CBCT three-dimensional image recognition method for adenoid body size evaluation

Abstract

The invention discloses a CBCT three-dimensional image recognition method for adenoid size evaluation, and relates to the technical field of medical practice, and the method comprises the steps: setting a boundary based on a CBCT image, then calculating the volume of an adenoid, carrying out the reinforcement and correction of the adenoid and nasopharyngeal airway boundaries based on strip convolution, and outputting the volume and the evaluation size of the adenoid through a multi-scale feature fusion module. According to the method, one-stop processing from image data input to adenoid volume calculation is realized through an automatic system, the three-dimensional adenoid boundary is accurately identified and automatically corrected based on a sine strip-shaped convolution model, and the volume ratio is output in combination with a multi-scale fusion algorithm; the method can eliminate the problems of manual calibration errors and multi-software collaboration in a traditional method, ensures the consistency of cross-patient data, remarkably improves the efficiency, unifies the platform and simplifies the operation process, compresses the traditional several-hour manual operation into full-automatic minute-level processing, and provides a standardized and high-precision diagnostic tool for clinic.

Description

Technical Field

[0001] The present invention relates to the field of medical practice technology, in particular to a CBCT three-dimensional image recognition method for adenoid size assessment. Background Art

[0002] As people pay more attention to their health, the need to accurately assess the size of adenoids has become more urgent. Adenoids are lymphatic tissues located above and behind the pharynx. They are an important part of the human immune system. Their main function is to protect the upper respiratory tract from invasion by external pathogens. Adenoids usually shrink gradually with age, but in some cases, such as infection or allergic reaction, adenoids may hypertrophy, causing a series of health problems. Adenoid hypertrophy may cause a variety of clinical symptoms, including obstructive sleep apnea, otitis media, nasal congestion, mouth breathing, etc. These symptoms may affect the patient's normal breathing, sleep quality, hearing and other physiological functions, and even affect the patient's overall quality of life. Therefore, accurate assessment of the size and degree of hypertrophy of the adenoids is crucial for early diagnosis and treatment. In the clinical diagnosis of adenoids, accurate evaluation methods are directly related to the treatment effect. Through scientific evaluation methods, doctors can more accurately judge whether the adenoids are enlarged, the degree of enlargement, and whether clinical symptoms will be caused. Correct evaluation can help doctors formulate more reasonable treatment plans, thereby improving efficacy and reducing side effects. On the contrary, inaccurate evaluation may lead to misdiagnosis or missed diagnosis, affecting the pertinence and effect of treatment. As the disease progresses, adenoids hypertrophy not only affects the patient's short-term health, but may also have a profound negative impact on his or her growth and development, mental health, social ability and other aspects. Therefore, accurate assessment of the size of the adenoids is of great significance for early diagnosis and timely intervention. Modern medical imaging technology, such as cone beam computed tomography equipment and three-dimensional reconstruction technology, can provide accurate adenoids size assessment, thereby helping doctors develop personalized treatment plans and improve treatment outcomes.

[0003] The current adenoids assessment method is to judge whether the adenoids are enlarged based on the patient's clinical symptoms, such as nasal congestion, snoring, sleep apnea and hearing loss. Doctors make a preliminary judgment based on the patient's symptoms. However, the severity of symptoms in this method varies from individual to individual, and the assessment results are relatively subjective. The size and degree of enlargement of the adenoids cannot be accurately assessed based on symptoms alone, which can easily lead to missed diagnosis or misdiagnosis. The current methods for evaluating adenoids include X-ray imaging assessment, which uses X-rays to take images of the adenoids to observe their size and shape and whether they compress the airway. However, the disadvantages of this method are radiation risks, and the images are two-dimensional and lack comprehensive three-dimensional information.

[0004] Currently, the existing methods for evaluating adenoids include CT scanning, which generates high-resolution three-dimensional images of adenoids, providing a clearer picture of the size of adenoids and their compression of surrounding structures, such as the upper airway and nasal cavity. However, the disadvantages of this method are high radiation exposure, high cost, complex operation, and the need for professional equipment and technical support. Currently, the existing methods for evaluating adenoids include magnetic resonance imaging, which uses a strong magnetic field and radio waves to generate detailed images of the adenoids. It is particularly suitable for imaging soft tissues and can clearly show the shape and size of the adenoids. However, the disadvantages of this method are that the examination time is long, the cost is high, the operation is complicated, and the patient's tolerance is high. The current existing adenoids assessment methods include endoscopic examination, which uses an endoscope to directly observe the adenoids through the nasal cavity or mouth. It is often used to check symptoms such as nasal congestion and snoring caused by adenoid hypertrophy. However, the disadvantage of this method is that it is limited to local observation, and some patients may not be able to adapt to it. In view of the above limitations, it is urgent to introduce more scientific and accurate evaluation methods to improve the accuracy, efficiency and standardization of diagnosis. Summary of the invention

[0005] The purpose of the present invention is to provide a CBCT three-dimensional image recognition method for adenoid size assessment, which solves the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a CBCT three-dimensional image recognition method for adenoid size assessment, comprising the following steps: Step S1, after setting the boundary based on the CBCT image, calculate the volume of the adenoids; Step S11, acquisition and preprocessing of CBCT image data; Step S12, correcting the head position and determining the FH plane; Step S13, determining the boundaries of adenoids and nasopharyngeal airway; Step S2: strengthening and correcting the boundaries of adenoids and nasopharyngeal airway based on strip convolution; Step S21, constructing a strip convolution using a sine function; Step S22, segmenting adenoids and nasopharyngeal airway; Step S23, extracting the adenoids and nasopharyngeal airway boundary boxes; Step S24, adenoids and nasopharyngeal airway margin correction; Step S25, calculating the volume of adenoids and nasopharyngeal airway; Step S3, the multi-scale feature fusion module outputs the volume; Step S31, preprocessing the input image and adapting the low-resolution image input; Step S32: lightweight multi-scale feature extraction; Step S33, branch processing module, specialization of low computational load; Step S34, dynamic convolution module, fast feature integration and quantization dynamic convolution; Step S35, attention fusion module; Step S4: Evaluate the size of the adenoids.

[0007] Optionally, the specific process of acquiring and preprocessing the CBCT image data in step S11 is as follows: The patient's three-dimensional image data is obtained through the CBCT scanning equipment, and then the three-dimensional image data is cleaned, denoised and three-dimensionally reconstructed to generate a visual three-dimensional model.

