gray value of the vocal cords in an open state (the second preset pixel value, for example, 0 black), the higher the confidence level of determining that it is in an open state. The two are mapped to the corresponding state confidence quantification results through a positive correlation function.

[0091] The first confidence level determination result refers to the quantitative value of the confidence level of the image belonging to the vocal cord closure and open state after analyzing and calculating the average pixel gray value of the image center region based on the pixel gray value determination mapping condition. This value is generated by mapping the proximity of the feature parameters to the preset typical pixel values ​​through a positive correlation function, and is used for subsequent comprehensive determination of the vocal cord state subtype.

[0092] Specifically, the first feature parameter (average gray value of pixels in the central region) is input into a preset pixel gray value determination mapping condition. By calculating the proximity between this average value and a first preset pixel value representing the vocal cord closure state, and the proximity between this average value and a second preset pixel value representing the vocal cord open state, the proximity is quantified into a normalized first confidence level and a second confidence level. The first confidence level is positively correlated with the credibility of the closure state determination, and the second confidence level is positively correlated with the credibility of the open state determination. Thus, the first confidence level and the second confidence level are the first confidence level determination result.

[0093] For example, let the first preset pixel value (representing vocal cord closure) be 211 and the second preset pixel value (representing vocal cord openness) be 0. Now, the average grayscale value of the pixels in the central region of an image is 150, the absolute difference between it and 221 is 65, and the absolute difference between it and 0 is 211. The system uses linear normalization to map the difference to the 0-1 interval, obtaining a first confidence level of 0.94 (close to 211) and a second confidence level of 0.06 (far from 0). Therefore, the image outputs a confidence level of 0.94 for the closed state and a confidence level of 0.06 for the open state based only on the average grayscale value, as the first confidence level determination result.

[0094] (2) Based on the pixel discreteness determination mapping condition, the second feature parameter is used to determine the vocal cord state subtype and the second confidence determination result is determined.

[0095] The pixel dispersion determination mapping condition refers to a predefined set of rules used to establish a mathematical relationship between the dispersion of pixel values ​​in the central region of the image (the second feature parameter) and the confidence level of the vocal cord state. Specifically, it achieves a quantitative mapping from feature values ​​to state determination confidence levels through functional relationships where the magnitude of the dispersion value is positively correlated with the confidence level of the vocal cord closure state and negatively correlated with the confidence level of the vocal cord open state. The pixel dispersion determination mapping condition can include: the magnitude of the second feature parameter value is positively correlated with the first confidence level of the identification result of the vocal cord closure state subtype; the magnitude of the second feature parameter value is negatively correlated with the second confidence level of the identification result of the vocal cord open state subtype. That is, the larger the dispersion of pixel values ​​in the central region of the image (the second feature parameter), the higher the confidence level of the identification of the vocal cord closure state; at the same time, the smaller the dispersion value, the higher the confidence level of the identification of the vocal cord open state. The two are mapped to the corresponding state confidence quantification results through positive and negative correlation functions, respectively.

[0096] Specifically, the second feature parameter can be used to input the pixel dispersion determination mapping condition. According to the predefined function, the standard deviation value is mapped to a pair of confidence scores: when the standard deviation is larger, the first confidence score of the vocal cord closure state is given a higher weight, while the second confidence score of the vocal cord open state is correspondingly suppressed; conversely, the smaller the standard deviation, the first confidence score decreases and the second confidence score increases. The mapping function ensures that the standard deviation is monotonically positively correlated with the first confidence score and monotonically negatively correlated with the second confidence score. Finally, the second confidence score determination result, which is only related to the dispersion, is output for subsequent comprehensive decision-making.

[0097] For example, let the second feature parameter be the standard deviation of gray level in the central region σ, and the preset judgment functions be softmax(α·σ) and softmax(−α·σ), where α is a positive coefficient; when σ is 18, the function outputs a closure confidence level of 0.85 and an open confidence level of 0.15, which is the second confidence level judgment result.

