An artificial intelligence-based nasopharyngeal mucosa hardness image processing method and system

Abstract

The application discloses a nasopharyngeal mucosa hardness image processing method and system based on artificial intelligence, relates to the technical field of image processing, and comprises the following steps: collecting nasopharyngeal image data, denoising, normalizing and unifying the coordinate system of the image, and constructing multi-modal image data; calculating the local displacement field of the multi-modal image data, constructing a micro-deformation field intensity map, and inversely calculating a local hardness map; calculating a global hardness map based on the local hardness map, calculating a hardness gradient map after filtering processing, and constructing a candidate boundary response map; adaptively thresholding the candidate boundary response map to extract a lesion boundary map; extracting hardness gradient information at the lesion boundary based on the lesion boundary map and the hardness gradient map, calculating lesion grade evaluation features, constructing a lesion grading score value, and dividing the lesion grade. The application can improve the accuracy of nasopharyngeal lesion boundary recognition and grading evaluation.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to an artificial intelligence-based method and system for image processing of nasopharyngeal mucosal stiffness. Background Technology

[0002] Early identification and grading of nasopharyngeal lesions are crucial for clinical diagnosis and treatment planning. Current nasopharyngeal endoscopy mainly relies on white light imaging and narrow-band imaging to observe mucosal surface morphology, color changes, and vascular texture characteristics. Although optical coherence tomography can provide some information on deep structures, differences exist between different imaging modalities in terms of viewing angle, resolution, acquisition time, and spatial location, making it difficult to directly form a unified and stable comprehensive judgment basis.

[0003] Currently, the identification of nasopharyngeal lesions in clinical practice largely relies on subjective judgment based on the doctor's experience. This is easily affected by image noise, local occlusion, changes in lighting, unclear lesion boundaries, and differences in the experience of different doctors. In particular, some early lesions only show subtle changes in mucosal stiffness, microdeformation ability, or local tissue mechanical properties, making it difficult to detect them in a timely and accurate manner based solely on surface image features.

[0004] Furthermore, existing image processing methods typically focus on color, texture, or morphological analysis of lesion areas, failing to adequately utilize the minute displacement changes, hardness distribution differences, and abrupt hardness gradient changes at lesion boundaries over continuous time intervals. This makes it difficult to achieve quantitative assessment of lesion boundaries and abnormality levels. Therefore, there is an urgent need for an image processing method and system that can integrate multimodal nasopharyngeal images, analyze local micro-deformation features, reflect mucosal hardness distribution, and assist in lesion boundary identification and grading evaluation. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] Therefore, this invention provides an artificial intelligence-based method and system for processing nasopharyngeal mucosal stiffness images, which solves the problem that existing nasopharyngeal endoscopy image analysis is unable to stably and quantitatively reflect changes in nasopharyngeal mucosal stiffness from multimodal images, resulting in insufficient accuracy in lesion boundary identification and lesion grade assessment.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides an artificial intelligence-based image processing method for nasopharyngeal mucosal stiffness, comprising, The nasopharyngeal image data was collected, denoised, normalized, and a unified coordinate system was established to construct multimodal image data. Calculate the local displacement field of multimodal image data, construct a micro-deformation field intensity map, and invert the local hardness map based on the intensity map; The global hardness map is calculated by weighting the local hardness map and then filtered. The hardness gradient map is calculated based on the filtered global hardness map, and candidate boundary response maps are constructed. Adaptive threshold segmentation is used to extract lesion boundary maps from the boundary response maps; Based on the lesion boundary and hardness gradient map, the hardness gradient information corresponding to the lesion boundary position is extracted, the lesion boundary gradient map is constructed, the features for evaluating the lesion grade are calculated, the lesion grading score is constructed, and the lesion grade is classified.

[0008] As a preferred embodiment of the artificial intelligence-based nasopharyngeal mucosal stiffness image processing method of the present invention, wherein: the collected nasopharyngeal image data is denoised and normalized, and a unified coordinate system is established, and the nasopharyngeal image data is constructed into multimodal image data, referring to the collection of nasopharyngeal data by an endoscopy data gateway deployed in the cloud. The collected data is then denoised to obtain... ; Calculate normalized data for the denoised image. ; Establish the mapping relationship between narrowband images and optical coherence tomography images to white light images. , ; A two-dimensional mapped image is obtained by projecting the optical coherence tomography image along the depth direction. ; The narrowband image and the projected optical coherence tomography image are mapped onto the white light image coordinate system, respectively. , .