[0008] Optionally, the specific process of correcting the head position and determining the FH plane in step S12 is as follows: Based on the visualized three-dimensional model outputted in step S11, the connecting line of the left and right orbital points on the coronal plane is made parallel to the horizontal line, and the orbitoauricular plane on the sagittal plane is made parallel to the horizontal plane; The specific process of determining the adenoids and nasopharyngeal airway boundaries in step S13 is as follows: The boundaries of the adenoids and nasopharyngeal airway were determined in the three-dimensional model, including the sagittal anterior boundary, sagittal posterior boundary, sagittal superior boundary, sagittal inferior boundary, coronal anterior boundary, coronal lateral boundary, coronal posterior boundary, axial anterior boundary, axial posterior boundary and axial lateral boundary. A rectangular model was constructed in the three-dimensional space by marking the boundaries of the adenoids and nasopharyngeal airway in the coronal, sagittal and axial planes.

[0009] Optionally, the specific process of constructing the strip convolution using the form of a sine function in step S21 is: Sine strip convolution is an operation that slides a one-dimensional sinusoidal strip convolution kernel along one direction of the image to extract edges with linear and periodic characteristics. Sine strip convolution combines the periodicity of the sine function and enhances the edges, textures, and periodic structures in the image through the periodic characteristics of the sine wave. The sinusoidal strip convolution kernel can be expressed by a sinusoidal function, and the calculation formula is as follows: in: W(x) Refers to the sinusoidal strip convolution kernel; A Refers to the amplitude, which determines the strength of the sine wave; f Refers to the frequency, controls the periodicity of the sine wave, and affects the scale of the detected image features; Refers to the phase, which controls the starting position of the sine wave; xRefers to spatial variables, indicating the position coordinates along a certain direction in the image; When convolving the input image, the sinusoidal strip convolution kernel W(x) Slide along one direction of the image and perform local weighted sum on the image to extract edge, texture and periodic structure information in the image; The calculation formula of sinusoidal strip convolution is: ; in: S( x, y ) refers to the image at position ( x, y )’s convolution result; I ( x + i, y ) refers to the image at position ( x + i, y )’s pixel value; Refers to a sinusoidal strip convolution kernel that slides in multiple directions; k Refers to the length of the sinusoidal strip convolution kernel, which determines the convolution operation; Sine Strip Convolution Kernel It is periodic, which can help highlight the periodically changing parts of the image and enhance the edge detection effect; When performing sinusoidal strip convolution, the sinusoidal strip convolution kernel slides in the three-dimensional CBCT image to calculate the weighted sum of each position in the image. The specific steps are as follows: For each three-dimensional pixel I ( x, y, z), use the sinusoidal strip convolution kernel to perform weighted calculation with the surrounding pixels of the point; Output S ( x, y, z) is a three-dimensional pixel value in the image, indicating the edge strength at that location; The calculation formula for three-dimensional sinusoidal strip convolution is: ; in: S ( x, y, z) refers to the three-dimensional image pixel value, indicating ( x, y, z) edge strength at position; I ( x, y, z) refers to the image at position ( x, y, z) pixel value; W ( i, j, l ) refers to the three-dimensional sinusoidal strip convolution kernel.

[0010] Optionally, the adenoids and nasopharyngeal airway edge correction in step S24 includes: image input, edge extension, sinusoidal strip convolution kernel of the expanded image, and edge correction; Image input: The pixel value of the input original image is represented as I ( x, y ), whose dimensions are M × N , edge processing requires expansion of edge pixels: Extended Edge: In the original image pixel value I ( x, y ) is mirrored around, and the width of the expansion is w , w is the filter radius to obtain the expanded image pixel value I ext ( x, y ); The calculation formula of the expanded image pixel value is: in: I ext ( x, y ) refers to the expanded image pixel value; I ( x, y ) refers to the original image pixel value; M Refers to the width of the image; N Refers to the height of the image; I ( x, y ), 0≤ x < M Refers to I ( x, y ) is within the valid range of the image, that is, 0≤ x < M When directly using I ( x, y )’s original intensity value; I ( w - x, y ), x <0 refers to processing the boundary pixels on the left side of the original image, and using the corresponding pixels on the right side of the image through mirror expansion ( w - x, y ) to replace the intensity value of the image edge, and maintain the continuity and smoothness of the image edge during edge correction; I ( 2M - x - 1, y ), x ≥ M Refers to processing the boundary pixels on the right side of the image, using the corresponding pixels on the left side of the image through mirroring expansion (2M - x - 1, y ) to replace the intensity value of the image edge, and maintain the continuity and smoothness of the image edge during edge correction; The calculation formula of the sinusoidal strip convolution kernel of the expanded image is as follows: ; ; in: K ( x, y ) refers to the expanded image pixel value I ext ( x, y )’s sinusoidal strip convolution kernel; G ( x, y ) refers to the Gaussian weighting function; σ Refers to the standard deviation of the Gaussian kernel, which is used to control the degree of smoothing.

[0011] Optionally, the specific process of calculating the adenoids and nasopharyngeal airway volumes in step S25 is as follows: Based on the 3D reconstructed model, the number of voxels in the adenoids and nasopharyngeal airway area was calculated. V voxel It is the basic unit in a three-dimensional image, and each voxel represents a spatial position in the image; The calculation formula for adenoid and nasopharyngeal airway volume is as follows: ; ;in: V Refers to the total volume of the adenoids and nasopharyngeal airway. The total volume of the adenoids is obtained by summing up all voxels in the adenoids and nasopharyngeal airway area. V voxel Refers to the volume of each voxel; d x ×d y ×d z Refers to the volume of each voxel. The voxel volume is usually determined by the resolution of the image, that is, the actual size of each voxel. If the resolution of the image is d x ×d y ×d z , then the volume of each voxel is V voxel .