[0098] (3) Based on the compactness determination mapping condition, the vocal cord state subtype is determined for the third feature parameter, and the third confidence determination result is determined.

[0099] The compactness determination mapping condition refers to a predefined set of rules used to establish a mathematical relationship between the quantitative index of shape compactness of the central region of an image (the third feature parameter) and the confidence level of vocal cord state. Specifically, it achieves a quantitative mapping from morphological features to state determination confidence level by using functional relationships where the compactness value is positively correlated with the confidence level of vocal cord closure and negatively correlated with the confidence level of vocal cord openness. The compactness determination mapping condition includes: the value of the third feature parameter is positively correlated with the first confidence level of the identification result of the vocal cord closure subtype; the value of the third feature parameter is negatively correlated with the second confidence level of the identification result of the vocal cord openness subtype. That is, when the value of the shape compactness index (third feature parameter) of the central region of the image is larger, the confidence level of the identification of the vocal cord closure state is higher; while when the value is smaller, the confidence level of the identification of the vocal cord open state is higher. The two achieve a precise mapping from morphological features to state confidence level through positive and negative correlation functions, respectively.

[0100] In this embodiment, the third feature parameter can be input into the compactness determination mapping condition. A preset monotonic mapping function is used to convert the compactness value into a pair of complementary confidence levels: when the compactness is higher, the function assigns a monotonically increasing first confidence level to the vocal cord closure subtype, while the second confidence level to the vocal cord open subtype monotonically decreases; conversely, the lower the compactness, the lower the first confidence level and the higher the second confidence level. The mapping function ensures that the compactness value is strictly positively correlated with the closure confidence level and strictly negatively correlated with the open confidence level, and finally outputs a third confidence level determination result that is only related to compactness for subsequent fusion decision-making.

[0101] For example, let the third feature parameter be the compactness factor C. The preset mapping function adopts a piecewise linear approach: when C ≥ 0.08, the output closure confidence is 0.92 and the open confidence is 0.08; when C drops to 0.05, the closure confidence linearly drops to 0.60 and the open confidence rises to 0.40; when C continues to drop to 0.02, the closure confidence is only 0.10 and the open confidence rises to 0.90. Therefore, if an image is calculated to have C = 0.075, the output third confidence judgment result is 0.88 for the closed state and 0.12 for the open state.

[0102] (4) Based on the first confidence level judgment result, the second confidence level judgment result and / or the third confidence level judgment result, determine the vocal cord state subtype recognition result of the nasal laryngoscopy image to be identified.

[0103] In this embodiment, the closed and open confidence scores obtained from the three independent mapping channels of pixel grayscale mean, pixel dispersion, and compactness are weighted and summed or logically compared according to a preset fusion rule, thereby integrating the local confidence scores of the three channels into an overall confidence score vector. The vocal cord state subtype corresponding to the higher one is used as the final judgment result output. In the whole process, only a single channel can be adopted, or the confidence scores of all or any two channels can be used for comprehensive decision-making.

[0104] S250. If the main structural category is at least one non-glottic region category, the nasal cavity laryngoscope image to be identified is input into the pre-trained second structural recognition model to determine the anatomical location subtype recognition result under the main structural category to which the nasal cavity laryngoscope image to be identified belongs.

[0105] The technical solution of this application, when determining the vocal cord state subtype identification result of the nasopharyngoscope image to be identified, firstly, performs edge detection on the nasopharyngoscope image to be identified, identifies and locates the central region of the image, and then, based on each pixel point in the central region of the image, determines the central region features of the nasopharyngoscope image to be identified. Therefore, based on the central region features and the vocal cord state determination mapping conditions, the vocal cord state subtype identification result of the nasopharyngoscope image to be identified is determined. This technical solution, through edge detection, accurately locates the core region of the vocal cords and extracts quantitative features, and combines multi-dimensional determination mapping conditions to generate confidence assessments. This effectively avoids subjective errors and interference from irrelevant regions, significantly improving the objectivity, repeatability, and automation of vocal cord functional state classification, providing standardized decision support for clinical practice.