[0009] As a preferred embodiment of the artificial intelligence-based nasopharyngeal mucosal stiffness image processing method of the present invention, wherein: the calculation of the local displacement field of multimodal image data refers to performing a two-dimensional Fourier transform on the local window of each modality image to obtain the frequency domain representation. ; For time The same position below Perform a Fourier transform on the local window image and take its conjugate complex number to obtain... ; based on and Calculate displacement-related response ; Displacement-related response Perform inverse Fourier transform to calculate displacement matching correlation peak distribution ; The translation of the peak position is taken as the local displacement. ; based on Constructing the displacement field .

[0010] As a preferred embodiment of the artificial intelligence-based nasopharyngeal mucosal stiffness image processing method of the present invention, wherein: the construction of a micro-deformation field intensity map, and the inversion of the local stiffness map based on the intensity map, refers to the process based on the displacement field. Constructing micro-deformation field intensity maps ; Based on local displacement and micro-deformation field intensity diagram Inverse local hardness map .

[0011] As a preferred embodiment of the artificial intelligence-based nasopharyngeal mucosal stiffness image processing method of the present invention, wherein: the step of calculating the global stiffness map based on the weighted local stiffness map and performing filtering processing refers to the step of weighting the local stiffness map of each modality... Weighted calculation of global hardness map ; Global hardness map conduct Gaussian filtering and smoothing process yields a smoothed global hardness map. .

[0012] As a preferred embodiment of the artificial intelligence-based nasopharyngeal mucosal hardness image processing method of the present invention, wherein: the step of calculating the hardness gradient map based on the filtered global hardness map and constructing the candidate boundary response map refers to the calculation of the hardness gradient map based on the smoothed global hardness map. Calculate the first-order rate of change in both the horizontal and vertical directions. , ; based on and Calculate the hardness gradient map ; Based on the smooth global hardness map Calculate the local hardness mean constraint ; Based on hardness gradient map Local hardness mean constraint and smooth global hardness map Output candidate boundary response map .

[0013] As a preferred embodiment of the artificial intelligence-based nasopharyngeal mucosal stiffness image processing method of the present invention, wherein: the adaptive threshold segmentation of the boundary response map to extract the lesion boundary map refers to the extraction of the lesion boundary map based on the candidate boundary response map. Calculate its mean and standard deviation; based on The mean and standard deviation are used to calculate the segmentation threshold of the candidate boundary response map. ; Based on segmentation threshold Extracting lesion boundary map .

[0014] As a preferred embodiment of the artificial intelligence-based nasopharyngeal mucosal stiffness image processing method of the present invention, wherein: the stiffness gradient information corresponding to the lesion boundary position is extracted based on the lesion boundary and stiffness gradient map, a lesion boundary gradient map is constructed, and feature indicators for evaluating the lesion grade are calculated based on... and Constructing a gradient map of lesion boundaries ; Based on lesion boundary gradient map Calculate the average gradient value of the lesion boundary ; Based on lesion boundary gradient map Calculate the average hardness value of the lesion area ; Based on lesion boundary gradient map Calculate the hardness dispersion within the lesion area .

[0015] As a preferred embodiment of the artificial intelligence-based nasopharyngeal mucosal stiffness image processing method of the present invention, wherein: the construction of lesion grading scores and the classification of lesion grades refer to... , , Normalization process yields , , And construct a lesion grading score. ; Based on lesion grading score The lesions are graded to obtain the lesion grading result L; The degree of abnormality of the lesion is determined based on the lesion grading result L.

[0016] Secondly, the present invention provides an artificial intelligence-based nasopharyngeal mucosal stiffness image processing system, comprising, The image acquisition and preprocessing module acquires image data of the nasopharynx through the endoscopic data gateway, performs noise reduction and normalization preprocessing on the data, and obtains multimodal image data based on the unified coordinate system of the preprocessed data. The displacement field calculation module performs frequency domain transformation based on multimodal data, and further obtains the displacement matching correlation peak distribution through inverse Fourier transform. Then, it takes the translation of the peak position as the local displacement and constructs the displacement field based on the local displacement. The local hardness map and global hardness map construction module calculates the micro-deformation intensity map based on the displacement field constructed in the previous module, then inverts the local hardness map based on the local displacement peak value, and weights the hardness map of each mode to obtain the global hardness map. The lesion boundary construction module is used to determine the lesion boundary. It calculates the hardness gradient based on the smoothed global hardness map, further calculates the hardness gradient map, constructs the boundary response map, and then performs threshold segmentation to extract the lesion boundary map. The lesion analysis module is used to classify lesions. It constructs a lesion boundary gradient map based on the lesion boundary map and hardness gradient map, and constructs a lesion grading scoring system based on the average gradient value of the lesion boundary, the average hardness value of the lesion area, and the hardness dispersion within the lesion area. It further sets thresholds to classify lesions into different grades.