[0012] Optionally, the specific process in the output volume of the multi-scale feature fusion module in step S3 is as follows: The specific process of the input image preprocessing and low-resolution image input adaptation in step S31 is as follows: The resolution of the input image is reduced by downsampling, and the calculation volume is reduced from the source. This is done using stride convolution or bilinear interpolation. The calculation formula is as follows: ;in: I 、 Refers to the low-resolution image after downsampling; I x，y Refers to the original image; s Refers to the reduction ratio; the specific process of lightweight multi-scale feature extraction in step S32 is as follows: The core of the design of the multi-scale feature fusion module is to reduce the volume of feature maps and extract rich multi-scale information at the same time, including multi-scale parallel paths and compression operations; The multi-scale parallel path is as follows: When extracting features of different receptive fields, redundancy is reduced by sharing the feature map basic calculation and adding lightweight convolution kernels; The compression operations are as follows: After multi-scale feature fusion, 1×1 convolution is used to compress the channel. The calculation formula is as follows: ; in: Refers to the compressed feature map; F MSFM ( x, y ) refers to the original feature map input to the multi-scale feature fusion module; Conv 1×1 Refers to 1×1 convolution, which is used to reduce the number of channels; The branch processing module in step S33 is specialized in the specific process of low computational complexity: Through branch processing for detail enhancement, context capture and global information, specialized lightweight processing methods are used to further optimize the amount of computation, including: detail enhancement branch, context capture branch and global information branch; Detail enhancement branch: Use a high-pass filter to simulate gradient operations through convolution, extract edge detail information, and compress the volume; The calculation formula of the detail enhancement branch is as follows: ; in: Refers to the enhanced detail feature map; HighPass Refers to a high-pass filter, used to extract edge details; Context Capture Branch: Use dilated convolution to expand the receptive field and avoid reducing the resolution of feature maps; The calculation formula of the context capture branch is as follows: ; in: Refers to the detailed feature map after context capture; DilatedConv Refers to dilated convolution; rate Refers to the dilation rate, which is used to control the size of the receptive field; rate =2 means the convolution kernel interval is 2 pixels; Global information branch: Use global average pooling to directly extract global features, which has low computational complexity. After each branch output is lightweighted, the computational volume is further reduced. The calculation formula of the global information branch is as follows: ; in: Refers to the detail feature map after the global information branch; GlobalAvgPooling Refers to global average pooling, which is used to compress the feature map of each channel into a single value.

[0013] Optionally, the specific process in the output volume of the multi-scale feature fusion module in step S3 further includes: The specific process of the dynamic convolution module in step S34, the fast feature integration quantization dynamic convolution, is as follows: Weight generation: Using compressed feature input, the weight generation network is designed as a lightweight version, including a single-layer linear mapping. The calculation formula of dynamic weight is as follows: ; in: W i Refers to dynamic weight; Linear Refers to a single-layer fully connected network; Softmax refers to the normalized weight; Refers to i The compression characteristics of the branches, i Includes detail branches, context branches, and global branches; Feature integration: Dynamic convolution weights perform weighted summation of features. The lightweight design of the dynamic convolution module reduces computational costs while retaining the ability to adaptively adjust branch features. The calculation formula for feature integration is as follows: ; in: F DCM ( x, y ) refers to the feature map output by the dynamic convolution module, which represents the adaptive fusion result of multi-scale features; Refers to i The branches are at position ( x, y )’s eigenvalues; The specific process of the attention fusion module in step S35 is as follows: Spatial attention is calculated by a lightweight convolution with a kernel size of 1×1, and the calculation formula is as follows; ; in: F 空间注意力 ( x, y ) refers to the spatial attention weight map; Sigmoid Refers to the activation function, which normalizes the weight to the interval [0,1] to indicate the importance of spatial position; MaxPool Refers to the maximum pooling operation, which extracts the maximum value at each position in the feature map; AvgPool Refers to the average pooling operation, which extracts the mean of each position in the feature map; The channel attention mechanism is completed through global average pooling and single-layer perceptron, and the calculation formula is as follows: ; in: F 通道注意力 ( x, y ) refers to the channel attention vector; Linear Refers to a single-layer fully connected network; GAP Refers to global average pooling; F DCM Refers to the multi-scale features output by the dynamic convolution module; The calculation formula of the final fusion feature is as follows: ; in: F AFM ( x, y ) refers to the optimized feature map; By simplifying the computational path, the attention mechanism achieves a balance between low computational volume and efficient feature enhancement.

[0014] Optionally, the calculation process of the estimated size of the adenoids in step S4 is as follows: The size of the adenoids was assessed by calculating the volume ratio of the adenoids to the nasopharyngeal cavity using the following formula: Adenoid size index = adenoid volume / (adenoid volume + nasopharyngeal airway volume).

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention has full-process automated operation, from the input of CBCT data to the calculation of adenoid volume, which is automatically completed by the AI ​​system, reducing tedious manual operations, greatly improving work efficiency, saving a lot of time, and being able to process more data. This is different from the posture movements and calculation processes in traditional technologies, which usually require doctors to manually calibrate adenoid boundaries, adjust reference layouts, and perform boundary corrections and other tedious workflows.

[0016] The present invention automatically processes the three-dimensional data of adenoids through a sinusoidal strip convolution model, and uses a precise algorithm to identify and correct the adenoids boundaries and calculate the volume, which greatly reduces manual operations and provides higher accuracy and consistency. Through the automated process, the measurements of all patients will remain highly consistent, avoiding deviations caused by operational deviations. This is different from existing methods such as three-dimensional model analysis, boundary drawing, and boundary correction, which often rely on manual operations and empirical judgments, and the measurement accuracy is greatly affected by the operator's skills and additional factors.

[0017] The system of the present invention seamlessly integrates modules such as data input, 3D model reconstruction, boundary correction, and volume calculation, providing a unified and comprehensive platform, reducing the complexity of data processing, greatly improving the convenience of operation, and quickly completing the entire measurement process, which is different from the existing technology. In the existing technology, there may be problems such as data incompatibility or poor information transmission between multiple software tools, resulting in cumbersome operation processes and problems that cannot be analyzed. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 This is a schematic diagram of the structure of connecting bilateral infraorbital points in the coronal position of the present invention; Figure 2 It is a structural schematic diagram of the sagittal FH plane of the present invention; Figure 3 It is a structural schematic diagram of the sagittal adenoids and nasopharyngeal airway boundary marking of the present invention; Figure 4 It is a structural schematic diagram of the coronal adenoids and nasopharyngeal airway boundaries of the present invention; Figure 5 It is a structural schematic diagram of the axial adenoids and nasopharyngeal airway boundary marking of the present invention; Figure 6It is a schematic diagram of the structure of the three-dimensional cuboid model with combined sagittal, coronal and axial boundary annotations of the present invention; Figure 7 It is a structural schematic diagram of the multi-scale feature fusion module of the present invention.