[0106] Example 3

[0107] Figure 4This is a schematic diagram of a nasal cavity and pharyngeal image structure recognition device provided in an embodiment of this application. The device includes: a main structure category recognition module 310, a first subtype structure category recognition module 320, and a second subtype structure category recognition module 330.

[0108] The main structure category recognition module 310 is used to input the nasal cavity laryngoscope image to be recognized into a pre-trained first structure recognition model to determine the main structure category to which the nasal cavity laryngoscope image to be recognized belongs; wherein, the main structure category includes a laryngeal glottic region category and at least one non-glottic region category;

[0109] The first subtype structure category identification module 320 is used to determine the vocal cord state subtype identification result of the nasal cavity laryngoscope image to be identified based on the central region features of the nasal cavity laryngoscope image to be identified if the main structure category is the laryngeal glottic region category.

[0110] The second subtype structure category recognition module 330 is used to input the nasal cavity laryngoscope image to be identified into a pre-trained second structure recognition model if the main structure category is at least one non-glottic region category, and to determine the anatomical location subtype recognition result under the main structure category to which the nasal cavity laryngoscope image to be identified belongs.

[0111] This application provides a method, apparatus, device, and medium for nasal and laryngeal image structure recognition. The method involves inputting a nasal and laryngeal image to be recognized into a pre-trained first structure recognition model to determine the main structural category to which the image belongs. The main structural category includes a laryngeal glottic region category and at least one non-glottic region category. If the main structural category is the laryngeal glottic region category, the vocal cord state subtype of the image is determined based on the central region features of the image. If the main structural category is at least one non-glottic region category, the image is input into a pre-trained second structure recognition model to determine the anatomical location subtype under the main structural category to which the image belongs. The technical solution of this application introduces a hierarchical recognition strategy. First, it uses a first structural recognition model to distinguish the macroscopic categories of the glottic region and non-glottic region of the larynx. Then, it accurately analyzes the dynamic functional state of the vocal cords based on the central region features for the glottic region. For the non-glottic region, it refines the anatomical location through a second structural recognition model. This effectively solves the problems of feature confusion and category imbalance caused by conflicting task objectives in traditional single-stage models, and significantly improves the accuracy of vocal cord state discrimination and sub-site localization, as well as the model's generalization ability.

[0112] Based on the above-mentioned device, optionally, the first subtype structure category identification module includes:

[0113] The central region recognition unit is used to perform edge detection on the nasopharyngoscope image to be recognized, and to identify and locate the central region of the image.

[0114] The region feature determination unit is used to determine the central region features of the nasopharyngoscope image to be identified based on each pixel point within the central region of the image.

[0115] The vocal cord state recognition unit is used to determine the vocal cord state subtype recognition result of the nasopharyngoscopy image to be recognized based on the central region features and vocal cord state determination mapping conditions.

[0116] Based on the above-mentioned device, optionally, a central region recognition unit is used to input the nasopharyngoscope image to be recognized into an edge detection model, extract edge contour points of the central region of the image; sequentially traverse each edge contour point, discard edge contour points whose distance between two adjacent points is less than a preset number of pixels, and supplement edge contour points whose distance between two adjacent points exceeds a preset number of pixels by interpolation points with a preset number of pixels as the interval, to obtain a sparse and uniform contour point sequence; and determine the image region enclosed by the contour point sequence as the image central region.

[0117] Based on the above-mentioned device, optionally, a region feature determination unit is used to perform mean calculation on the gray values ​​of all pixels in the central region to obtain a first feature parameter; perform standard deviation calculation on the gray values ​​of all pixels in the central region to obtain a second feature parameter; calculate the perimeter and area of ​​the contour information of the central region, and perform compactness factor calculation based on the ratio of perimeter to area to obtain a third feature parameter; and use at least one of the first feature parameter, the second feature parameter, and the third feature parameter as the central region feature of the nasopharyngoscope image to be identified.