[0017] The beneficial effects of this invention are as follows: This invention acquires multiple modal images, including white light imaging, narrowband imaging, and optical coherence tomography, through an endoscopic data gateway, and processes them using techniques such as denoising, registration, and normalization to ensure data consistency and accuracy. Then, the local displacement field is calculated using the phase correlation method to construct a micro-deformation field intensity map, and local hardness maps are obtained through Fourier transform and inverse Fourier transform. Finally, by combining the global hardness map and hardness gradient, lesion boundaries are accurately identified, and lesions are graded and evaluated using a hardness grading algorithm. This invention has significant effects in medical image analysis and can efficiently help doctors diagnose lesion areas. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart of the artificial intelligence-based nasopharyngeal mucosal stiffness image processing method in Example 1.

[0020] Figure 2 This is a structural diagram of the artificial intelligence-based nasopharyngeal mucosal stiffness image processing system in Example 1. Detailed Implementation

[0021] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0022] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0024] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides an artificial intelligence-based image processing method for nasopharyngeal mucosal stiffness, including the following steps: S1. Acquire nasopharyngeal images, preprocess the images, and unify the coordinate system of the preprocessed images to obtain multimodal image data; An endoscopy data gateway is deployed in the cloud to collect nasopharyngeal data. The endoscopy data gateway connects to the endoscopy terminal to collect data. The endoscopy terminal collects data such as white light imaging (WLI), narrow band imaging (NBI), optical coherence tomography (OCT), the precise position and angle of the endoscope in the nasopharynx, and timestamps. The collected data was denoised, and the denoised image is as follows:

[0025] in In the diagram, w corresponds to white light imaging (WLI), n corresponds to narrowband imaging (NBI), and o corresponds to optical coherence tomography (OCT). (x, y) represents the target pixel coordinates currently being calculated for denoising. Ω represents the neighborhood window, preferably 3×3 to suppress small random noise without excessively blurring the contours. |Ω| represents the number of pixels contained in the neighborhood window, and (u, v) represents the relative offset coordinates within the neighborhood window Ω, where u, v ∈ {-1, 0, 1}. Indicates the first The original image in neighborhood pixel coordinates The original pixel value at that location.

[0026] Perform normalization on the denoised image:

[0027] in For the first The minimum pixel value in a denoised image. For the first The maximum pixel value in a denoised image. represents the pixel values ​​after preprocessing the original image.

[0028] Using the coordinate system of the white light image as the reference coordinate system, and based on the position, angle, and timestamp information recorded during endoscopic acquisition, a mapping relationship between narrowband images and optical coherence tomography images and white light images is established:

[0029]

[0030] in, , , These represent the positional parameters used during the acquisition of white light images, narrowband images, and optical coherence tomography images, respectively. , , These represent the collected angle parameters, , , These represent timestamps, and These represent the mapping relationships between narrowband images and optical coherence tomography images relative to white light images, respectively.

[0031] For optical coherence tomography images, since they contain depth dimension information, they are first projected along the depth direction to obtain a two-dimensional mapped image:

[0032] in, This indicates the number of depth sampling layers in an optical coherence tomography (OCT) image. This represents a two-dimensional optical coherence tomography image after projection.

[0033] The narrowband image and the projected optical coherence tomography image are mapped onto the white light image coordinate system, respectively: By combining the white light image, the registered narrowband image, and the registered optical coherence tomography image, multimodal image data in a unified coordinate system is obtained. .