[0019] Figure 8 Flow chart of the steps of the CBCT 3D image recognition method for adenoid size assessment. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Example 1, please refer to Figures 1 to 8 This embodiment provides a technical solution: a CBCT three-dimensional image recognition method for adenoid size assessment, comprising the following steps: Step S1, after setting the boundary based on the CBCT image, calculate the volume of the adenoids; Step S11, acquisition and preprocessing of CBCT image data; Step S12, correcting the head position and determining the FH plane; Step S13, determining the boundaries of adenoids and nasopharyngeal airway; Step S2: strengthening and correcting the boundaries of adenoids and nasopharyngeal airway based on strip convolution; Step S21, constructing a strip convolution using a sine function; Step S22, segmenting adenoids and nasopharyngeal airway; Step S23, extracting the adenoids and nasopharyngeal airway boundary boxes; Step S24, adenoids and nasopharyngeal airway margin correction; Step S25, calculating the volume of adenoids and nasopharyngeal airway; Step S3, the multi-scale feature fusion module outputs the volume; Step S31, preprocessing the input image and adapting the low-resolution image input; Step S32: lightweight multi-scale feature extraction; Step S33, branch processing module, specialization of low computational load; Step S34, dynamic convolution module, fast feature integration and quantization dynamic convolution; Step S35, attention fusion module; Step S4: Evaluate the size of the adenoids.

[0022] In this embodiment: In the imaging mode, the present invention uses CBCT equipment, which is different from the prior art that uses radiation and CT, but the method is two-dimensional or low-resolution three-dimensional, and the soft tissue contrast is poor. The present invention uses three-dimensional high-resolution imaging and accurately restores the spatial relationship between adenoids and airways; different from the prior art in that it may rely more on manual annotation and threshold segmentation when processing boundaries, which may cause blurred edges. The present invention uses sinusoidal strip convolution to strengthen anatomical boundaries, weight correction to eliminate artifacts, and high segmentation accuracy; different from the prior art in terms of computational efficiency, the prior art may use manual measurement, which is time-consuming and has high redundancy in traditional network calculations, and the present invention uses MSFM through dynamic convolution and attention mechanism, and the reasoning speed is faster; different from the prior art in terms of clinical applicability, the prior art may use MRI, which makes the cost high and the endoscope is highly invasive, the present invention samples non-invasive CBCT and fully automatic analysis, thereby reducing the cost to a certain extent, and is suitable for large-scale screening; different from the prior art in terms of result standardization, the prior art may be based on the analysis of physicians, but experience differences lead to evaluation deviations, the present invention samples the standardized formula for comprehensive analysis, so that the comparability of results from different medical institutions is enhanced.

[0023] See also Figures 1 to 8 The specific process of acquiring and preprocessing CBCT image data in step S11 is as follows: The patient's three-dimensional image data is obtained through the CBCT scanning equipment, and then the three-dimensional image data is cleaned, denoised and three-dimensionally reconstructed to generate a visual three-dimensional model.

[0024] In this embodiment: In the process of three-dimensional image analysis, the first step is the data acquisition stage. Using a cone beam computed tomography (CBCT) device, detailed three-dimensional images of the patient can be obtained. In order to ensure the clarity and low noise of the image, the setting of scanning parameters is crucial, including the resolution is usually set to 0.3 mm or higher to capture fine structures; the voltage is adjusted between 70-120 kVp to obtain appropriate X-ray penetration; the current is controlled at 5-10 mA to balance the image quality and the patient's radiation dose. The precise setting of these parameters is crucial for subsequent diagnosis and treatment planning; The second is the image preprocessing stage. First, in order to remove the noise introduced during the scanning process, median filtering or Gaussian filtering technology is applied. These methods can effectively reduce noise while retaining edge information; Finally, the 3D reconstruction stage is entered to reconstruct the CBCT images in three dimensions and generate a visualization model. This step involves complex image processing and computer vision technologies, including image segmentation, feature extraction, and 3D modeling. The final 3D model can intuitively display the structure and provide strong visual support for clinical diagnosis and treatment. By implementing these technologies, details can be extracted from the original CBCT images. Accurate 3D information can be extracted.

[0025] See also Figures 1 to 8 In step S12, the head position is corrected and the specific process of determining the FH plane is as follows: Based on the visualized three-dimensional model outputted in step S11, the connecting line of the left and right orbital points on the coronal plane is made parallel to the horizontal line, and the orbitoauricular plane on the sagittal plane is made parallel to the horizontal plane; In this embodiment: First, by connecting the bilateral infraorbital points on the coronal plane and making them parallel to the horizontal plane, as shown in FIG. Figure 1 As shown, the bilateral infraorbital points were marked on the image, and then the coronal plane was adjusted to pass through these two points to ensure that the connecting line of the bilateral infraorbital points was parallel to the horizontal plane; Secondly, in the sagittal plane, connect the infraorbital point with the auricular point and ensure that this plane (orbitoauricular plane) is parallel to the horizontal plane, e.g. Figure 2 As shown, connect the infraorbital point and the ear point and adjust the head position so that the connection line is parallel to the horizontal plane to ensure that the head is in a standardized posture during imaging examinations. Correcting the head position before measurement can help ensure that the head maintains a consistent angle and position during CBCT imaging examinations, thereby improving the accuracy and repeatability of imaging data and reducing measurement errors caused by angle deviations. It is particularly suitable for accurate evaluation of facial anatomical structures, adenoids and other areas.

[0026] The specific process of determining the adenoids and nasopharyngeal airway boundaries in step S13 is as follows: The boundaries of the adenoids and nasopharyngeal airway were determined in the three-dimensional model, including the sagittal anterior boundary, sagittal posterior boundary, sagittal superior boundary, sagittal inferior boundary, coronal anterior boundary, coronal lateral boundary, coronal posterior boundary, axial anterior boundary, axial posterior boundary and axial lateral boundary. A rectangular model was constructed in the three-dimensional space by marking the boundaries of the adenoids and nasopharyngeal airway in the coronal, sagittal and axial planes.

[0027] In this embodiment, first, the sagittal adenoids and nasopharyngeal airway boundaries are marked. The marking method of the sagittal boundaries can be divided into the following parts: Figure 3 As shown: Anterior border in sagittal plane: The anterior border is defined as the edge of the adenoids. Figure 3 The marked line 1A in the figure marks this boundary, indicating the position of the anterior edge of the adenoid in the sagittal plane; Posterior limit in sagittal plane: The posterior limit is a line passing through the spheno-occipital fusion and perpendicular to the FH plane. Figure 3 In the figure, the marker line 23A indicates the position of the posterior limit and defines the exact position of the posterior limit in the sagittal plane; The upper limit of the sagittal plane: The upper limit passes through the spheno-occipital cartilage and is parallel to the FH plane. The marking line 2A marks the upper limit. The lower limit of the sagittal plane: The lower limit passes through the tip of the uvula and is parallel to the FH plane. The marking line 4A marks the position of the lower limit.