[0118] Based on the above-mentioned device, optionally, the vocal cord state determination mapping conditions include: pixel grayscale mean determination mapping conditions, pixel dispersion determination mapping conditions, and compactness determination mapping conditions. The vocal cord state recognition unit is specifically used to perform vocal cord state subtype determination mapping on the first feature parameter based on the pixel grayscale mean determination mapping conditions to determine a first confidence level determination result; wherein, the pixel grayscale mean determination mapping conditions include: the closeness between the value of the first feature parameter and the value of a first preset pixel is positively correlated with the first confidence level of the vocal cord closure state subtype recognition result; the closeness between the value of the first feature parameter and the value of a second preset pixel is positively correlated with the second confidence level of the vocal cord open state subtype recognition result; and the second feature parameter is determined based on the pixel dispersion determination mapping conditions to determine a second confidence level determination result; wherein, the pixel The dispersion determination mapping conditions include: the magnitude of the second feature parameter value is positively correlated with the first confidence level of the vocal cord closure state subtype identification result; the magnitude of the second feature parameter value is negatively correlated with the second confidence level of the vocal cord open state subtype identification result; based on the compactness determination mapping conditions, the third feature parameter is used to determine the vocal cord state subtype, and the third confidence level determination result is determined; wherein, the compactness determination mapping conditions include: the magnitude of the third feature parameter value is positively correlated with the first confidence level of the vocal cord closure state subtype identification result; the magnitude of the third feature parameter value is negatively correlated with the second confidence level of the vocal cord open state subtype identification result; based on the first confidence level determination result, the second confidence level determination result, and / or the third confidence level determination result, the vocal cord state subtype identification result of the nasopharyngoscopy image to be identified is determined.

[0119] Based on the above-described device, optionally, the at least one non-glottic region category includes: nasal cavity structure category, nasopharyngeal structure category, hypopharyngeal structure category, and / or pharyngeal structure category.

[0120] Based on the above-mentioned device, optionally, the anatomical location subtype identification results under the nasal cavity structure category include: left nasal cavity structure category or right nasal cavity structure category; the anatomical location subtype identification results under the nasopharyngeal structure category include: left nasopharyngeal structure category or right nasopharyngeal structure category; the anatomical location subtype identification results under the hypopharyngeal structure category include: left piriform fossa structure category, right piriform fossa structure category or midline hypopharyngeal structure category; the anatomical location subtype identification results under the pharyngeal structure category include: proximal and distal hypopharynx or root of tongue and vallecula.

[0121] The nasal cavity and pharyngeal image structure recognition device provided in this application embodiment can execute the nasal cavity and pharyngeal image structure recognition method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects of executing the method.

[0122] It is worth noting that the various units and modules included in the above system are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy differentiation and are not used to limit the protection scope of the embodiments of this application.

[0123] Example 4

[0124] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 5 A block diagram is shown of an exemplary electronic device 40 suitable for implementing embodiments of the present application. Figure 5 The electronic device 40 shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0125] like Figure 5 As shown, electronic device 40 is represented in the form of a general-purpose computing device. The components of electronic device 40 may include, but are not limited to: one or more processors or processing units 401, system memory 402, and bus 403 connecting different system components (including system memory 402 and processing unit 401).

[0126] Bus 403 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0127] Electronic device 40 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 40, including volatile and non-volatile media, removable and non-removable media.

[0128] System memory 402 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 404 and / or cache memory 405. Electronic device 40 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 406 may be used to read and write non-removable, non-volatile magnetic media (… Figure 5 Not shown; usually referred to as a "hard drive"). Although Figure 5 As not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 403 via one or more data media interfaces. Memory 402 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of this application.

[0129] A program / utility 408 having a set (at least one) of program modules 407 may be stored, for example, in memory 402. Such program modules 407 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 407 typically perform the functions and / or methods described in the embodiments of this application.