[0034] S2. Using multimodal image data of the same nasopharyngeal region at consecutive time points, the local displacement field is calculated using the phase correlation method, and a micro-deformation field intensity map is further constructed. Finally, the local hardness map is inverted based on the displacement field gradient. For each modality image First, capture the time. The same position in the image below Local window Window size is A square region (adjustable according to actual needs) is defined, and then a two-dimensional Fourier transform is performed on this local window image to obtain its frequency domain representation. By analyzing time... The same position below Perform a Fourier transform on the local window image and take its conjugate complex number to obtain... The product of the two values ​​and the modulus are normalized to obtain the displacement-related response. : in, and This represents the horizontal and vertical frequency indices in the frequency domain after a two-dimensional Fourier transform of a local window image. To prevent stable terms with a denominator of zero, the value is taken as... Prioritize This is sufficient to avoid numerical instability when the frequency domain components are extremely small, and at the same time, it will not significantly change the distribution of the shift-matching correlation peak.

[0035] The displacement matching correlation peak distribution is then obtained through inverse Fourier transform: in This refers to the horizontal and vertical displacement of the image at a later time relative to the image at a previous time. This represents the local correlation response obtained after the inverse Fourier transform, in the context of... Within this small local region centered on the image, the image from the next moment is translated horizontally and vertically relative to the image from the previous moment. Then, the degree of matching between the two.

[0036] And the translation of the peak position is taken as the local displacement:

[0037] Here, argmax does not take the maximum value itself, but rather "the argument that makes the function reach its maximum value," that is, it finds which one... let Reaching the maximum response value, The shift amount corresponding to obtaining the maximum response value .

[0038] The displacement field is constructed based on the translation of the maximum response value. Further construct the micro-deformation field intensity map :

[0039] in This is a scalar graph, representing points. The intensity of micro-deformation at each pixel location corresponds to only one numerical value representing the strength of the micro-deformation. Based on local displacement peak and micro-deformation field intensity diagram Inverse local hardness map :

[0040] in Indicates the local hardness value; This is the hardness scale factor, with a value of [value missing]. Preferred It can better preserve local hardness differences, while keeping the hardness distribution within a relatively stable range that is easy to display and analyze; This is the displacement gradient adjustment coefficient, with a value of [value missing]. Preferred This ensures that the displacement amplitude information and displacement gradient information are relatively balanced, reflecting both the overall degree of micro-deformation and enhancing the identification effect of areas with changes in local tissue mechanical properties. The stability coefficient takes a value of [value missing]. Preferred It can effectively suppress numerical anomalies without significantly weakening the model's ability to represent minute differences in hardness. It is an index for local deformation inhomogeneity, describing the uniformity and continuity of local deformation distribution in tissues.

[0041] S3, based on white light imaging Local hardness maps from narrowband imaging and optical coherence tomography are used to calculate global hardness maps. Based on the global hardness maps, hardness gradients are calculated, and regions with large gradients are identified by setting thresholds, thereby determining lesion boundaries. Using a weighted average method, the stiffness map of each mode is weighted according to its importance and reliability:

[0042] in It is the location in the global hardness map. The hardness value at that location; These are the weighting coefficients for the corresponding modes, representing the contributions of different modes. The preferred value is 0.4. White light imaging is the most commonly used imaging modality in endoscopy. It has high resolution and can clearly display the surface structure and tissue features of the nasopharynx. It is given a high weight so that it occupies a major position in the generation of hardness map and can maximize the preservation of the features of surface hardness differences. The preferred value is 0.3, which can fully utilize its advantages in vascular information without causing the hardness map results to rely excessively on low-resolution vascular information due to excessive weight. A value of 0.3 is preferred to ensure that OCT can supplement the hardness information of deep tissues without affecting the accuracy of surface structure.

[0043] Based on global hardness map To suppress high-frequency noise caused by local displacement estimation errors, modal registration micro-deviations, and numerical fluctuations during hardness inversion, while maintaining the spatial continuity of hardness distribution in the lesion area, the global hardness map was... Gaussian filtering smoothing process yields a smoothed global hardness map in pixel coordinates. Hardness value at :

[0044] in, Indicates the global hardness map in The original hardness value at the location; and Indicates the current pixel The relative offset of the center's neighborhood, and ; express The weight coefficients of the Gaussian convolution kernel at corresponding positions, the preferred Gaussian kernel G is... This conforms to the decay law of a two-dimensional Gaussian function, where the weight increases as the distance from the center increases, thus avoiding the problem of excessively blurred boundaries caused by uniform averaging.

[0045] Based on the smoothed global hardness map Identify lesion boundaries and calculate hardness gradients, then smooth the global hardness map. The first-order rate of change is calculated in both the horizontal and vertical directions to characterize the rate of change of hardness in space, thus obtaining the horizontal gradient. and vertical gradient :

[0046] in, Represents pixels along Change in hardness in a specific direction. Represents pixels along Change in hardness in a specific direction. , , and These represent the smoothness hardness values ​​at adjacent positions of the current pixel.