[0028] The superior and inferior boundaries of the nasopharyngeal airway were defined identically to the borders of the adenoids, whereas the perimeter was the edge of the upper airway; The second is the coronal adenoid and nasopharyngeal airway boundary marking: The coronal boundary marking method can be divided into the following parts: Figure 4 As shown: Anterior border in the coronal plane: The anterior border is defined as the front edge of the adenoids. Figure 4 The mid-anterior boundary is marked as marker line five 1B; Lateral boundary of the coronal plane: The lateral boundary is composed of the medial plates of the sphenoid bones on both sides. Figure 4 The marked line 63B in FIG. 3 indicates the position of the lateral limit; Posterior limit in the coronal plane: The posterior limit is defined by a line passing through the sphenoid cartilages and perpendicular to the FH plane. Figure 4 Indicated by the marking line 72B; Then the axial adenoids and nasopharyngeal airway boundaries are marked: the axial boundary marking method can be divided into the following parts: Figure 5 As shown: Anterior border in axial position: The anterior border is defined as the edge of the adenoids. Figure 5 The line marked in Figure 8.1C marks this boundary; Posterior limit in axial position: the line passing through the spheno-occipital fusion and perpendicular to the FH plane. Figure 5 The marked line 93C in the figure indicates the position of the posterior limit, which means that the posterior limit passes through the butterfly cartilage junction and is perpendicular to the FH plane; Lateral limit of axial position: The lateral limit is composed of the medial plate of the sphenoid bone on both sides. Figure 5 In the figure, the marked line + 2C indicates the location of the lateral boundary; By accurately marking the adenoids and nasopharyngeal airway boundaries in the coronal, sagittal, and axial planes, such as Figure 6As shown in the figure, a rectangular model can be constructed in three-dimensional space, in which the 1D area is used to clearly mark the position and range of the adenoids, and the 2D area is used to clearly mark the position and range of the upper airway. This multi-dimensional labeling method can not only accurately reflect the morphological characteristics of the adenoids in different directions, but also help to better understand its spatial position and volume size in the anatomical structure. The complex three-dimensional visualization technology provides an important reference for medical diagnosis and treatment planning, allowing doctors to more accurately evaluate the pathological state of the adenoids. The traditional method is to use lateral radiographs to perform two-dimensional line distance measurements, while this method innovatively uses CBCT to measure the volume of adenoids and nasopharynx, and evaluates the size of adenoids by calculating their volume ratio. Secondly, it automatically measures through the AI ​​system, thereby reducing tedious manual operations, greatly improving work efficiency, saving a lot of time, and being able to process more data.

[0029] See also Figures 1 to 8 In step S21, the specific process of constructing strip convolution using the form of sine function is as follows: Sine strip convolution is an operation that slides a one-dimensional sinusoidal strip convolution kernel along one direction of the image to extract edges with linear and periodic characteristics. Sine strip convolution combines the periodicity of the sine function and enhances the edges, textures, and periodic structures in the image through the periodic characteristics of the sine wave. The sinusoidal strip convolution kernel can be expressed by a sinusoidal function, and the calculation formula is as follows: in: W(x) Refers to the sinusoidal strip convolution kernel; A Refers to the amplitude, which determines the strength of the sine wave; f Refers to the frequency, controls the periodicity of the sine wave, and affects the scale of the detected image features; Refers to the phase, which controls the starting position of the sine wave; x Refers to spatial variables, indicating the position coordinates along a certain direction in the image; When convolving the input image, the sinusoidal strip convolution kernel W(x) Slide along one direction of the image and perform local weighted sum on the image to extract edge, texture and periodic structure information in the image; The calculation formula of sinusoidal strip convolution is: ; in: S( x, y ) refers to the image at position ( x, y )’s convolution result; I ( x + i, y ) refers to the image at position ( x + i, y )’s pixel value; Refers to a sinusoidal strip convolution kernel that slides in multiple directions; k Refers to the length of the sinusoidal strip convolution kernel, which determines the convolution operation; Sine Strip Convolution Kernel It is periodic, which can help highlight the periodically changing parts of the image and enhance the edge detection effect; When performing sinusoidal strip convolution, the sinusoidal strip convolution kernel slides in the three-dimensional CBCT image to calculate the weighted sum of each position in the image. The specific steps are as follows: For each three-dimensional pixel I ( x, y, z), use the sinusoidal strip convolution kernel to perform weighted calculation with the surrounding pixels of the point; Output S ( x, y, z) is a three-dimensional pixel value in the image, indicating the edge strength at that location; The calculation formula for three-dimensional sinusoidal strip convolution is: ; in: S ( x, y, z) refers to the three-dimensional image pixel value, indicating ( x, y, z) edge strength at position; I ( x, y, z) refers to the image at position ( x, y, z) pixel value; W ( i, j, l ) refers to the three-dimensional sinusoidal strip convolution kernel.

[0030] The grayscale contrast between adenoids and nasopharyngeal airway was enhanced by three-dimensional sinusoidal strip convolution; It is worth noting that the step S22 of adenoids and nasopharyngeal airway segmentation is further explained; Step S22 includes threshold segmentation: using a threshold method to extract the region of adenoids and nasopharyngeal airway, and setting an appropriate grayscale threshold to distinguish adenoids and nasopharyngeal airway from other tissues based on the comparison of grayscale values ​​with surrounding tissues; Step S22 includes edge detection: using methods such as sinusoidal strip convolution to enhance the edges in the image and accurately extract the boundaries.

[0031] Step S22 includes three-dimensional morphological operations: applying morphological operations such as dilation, erosion, and opening operations to modify the shapes of the adenoids and nasopharyngeal airway regions, fill discontinuous edges or remove noise, and ensure the continuity of boundaries; Step S22 includes deep learning segmentation: in complex situations, a deep learning algorithm, namely U-Net, can be used to automatically segment the adenoids and nasopharyngeal airway areas, which can better handle complex situations in the image and obtain more accurate segmentation results.

[0032] It is worth noting that the extraction of the adenoids and nasopharyngeal airway boundary boxes in step S23 is further explained; The segmented adenoids and nasopharyngeal airway regions and the corresponding two-dimensional slice data are used to reconstruct the three-dimensional model, at which point its contour and volume in three-dimensional space have been formed; Three-dimensional reconstruction is performed using image processing software, such as Mimics and 3D Slicer, to generate a three-dimensional volume model of the adenoids and nasopharyngeal airway.