[0130] Electronic device 40 can also communicate with one or more external devices 409 (e.g., keyboard, pointing device, display 410, etc.), and with one or more devices that enable a user to interact with electronic device 40, and / or with any device that enables electronic device 40 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 411. Furthermore, electronic device 40 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 412. As shown, network adapter 412 communicates with other modules of electronic device 40 via bus 403. It should be understood that, although... Figure 5 Not shown, other hardware and / or software modules may be used in conjunction with electronic device 40, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0131] The processing unit 401 executes various functional applications and page processing by running programs stored in the system memory 402, such as implementing the nasal cavity and pharyngeal image structure recognition method provided in the embodiments of this application.

[0132] Example 5

[0133] This application also provides a storage medium containing computer-executable instructions, which, when executed by a computer processor, are used to perform a method for recognizing the structure of a nasal cavity and pharynx image. The method includes:

[0134] The nasal cavity and laryngoscope image to be identified is input into a pre-trained first structure recognition model to determine the main structure category to which the nasal cavity and laryngoscope image to be identified belongs; wherein, the main structure category includes the glottic region category and at least one non-glottic region category;

[0135] If the main structural category is the laryngeal glottic region category, then based on the central region features of the nasopharyngeal endoscope image to be identified, the vocal cord state subtype identification result of the nasopharyngeal endoscope image to be identified is determined.

[0136] If the main structural category is at least one non-glottic region category, the nasal cavity laryngoscope image to be identified is input into a pre-trained second structural recognition model to determine the anatomical location subtype recognition result under the main structural category to which the nasal cavity laryngoscope image to be identified belongs.

[0137] The computer storage medium in this application embodiment can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0138] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0139] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including—but not limited to—wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0140] Computer program code for performing the operations of the embodiments of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0141] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

Claims

1. A method for recognizing the structure of a nasal cavity and pharynx in an image, characterized in that, The method includes: The nasal cavity and laryngoscope image to be identified is input into a pre-trained first structure recognition model to determine the main structure category to which the nasal cavity and laryngoscope image to be identified belongs; wherein, the main structure category includes the glottic region category and at least one non-glottic region category; If the main structural category is the laryngeal glottic region category, then based on the central region features of the nasopharyngeal endoscope image to be identified, the vocal cord state subtype identification result of the nasopharyngeal endoscope image to be identified is determined. If the main structural category is at least one non-glottic region category, the nasal cavity laryngoscope image to be identified is input into a pre-trained second structural recognition model to determine the anatomical location subtype recognition result under the main structural category to which the nasal cavity laryngoscope image to be identified belongs.

2. The method according to claim 1, characterized in that, The determination of the vocal cord state subtype of the nasal cavity and laryngoscope image to be identified based on the central region features includes: Edge detection is performed on the nasopharyngoscope image to be identified, and the central region of the image is identified and located; Based on each pixel within the central region of the image, the central region features of the nasopharyngoscope image to be identified are determined. Based on the central region features and vocal cord state determination mapping conditions, the vocal cord state subtype identification result of the nasopharyngoscope image to be identified is determined.

3. The method according to claim 2, characterized in that, The step of performing edge detection on the nasopharyngoscope image to be identified, and identifying and locating the central region of the image, includes: The nasal laryngoscope image to be identified is input into the edge detection model to extract the edge contour points of the central region of the image. The edge contour points are traversed sequentially. Edge contour points with a distance of less than a preset number of pixels between adjacent points are discarded. Edge contour points with a distance of more than a preset number of pixels between adjacent points are supplemented by interpolation points with a preset number of pixels as the interval, so as to obtain a sparse and uniform contour point sequence. The image region enclosed by the sequence of contour points is defined as the image center region.

4. The method according to claim 2, characterized in that, The step of determining the central region features of the nasopharyngoscope image to be identified based on the grayscale values ​​of each pixel within the central region of the image includes: The first feature parameter is obtained by averaging the gray values ​​of all pixels in the central region. The standard deviation of the gray values ​​of all pixels in the central region is calculated to obtain the second feature parameter. The perimeter and area of ​​the central region are calculated, and a compactness factor is calculated based on the ratio of perimeter to area to obtain the third feature parameter. At least one of the first feature parameter, the second feature parameter, and the third feature parameter is used as the central region feature of the nasopharyngoscope image to be identified.