[0047] Based on horizontal gradient and vertical gradient Calculate the hardness gradient map to characterize the intensity of the overall hardness change at that location. :

[0048] To avoid misidentifying local noise points as lesion edges due to relying solely on gradient magnitude, tissue background adaptive modulation is applied to the gradient response. A local hardness mean constraint is introduced, and candidate boundary response maps are constructed. :

[0049]

[0050] in, Indicated by The average hardness value of the local neighborhood centered on the target; This represents a local background statistics window, with the preferred selection... ,Right now ; This indicates the number of pixels contained within the window; Indicates the candidate boundary response value; These are the gradient response weights, with values ​​ranging from 1 to 2. Preferred Since the lesion boundary mainly exhibits abrupt changes in hardness gradient, gradient information should be the primary consideration. The local hardness deviation weighting coefficient is set to a value of [value missing]. Preferred Supplementing information on regional hardness anomalies appropriately can prevent the pure gradient method from missing detections in areas with gradual changes in hardness.

[0051] Obtain candidate boundary response map Then, adaptive threshold segmentation is performed to extract the lesion boundary map. :

[0052]

[0053] in, The threshold value for segmenting the candidate boundary response map; express The mean value across the entire image; express The standard deviation across the entire image; This is the threshold adjustment coefficient, with a value of [value missing]. Preferred ;when When the threshold is small, the threshold is low, and boundary detection is more thorough, but it is easy to introduce noisy edges; when When the value is large, the threshold is high, which can reduce false boundaries but may cause weak boundary areas to be missed; when the pixel value is 1, it means that the point belongs to the lesion boundary, and when the pixel value is 0, it means that the point does not belong to the lesion boundary.

[0054] S4, Based on lesion boundary map and hardness gradient map Constructing a gradient map of lesion boundaries Then calculate the average gradient value of the lesion boundary. Average hardness value of the lesion area Hardness dispersion within the lesion area These three types of features are normalized and fused to construct a lesion grading score. by As a basis for locating the lesion area, As a basis for the intensity of hardness changes at the lesion boundary, hardness gradient information corresponding to the lesion boundary location is extracted to construct a lesion boundary gradient map. :

[0055] in, =1 indicates that the location belongs to the boundary of the lesion. =0 indicates that the location is not within the boundary of the lesion.

[0056] Based on lesion boundary gradient map The gradient response within the boundary region is statistically analyzed to calculate the average gradient value of the lesion boundary. :

[0057] in, This represents the sum of the gradient responses of all boundary pixels; Indicates the total number of boundary pixels; To prevent stable terms with a denominator of zero, the value is taken as... Preferred This prevents division by zero anomalies during the calculation process and does not significantly affect the statistical results in normal images.

[0058] To further reflect the overall stiffness level of the lesion area itself, in the lesion boundary map Under the constraint, the global smoothness values ​​within the boundary-enclosed area are statistically analyzed to calculate the average hardness value of the lesion area. :

[0059] in, Used to limit the statistical area, and only the locations marked as lesions at the boundaries are included in the calculation; Indicates the smoothing of the global hardness map at the pixel level. The hardness value at that location; As a stable term, its value and function are the same as in the previous formula.

[0060] To characterize whether the hardness distribution within the lesion is uniform, the hardness dispersion within the lesion area is calculated. :

[0061] Based on calculations , and After considering the three types of features, they are normalized and fused to construct a lesion grading score. : First of all , and Three types of feature normalization processing:

[0062]

[0063] The normalized three types of features are fused to construct a lesion grading score:

[0064] in , , , , , These represent the minimum and maximum values ​​of the corresponding indicators in the current data, respectively. , , Let represent the boundary gradient weight coefficient, the average hardness weight coefficient, and the dispersion weight coefficient, respectively, and satisfy . ,in Values Preferred Because stiffness mutations usually most directly reflect the transition strength between lesions and normal tissues, the boundary average gradient... It accounts for the main weight. Values Preferred The average hardness of the lesion area describes the overall stiffness level and has high auxiliary significance, thus it is assigned the second highest weight. Values Preferred The internal hardness dispersion of lesions mainly reflects regional heterogeneity and is a relatively supplementary feature.