[0033] The correction of the edge of adenoids and nasopharyngeal airway in step S24 includes: image input, edge extension, sinusoidal strip convolution kernel of the extended image and edge correction; Image input: The pixel value of the input original image is represented as I ( x, y ), whose dimensions are M × N , edge processing requires expansion of edge pixels: Extended Edge: In the original image pixel value I ( x, y ) is mirrored around, and the width of the expansion is w , w is the filter radius to obtain the expanded image pixel value I ext ( x, y ); The calculation formula of the expanded image pixel value is: in: I ext ( x, y ) refers to the expanded image pixel value; I ( x, y ) refers to the original image pixel value; M Refers to the width of the image; N Refers to the height of the image; I ( x, y ), 0≤x < M Refers to I ( x, y ) is within the valid range of the image, that is, 0≤ x < M When directly using I ( x, y )’s original intensity value; I ( w - x, y ), x <0 refers to processing the boundary pixels on the left side of the original image, and using the corresponding pixels on the right side of the image through mirror expansion ( w - x, y ) to replace the intensity value of the image edge, and maintain the continuity and smoothness of the image edge during edge correction; I ( 2M - x - 1, y ), x ≥ M Refers to processing the boundary pixels on the right side of the image, using the corresponding pixels on the left side of the image through mirroring expansion ( 2M - x - 1, y ) to replace the intensity value of the image edge, and maintain the continuity and smoothness of the image edge during edge correction; The calculation formula of the sinusoidal strip convolution kernel of the expanded image is as follows: ; ; in: K ( x, y ) refers to the expanded image pixel value I ext ( x, y )’s sinusoidal strip convolution kernel; G ( x, y ) refers to the Gaussian weighting function; σ Refers to the standard deviation of the Gaussian kernel, which is used to control the degree of smoothing; Edge Correction: In the edge part, since the value of the extended image is not the real image value, the response result is inaccurate, so weight correction needs to be introduced; The calculation formula of weight in edge correction is as follows: ; ;in: W ( x,y ) refers to the weight map, in coordinates ( x,y ) is the effective area weight value after Gaussian weighting, which is used to correct the convolution response later to reduce the interference of the extended area on edge detection; M (x,y ) refers to the binary mask matrix, which is used to limit the effective area range. M ( x,y ) is used to retain only the original image area, the size is M × N , shielding the invalid area introduced by image expansion; Otherwise Refers to other situations; G ( u, v ) refers to the Gaussian kernel function, which is used to perform weighted averaging on the mask area. G ( u, v ) is used to suppress high-frequency noise and smooth edge transition areas; ( u, v ) refers to the support range of the convolution kernel, that is, the offset of the convolution kernel in the horizontal and vertical directions; Corrected response: The modified convolution response is calculated as follows: ;in: R 、 ( x, y ) refers to the modified convolution response map; By dividing by the weight W(x, y) , eliminating the edge area response error caused by image expansion, expanding the area M (x + u, y + v) The value is 0, and its original convolution response value R ( x, y ) is unreliable due to the lack of true pixel values, and such invalid responses can be suppressed by weight correction; R ( x, y ) refers to the original convolution response map; ε Refers to a very small constant to prevent the denominator from being zero; Remove the extended area: Crop the corrected response plot R 、 ( x, y ), retain the original image area 0 ≤ x < M , 0 ≤ y < N; Thereby, artifact interference can be avoided. The response value of the extended area may introduce noise or artifacts due to the lack of real data. Only the valid anatomical area data is retained after cropping. In addition, it is coordinated with the head position correction and combined with the coordinate system of the head position correction, the sagittal plane, coronal plane and axial plane to ensure that the three-dimensional volume calculation is only based on the real anatomical structure. The edge correction solves the boundary blur problem that may be caused by traditional methods, such as the threshold segmentation method, thereby improving the segmentation accuracy.

[0034] The specific process of calculating the adenoid and nasopharyngeal airway volumes in step S25 is as follows: Based on the 3D reconstructed model, the number of voxels in the adenoids and nasopharyngeal airway area was calculated. V voxel It is the basic unit in a three-dimensional image, and each voxel represents a spatial position in the image; The calculation formula for adenoid and nasopharyngeal airway volume is as follows: ; ;in: V Refers to the total volume of the adenoids and nasopharyngeal airway. The total volume of the adenoids is obtained by summing up all voxels in the adenoids and nasopharyngeal airway area. V voxel Refers to the volume of each voxel; d x ×d y ×d z Refers to the volume of each voxel. The voxel volume is usually determined by the resolution of the image, that is, the actual size of each voxel. If the resolution of the image is d x ×d y ×d z , then the volume of each voxel is V voxel .

[0035] In this embodiment: For the strengthening and correction steps of the adenoids and nasopharyngeal airway boundaries based on sinusoidal strip convolution, in the cone beam computed tomography (CBCT) image training process, in order to meet the real-time computing requirements of on-site deployment, an improved sinusoidal-sinusoidal strip convolution MSC model is adopted. By introducing the sinusoidal strip convolution model, the medical image segmentation technology is facilitated. Through multi-directional strip convolution operations, the adenoids and nasopharyngeal airway boundaries can be effectively strengthened and corrected. Specifically, the sinusoidal-sinusoidal strip convolution MSC model can better locate the target area by introducing multi-scale features, and at the same time, combined with multi-directional convolution operations, the accuracy of boundary recognition and correction is further optimized. This technology performs well in the adenoids and nasopharyngeal airway segmentation tasks, and can significantly improve the accuracy and robustness of segmentation, and provide more reliable imaging support for medical diagnosis and treatment. Strip convolution is through The one-dimensional convolution kernel slides along one direction of the image to extract edges with linear or periodic features. The sinusoidal strip convolution combines the periodicity of the sine function and enhances the edges, textures and periodic structures in the image through the periodic characteristics of the sine wave. This method uses accurate adenoid and nasopharyngeal airway boundary recognition and volume calculation methods to construct a three-dimensional structural model of the adenoid based on cone beam computed tomography (CBCT) technology, and achieves efficient volume calculation through voxel accumulation. This method combines resolution calibration and error correction to ensure the accuracy of volume calculation results, providing standardized data support for the diagnosis, evaluation and treatment of adenoid hyperplasia. Through automated adenoid and nasopharyngeal airway boundary correction and enhancement technology, the deep learning model is combined with the morphological algorithm to adaptively adjust the boundaries in the adenoid image and enhance the details to ensure the integrity and clarity of the boundaries. Through feature extraction and local contrast enhancement, the problems of boundary fuzziness and discontinuity in traditional methods are eliminated, the accuracy of boundary recognition is improved, and the subjective errors of human operation are reduced.