5. The method according to claim 4, characterized in that, The vocal cord state determination mapping conditions include: pixel grayscale mean determination mapping conditions, pixel dispersion determination mapping conditions, and compactness determination mapping conditions. The determination of the vocal cord state subtype identification result based on the central region features and the preset vocal cord state determination conditions includes: Based on the pixel grayscale mean determination mapping condition, the first feature parameter is mapped to a vocal cord state subtype to determine a first confidence level determination result; wherein, the pixel grayscale mean determination mapping condition includes: the degree of closeness between the value of the first feature parameter and the value of the first preset pixel is positively correlated with the first confidence level of the determination result of the vocal cord closure state subtype; the degree of closeness between the value of the first feature parameter and the value of the second preset pixel is positively correlated with the second confidence level of the determination result of the vocal cord open state subtype. Based on the pixel dispersion determination mapping condition, the second feature parameter is used to determine the vocal cord state subtype, and a second confidence level determination result is determined; wherein, the pixel dispersion determination mapping condition includes: the magnitude of the second feature parameter value is positively correlated with the first confidence level of the recognition result of the vocal cord closure state subtype; the magnitude of the second feature parameter value is negatively correlated with the second confidence level of the recognition result of the vocal cord open state subtype. Based on the compactness determination mapping conditions, the vocal cord state subtype is determined using the third feature parameter, and a third confidence level determination result is obtained; wherein, the compactness determination mapping conditions include: the magnitude of the third feature parameter value is positively correlated with the first confidence level of the determination result for the vocal cord closure state subtype; the magnitude of the third feature parameter value is negatively correlated with the second confidence level of the determination result for the vocal cord open state subtype. Based on the first confidence level determination result, the second confidence level determination result, and / or the third confidence level determination result, the vocal cord state subtype identification result of the nasopharyngoscopy image to be identified is determined.

6. The method according to claim 1, characterized in that, The at least one non-glottic region category includes: nasal cavity structure category, nasopharyngeal structure category, hypopharyngeal structure category and / or pharyngeal structure category.

7. The method according to claim 1 or 6, characterized in that, The anatomical location subtype identification results under the nasal cavity structure category include: left nasal cavity structure category or right nasal cavity structure category; the anatomical location subtype identification results under the nasopharyngeal structure category include: left nasopharyngeal structure category or right nasopharyngeal structure category; the anatomical location subtype identification results under the hypopharyngeal structure category include: left piriform fossa structure category, right piriform fossa structure category or midline hypopharyngeal structure category; the anatomical location subtype identification results under the pharyngeal structure category include: proximal and distal hypopharynx or root of tongue and vallecula.

8. A device for recognizing the structure of a nasal cavity and pharynx, characterized in that, The device includes: The main structure category recognition module is used to input the nasal cavity laryngoscope image to be recognized into a pre-trained first structure recognition model to determine the main structure category to which the nasal cavity laryngoscope image to be recognized belongs; wherein, the main structure category includes the laryngeal glottic region category and at least one non-glottic region category; The first subtype structure category identification module is used to determine the vocal cord state subtype identification result of the nasal cavity laryngoscope image to be identified based on the central region features of the nasal cavity laryngoscope image to be identified if the main structure category is the laryngeal glottic region category. The second subtype structure category recognition module is used to input the nasal cavity laryngoscope image to be identified into a pre-trained second structure recognition model if the main structure category is at least one non-glottic region category, and to determine the anatomical location subtype recognition result under the main structure category to which the nasal cavity laryngoscope image to be identified belongs.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the nasal cavity and pharyngeal image structure recognition method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the nasal cavity and pharyngeal image structure recognition method according to any one of claims 1-7.

Images