[0065] Obtain lesion grading score Then, the lesions are classified into grades according to the preset grading thresholds to obtain the lesion grading results. :

[0066] in Indicates the lesion grading result, when When it indicates a low-grade abnormality lesion, When the time indicates an abnormal grade of lesion, This indicates a high-grade abnormality lesion; and These represent the grading thresholds, Values Preferred , Values Preferred When the score is low, it indicates that the boundary gradient is small, the overall hardness is low, and the interior is relatively uniform, corresponding to a low degree of abnormality. When the score is in the middle range, it indicates that the lesion has shown a certain degree of edge mutation and internal hardness abnormality, corresponding to a medium level. When the score is high, it indicates that the hardness mutation at the boundary is strong, the region is generally hard, and the internal heterogeneity is obvious.

[0067] This embodiment also provides an artificial intelligence-based nasopharyngeal mucosal stiffness image processing system, including: The image acquisition and preprocessing module acquires image data of the nasopharynx through the endoscopic data gateway, performs noise reduction and normalization preprocessing on the data, and obtains multimodal image data based on the unified coordinate system of the preprocessed data. The displacement field calculation module performs frequency domain transformation based on multimodal data, and further obtains the displacement matching correlation peak distribution through inverse Fourier transform. Then, it takes the translation of the peak position as the local displacement and constructs the displacement field based on the local displacement. The local hardness map and global hardness map construction module calculates the micro-deformation intensity map based on the displacement field constructed in the previous module, then inverts the local hardness map based on the local displacement peak value, and weights the hardness map of each mode to obtain the global hardness map. The lesion boundary construction module is used to determine the lesion boundary. It calculates the hardness gradient based on the smoothed global hardness map, further calculates the hardness gradient map, constructs the boundary response map, and then performs threshold segmentation to extract the lesion boundary map. The lesion analysis module is used to classify lesions. It constructs a lesion boundary gradient map based on the lesion boundary map and hardness gradient map, and constructs a lesion grading scoring system based on the average gradient value of the lesion boundary, the average hardness value of the lesion area, and the hardness dispersion within the lesion area. It further sets thresholds to classify lesions into different grades.

[0068] In summary, this invention acquires multiple modalities of images, including white light imaging, narrowband imaging, and optical coherence tomography, through an endoscopic data gateway. Data consistency and accuracy are ensured through denoising, registration, and normalization techniques. Next, the local displacement field is calculated using the phase correlation method to construct a micro-deformation field intensity map, and local hardness maps are obtained through Fourier transform and inverse Fourier transform. Finally, by combining the global hardness map and hardness gradient, lesion boundaries are accurately identified, and a hardness grading algorithm is used for lesion grading evaluation. This invention demonstrates significant effectiveness in medical image analysis and can efficiently assist doctors in diagnosing lesion areas.

[0069] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based image processing method for nasopharyngeal mucosal stiffness, characterized in that, include: The nasopharyngeal image data was collected, denoised, normalized, and a unified coordinate system was established to construct multimodal image data. Calculate the local displacement field of multimodal image data, construct a micro-deformation field intensity map, and invert the local hardness map based on the intensity map; The construction of the micro-deformation field intensity map, and the inversion of the local hardness map based on the intensity map, refers to the process based on the displacement field. Constructing micro-deformation field intensity maps ; Based on local displacement and micro-deformation field intensity diagram Inverse local hardness map ; The global hardness map is calculated by weighting the local hardness map and then filtered. The hardness gradient map is calculated based on the filtered global hardness map, and candidate boundary response maps are constructed. The process of calculating the global hardness map based on weighted local hardness maps and then performing filtering refers to calculating the local hardness map of each mode. Weighted calculation of global hardness map ; Global hardness map conduct Gaussian filtering and smoothing process yields a smoothed global hardness map. ; The calculation of the hardness gradient map based on the filtered global hardness map and the construction of candidate boundary response maps refer to the smoothing of the global hardness map. Calculate the first-order rate of change in both the horizontal and vertical directions. , ; based on and Calculate the hardness gradient map ; Based on the smooth global hardness map Calculate the local hardness mean constraint ; Based on hardness gradient map Local hardness mean constraint and smooth global hardness map Output candidate boundary response map ; Adaptive threshold segmentation is used to extract lesion boundary maps from the boundary response maps; Based on the lesion boundary and hardness gradient map, the hardness gradient information corresponding to the lesion boundary position is extracted, the lesion boundary gradient map is constructed, the features for evaluating the lesion grade are calculated, the lesion grading score is constructed, and the lesion grade is classified.