[0036] Example 2, based on the above example: Please refer to Figures 1 to 8 , the specific process of input image preprocessing in step S31 and low-resolution image input adaptation is as follows: The resolution of the input image is reduced by downsampling, and the calculation volume is reduced from the source. This is done using stride convolution or bilinear interpolation. The calculation formula is as follows: ;in: I 、 Refers to the low-resolution image after downsampling; I x，y Refers to the original image; s Refers to the reduction ratio; the specific process of lightweight multi-scale feature extraction in step S32 is as follows: The core of the design of the multi-scale feature fusion module is to reduce the volume of feature maps and extract rich multi-scale information at the same time, including multi-scale parallel paths and compression operations; The multi-scale parallel path is as follows: When extracting features of different receptive fields, redundancy is reduced by sharing the feature map basic calculation and adding lightweight convolution kernels; The compression operations are as follows: After multi-scale feature fusion, 1×1 convolution is used to compress the channel. The calculation formula is as follows: ; in: Refers to the compressed feature map; F MSFM ( x, y ) refers to the original feature map input to the multi-scale feature fusion module; Conv 1×1 Refers to 1×1 convolution, which is used to reduce the number of channels; The branch processing module in step S33 is specialized in the specific process of low computational complexity: Through branch processing for detail enhancement, context capture and global information, specialized lightweight processing methods are used to further optimize the amount of computation, including: detail enhancement branch, context capture branch and global information branch; Detail enhancement branch: Use a high-pass filter to simulate gradient operations through convolution, extract edge detail information, and compress the volume; The calculation formula of the detail enhancement branch is as follows: ; in: Refers to the enhanced detail feature map; HighPass Refers to a high-pass filter, used to extract edge details; Context Capture Branch: Use dilated convolution to expand the receptive field and avoid reducing the resolution of feature maps; The calculation formula of the context capture branch is as follows: ; in: Refers to the detailed feature map after context capture; DilatedConv Refers to dilated convolution; rate Refers to the dilation rate, which is used to control the size of the receptive field; rate =2 means the convolution kernel interval is 2 pixels; Global information branch: Use global average pooling to directly extract global features, which has low computational complexity. After each branch output is lightweighted, the computational volume is further reduced. The calculation formula of the global information branch is as follows: ; in: Refers to the detail feature map after the global information branch; GlobalAvgPooling Refers to global average pooling, which is used to compress the feature map of each channel into a single value.

[0037] The specific process of the dynamic convolution module in step S34, the fast feature integration quantization dynamic convolution, is as follows: Weight generation: Using compressed feature input, the weight generation network is designed as a lightweight version, including a single-layer linear mapping. The calculation formula of dynamic weight is as follows: ; in: W i Refers to dynamic weight; Linear Refers to a single-layer fully connected network; Softmax refers to the normalized weight; Refers to i The compression characteristics of the branches, i Includes detail branches, context branches, and global branches; Feature integration: Dynamic convolution weights perform weighted summation of features. The lightweight design of the dynamic convolution module reduces computational costs while retaining the ability to adaptively adjust branch features. The calculation formula for feature integration is as follows: ; in: F DCM ( x, y ) refers to the feature map output by the dynamic convolution module, which represents the adaptive fusion result of multi-scale features; Refers to i The branches are at position ( x, y )’s eigenvalues; The specific process of the attention fusion module in step S35 is as follows: Spatial attention is calculated by a lightweight convolution with a kernel size of 1×1, and the calculation formula is as follows; ; in: F 空间注意力( x, y ) refers to the spatial attention weight map; Sigmoid Refers to the activation function, which normalizes the weight to the interval [0,1] to indicate the importance of spatial position; MaxPool Refers to the maximum pooling operation, which extracts the maximum value at each position in the feature map; AvgPool Refers to the average pooling operation, which extracts the mean of each position in the feature map; The channel attention mechanism is completed through global average pooling and single-layer perceptron, and the calculation formula is as follows: ; in: F 通道注意力 ( x, y ) refers to the channel attention vector; Linear Refers to a single-layer fully connected network; GAP Refers to global average pooling; F DCM Refers to the multi-scale features output by the dynamic convolution module; The calculation formula of the final fusion feature is as follows: ; in: F AFM ( x, y ) refers to the optimized feature map; By simplifying the computational path, the attention mechanism achieves a balance between low computational volume and efficient feature enhancement.

[0038] In this embodiment: for the multi-scale feature fusion module, that is, the MSFM module, the output volume is a multi-channel feature map that fuses multi-scale features. By fusing features from different scales, the model can capture detail information and global information and enhance the understanding of the input data. The output volume can maintain the same spatial dimension as the input feature map, or change due to operations such as pooling and upsampling. Due to the fusion of multi-scale features, the number of channels of the output volume usually increases, thereby containing more levels of feature information. The structure is as follows Figure 7 As shown in the figure, these feature information helps to improve the performance of the model in target detection tasks, provides richer high-level features, and helps the model to better perform tasks such as classification and positioning. In short, the output volume of the multi-scale feature fusion module effectively enhances the model's ability in complex multi-scale tasks through multi-scale feature fusion; By introducing a multi-scale feature fusion module, the output efficiency and accuracy of adenoids and nasopharyngeal airway volume calculation are significantly improved. The multi-scale feature fusion module can extract the global structural features and local detail information of adenoids under different receptive fields, and realize efficient segmentation and recognition of adenoids and nasopharyngeal airway regions. Through strip convolution operations and feature splicing mechanisms, the module can make full use of the multi-level information of the image, improve the computational efficiency of the model, and reduce the resource occupation of redundant calculations. In addition, the module performs adaptive weighted integration of features to highlight the significance of adenoids and nasopharyngeal airway regions, ensuring the accuracy and consistency of three-dimensional reconstruction and volume calculation results. The key point is to use the ability of multi-scale feature fusion to accelerate the adenoids and nasopharyngeal airway volume calculation process, including multi-scale feature extraction, feature fusion efficient algorithm design and volume fast output method based on this module. This method is suitable for a variety of medical imaging scenarios and provides efficient and reliable technical support for clinical diagnosis and treatment.

[0039] The calculation process of the estimated size of the adenoids in step S4 is as follows: The size of the adenoids was assessed by calculating the volume ratio of the adenoids to the nasopharyngeal cavity using the following formula: Adenoid size index = adenoid volume / (adenoid volume + nasopharyngeal airway volume).

[0040] In this embodiment: the calculation formula can be used to determine whether the adenoids completely block the nasopharyngeal cavity or the adenoids are of normal size, thereby quantifying the degree of airway obstruction caused by adenoid hypertrophy, providing an objective basis for clinical grading, i.e. mild, moderate and severe, replacing the subjective evaluation of traditional endoscopy or MRI, and achieving standardized and repeatable quantitative diagnosis.

[0041] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A CBCT three-dimensional image recognition method for adenoid size assessment, characterized in that: The following steps are involved: Step S1, after setting the boundary based on the CBCT image, the volume of the adenoids is calculated, step S1 includes: acquiring and preprocessing the CBCT image data, correcting the head position, determining the FH plane and the boundary of the adenoids and nasopharyngeal airway; Step S2, strengthening and correcting the boundaries of adenoids and nasopharyngeal airway based on strip convolution, step S2 includes: using a sine function to construct strip convolution, adenoids and nasopharyngeal airway segmentation, adenoids and nasopharyngeal airway boundary box extraction, adenoids and nasopharyngeal airway edge correction and adenoids and nasopharyngeal airway volume calculation; Step S3, multi-scale feature fusion module output volume, step S3 includes: input image preprocessing, low-resolution image input adaptation, lightweight multi-scale feature extraction, branch processing module, low computational specialization, dynamic convolution module, fast feature integration quantization dynamic convolution and attention fusion module; Step S4: Evaluate the size of the adenoids.

2. The CBCT three-dimensional image recognition method for adenoid size assessment according to claim 1, characterized in that: The specific process of acquiring and preprocessing the CBCT image data in step S1 is as follows: The patient's three-dimensional image data is obtained through the CBCT scanning equipment, and then the three-dimensional image data is cleaned, denoised and three-dimensionally reconstructed to generate a visual three-dimensional model.

3. The CBCT three-dimensional image recognition method for adenoid size assessment according to claim 2, characterized in that: The specific process of correcting the head position and determining the FH plane in step S1 is as follows: Based on the acquisition and preprocessing of CBCT image data, the visualization of the three-dimensional model output makes the connecting line of the left and right orbital points on the coronal plane parallel to the horizontal line, and the orbitoauricular plane on the sagittal plane parallel to the horizontal plane; The specific process of determining the adenoids and nasopharyngeal airway boundaries in step S1 is as follows: The boundaries of the adenoids and nasopharyngeal airway were determined in the three-dimensional model, including the sagittal anterior boundary, sagittal posterior boundary, sagittal superior boundary, sagittal inferior boundary, coronal anterior boundary, coronal lateral boundary, coronal posterior boundary, axial anterior boundary, axial posterior boundary and axial lateral boundary. A rectangular model was constructed in the three-dimensional space by marking the boundaries of the adenoids and nasopharyngeal airway in the coronal, sagittal and axial planes.

4. The CBCT three-dimensional image recognition method for adenoid size assessment according to claim 3, characterized in that: The specific process of constructing the strip convolution using the sine function in step S2 is as follows: Strip convolution is an operation that slides a one-dimensional sinusoidal strip convolution kernel along one direction of the image to extract edges with linear and periodic characteristics. Sine strip convolution combines the periodicity of the sine function and enhances the edges, textures, and periodic structures in the image through the periodic characteristics of the sine wave. When the input image is convolved, the sinusoidal strip convolution kernel will slide along one direction of the image and perform local weighted sum on the image, thereby extracting edge, texture and periodic structure information in the image; The sinusoidal strip convolution kernel is periodic, which can help highlight the periodically changing parts of the image and enhance the edge detection effect; When performing sinusoidal strip convolution, the sinusoidal strip convolution kernel slides in the three-dimensional CBCT image to calculate the weighted sum of each position in the image. The specific steps are as follows: For each three-dimensional pixel point, a sinusoidal strip convolution kernel is used to perform weighted calculation with the surrounding pixels of the point; The output is a three-dimensional pixel value in the image, representing the edge strength at that location.

5. The CBCT three-dimensional image recognition method for adenoid size assessment according to claim 4, characterized in that: The adenoids and nasopharyngeal airway edge correction in step S2 includes image input, edge expansion, a sinusoidal strip convolution kernel of the expanded image, and edge correction; Image input: The pixel value of the input original image is represented as I ( x,y ), whose dimensions are M×N , edge processing requires expansion of edge pixels: Extended Edge: In the original image pixel value I ( x,y ) is mirrored around, and the width of the expansion is w , w is the filter radius to obtain the expanded image pixel value I ext ( x,y ).

6. The CBCT three-dimensional image recognition method for adenoid size assessment according to claim 5, characterized in that: The specific process of calculating the adenoids and nasopharyngeal airway volumes in step S2 is as follows: Based on the 3D reconstructed model, the number of voxels in the adenoids and nasopharyngeal airway area was calculated. V voxel It is the basic unit in a three-dimensional image, and each voxel represents a spatial position in the image.

7. The CBCT three-dimensional image recognition method for adenoid size assessment according to claim 6, characterized in that: The specific process of the multi-scale feature fusion module output volume in step S3 is as follows: The specific process of the input image preprocessing and low-resolution image input adaptation in step S3 is as follows: Reduce the resolution of the input image through downsampling operations to reduce the computation volume from the source, which is done using strided convolution or bilinear interpolation; The specific process of lightweight multi-scale feature extraction in step S3 is as follows: The core of the design of the multi-scale feature fusion module is to reduce the volume of feature maps and extract rich multi-scale information at the same time, including multi-scale parallel paths and compression operations; The multi-scale parallel path is as follows: When extracting features of different receptive fields, redundancy is reduced by sharing the feature map basic calculation and adding lightweight convolution kernels; The compression operations are as follows: After multi-scale feature fusion, 1×1 convolution is used to compress the channels; The branch processing module in step S3 is specialized in the specific process of low computational complexity: Through branch processing for detail enhancement, context capture and global information, specialized lightweight processing methods are used to further optimize the amount of computation, including: detail enhancement branch, context capture branch and global information branch; Detail enhancement branch: Use a high-pass filter to simulate gradient operations through convolution, extract edge detail information, and compress the volume; Context Capture Branch: Use dilated convolution to expand the receptive field and avoid reducing the resolution of feature maps; Global information branch: Global average pooling is used to directly extract global features with low computational complexity. After each branch output is lightweight, the computational volume is further reduced.

8. The CBCT three-dimensional image recognition method for adenoid size assessment according to claim 7, characterized in that: The specific process of the multi-scale feature fusion module output volume in step S3 also includes: The specific process of the dynamic convolution module in step S3, the fast feature integration quantization dynamic convolution, is as follows: Weight generation: Using compressed feature input, the weight generation network is designed as a lightweight version, including a single-layer linear mapping; Feature integration: Dynamic convolution weights perform weighted summation of features. The lightweight design of the dynamic convolution module reduces computational costs while retaining the ability to adaptively adjust branch features. The specific process of the attention fusion module in step S3 is as follows: Spatial attention is calculated by lightweight convolution with a kernel size of 1×1; The channel attention mechanism is accomplished through global average pooling and a single-layer perceptron.

Images