2. The image processing method for nasopharyngeal mucosal stiffness based on artificial intelligence as described in claim 1, characterized in that: The collected nasopharyngeal image data is denoised and normalized, and a unified coordinate system is established. The nasopharyngeal image data is then constructed into multimodal image data, which refers to the nasopharyngeal data collected by the cloud-deployed endoscopic data gateway. The collected data is then denoised to obtain... ; Calculate normalized data for the denoised image. ; Establish the mapping relationship between narrowband images and optical coherence tomography images to white light images. , ; A two-dimensional mapped image is obtained by projecting the optical coherence tomography image along the depth direction. ; The narrowband image and the projected optical coherence tomography image are mapped onto the white light image coordinate system, respectively. , .

3. The image processing method for nasopharyngeal mucosal stiffness based on artificial intelligence as described in claim 2, characterized in that: The calculation of the local displacement field of multimodal image data refers to performing a two-dimensional Fourier transform on a local window of each modality image to obtain a frequency domain representation. ; For time The same position below Perform a Fourier transform on the local window image and take its conjugate complex number to obtain... ; based on and Calculate displacement-related response ; Displacement-related response Perform inverse Fourier transform to calculate displacement matching correlation peak distribution ; The translation of the peak position is taken as the local displacement. ; based on Constructing the displacement field .

4. The method for image processing of nasopharyngeal mucosal stiffness based on artificial intelligence as described in claim 1, characterized in that: The adaptive threshold segmentation extraction of lesion boundary maps from boundary response maps refers to extracting lesion boundary maps based on candidate boundary response maps. Calculate its mean and standard deviation; based on The mean and standard deviation are used to calculate the segmentation threshold of the candidate boundary response map; Based on segmentation threshold Extracting lesion boundary map .

5. The method for image processing of nasopharyngeal mucosal stiffness based on artificial intelligence as described in claim 4, characterized in that: The process involves extracting hardness gradient information corresponding to the lesion boundary location based on the lesion boundary and hardness gradient map, constructing a lesion boundary gradient map, and calculating feature indicators for assessing the lesion grade. and Constructing a gradient map of lesion boundaries ; Based on lesion boundary gradient map Calculate the average gradient value of the lesion boundary ; Based on lesion boundary gradient map Calculate the average hardness value of the lesion area ; Based on lesion boundary gradient map Calculate the hardness dispersion within the lesion area .

6. The method for image processing of nasopharyngeal mucosal stiffness based on artificial intelligence as described in claim 5, characterized in that: The construction of lesion grading scores and the classification of lesion grades refer to... , , Normalization process yields , , And construct a lesion grading score. ; Based on lesion grading score The lesions are graded to obtain the lesion grading result L; The degree of abnormality of the lesion is determined based on the lesion grading result L.

7. An artificial intelligence-based nasopharyngeal mucosal stiffness image processing system, based on the artificial intelligence-based nasopharyngeal mucosal stiffness image processing method according to any one of claims 1 to 6, characterized in that, include: The image acquisition and preprocessing module acquires image data of the nasopharynx through the endoscopic data gateway, performs noise reduction and normalization preprocessing on the data, and obtains multimodal image data based on the unified coordinate system of the preprocessed data. The displacement field calculation module performs frequency domain transformation based on multimodal data, and further obtains the displacement matching correlation peak distribution through inverse Fourier transform. Then, it takes the translation of the peak position as the local displacement and constructs the displacement field based on the local displacement. The local hardness map and global hardness map construction module calculates the micro-deformation intensity map based on the displacement field constructed in the previous module, then inverts the local hardness map based on the local displacement peak value, and weights the hardness map of each mode to obtain the global hardness map. The lesion boundary construction module is used to determine the lesion boundary. It calculates the hardness gradient based on the smoothed global hardness map, further calculates the hardness gradient map, constructs the boundary response map, and then performs threshold segmentation to extract the lesion boundary map. The lesion analysis module is used to classify lesions. It constructs a lesion boundary gradient map based on the lesion boundary map and hardness gradient map, and constructs a lesion grading scoring system based on the average gradient value of the lesion boundary, the average hardness value of the lesion area, and the hardness dispersion within the lesion area. It further sets thresholds to classify lesions into different grades.

Images