Soft tissue self-adaptive enhancement method and enhancement device of skull side position film

Abstract

The invention provides a soft tissue adaptive enhancement method and enhancement device for a skull side position film. The method comprises the following steps: extracting key anatomical points related to a mouth-nose area in the skull side position film by adopting a UNet-based feature point recognition model; extracting a mouth-nose area in the skull side position film according to the key anatomical points, and performing local enhancement; and carrying out weighted fusion on the enhanced mouth and nose area image and the original image of the skull side position film to form a soft tissue enhanced image of the skull side position film. According to the invention, through the feature point identification model, automatic and accurate positioning of key feature points is realized; independent image enhancement processing is carried out on the segmented mouth and nose areas, so that the contrast ratio, the definition and the detail visibility of the key areas can be remarkably improved on the premise of not influencing other areas of the image; and a fusion technology based on an adaptive weight map is adopted, so that smooth transition of the enhanced region and the original image at the boundary is ensured, and the overall consistency and visual authenticity of the image are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to a soft tissue adaptive enhancement method of a lateral cephalogram and a soft tissue adaptive enhancement device of a lateral cephalogram. BACKGROUND

[0002] The lateral cephalogram is an indispensable X-ray diagnostic image in the fields of orthodontics, craniofacial surgery, etc., and the oral and nasal soft tissue region, as a core component of facial aesthetics and function, is of great importance to precise diagnosis, treatment planning and efficacy evaluation. However, this region often appears in a low gray interval in X-ray imaging, with inherent defects of low contrast and blurred details.

[0003] Currently, global image enhancement methods (such as histogram equalization) are generally used to address the above problems, which can improve the overall contrast, but will also cause overexposure in high gray regions (such as the skull), and cannot achieve targeted local enhancement.

[0004] In related technologies, although there are methods of local enhancement using threshold segmentation, the segmentation accuracy is insufficient and the robustness is poor due to the individual differences and image conditions of the skull structure, making it difficult to accurately separate the oral and nasal soft tissue region, and this method relies on the manual delineation of the region of interest by doctors, which is tedious and inefficient. SUMMARY

[0005] To solve the above technical problems, the present application provides a soft tissue adaptive enhancement method of a lateral cephalogram.

[0006] The present application also provides a soft tissue adaptive enhancement device of a lateral cephalogram.

[0007] The technical solutions adopted by the present application are as follows: The first aspect of the present application provides a soft tissue adaptive enhancement method of a lateral cephalogram, including the following steps: using a feature point recognition model based on UNet (a convolutional neural network-based image segmentation model) to extract key anatomical points related to the oral and nasal region in the lateral cephalogram; extracting the oral and nasal region in the lateral cephalogram according to the key anatomical points, and performing local enhancement on the extracted oral and nasal region; and performing weighted fusion of the enhanced oral and nasal region image and the original image of the lateral cephalogram to form a soft tissue enhanced image of the lateral cephalogram.

[0008] The soft tissue adaptive enhancement method of a lateral cephalogram described above also has the following additional technical features: According to one embodiment of the present application, the UNet-based feature point recognition model is generated in the following manner: collect head lateral radiograph samples, and perform expert annotation of key anatomical points on the samples; pre-process and perform data enhancement processing on the annotated samples; convert each annotated key anatomical point in the samples into a Gaussian heat map form; divide the samples into a training set, a validation set, and a test set; construct a UNet-based feature point recognition model, and sample the training set, the validation set, and the test set to train the feature point recognition model, wherein a binary cross-entropy loss combined with a Sigmoid (a type of S-shaped function) activation is used as the objective function during training, an Adam (Adaptive Moment Estimation, an optimization algorithm combining a momentum mechanism and an adaptive learning rate) optimizer is selected, and a cosine annealing strategy is used to dynamically adjust the learning rate.

[0009] According to one embodiment of the present application, the oral-nasal region in the head lateral radiograph is extracted according to the key anatomical points, specifically including: connecting the lateral contour points and the medial contour points in the key anatomical points in sequence according to the anatomical order to form an inner contour line and an outer contour line; extending the end points of the inner contour line and the outer contour line at the two ends to the image boundary along the vertical direction, respectively, to construct a closed region on the image plane; taking the midpoint between the starting points of the inner contour line and the outer contour line as a seed point, performing a filling operation on the closed region to generate an initial position template; performing morphological dilation processing on the initial position template; applying the dilated initial position template as a spatial mask to the original head lateral radiograph to realize extraction of the oral-nasal region.

[0010] According to one embodiment of the present application, the extracted oral-nasal region is locally enhanced, specifically including: normalizing the gray values of all pixels in the extracted oral-nasal region to the [0, 1] interval; performing dynamic Gamma correction on the normalized gray values; expanding the oral-nasal region mask by a specific radius, counting the maximum value max and the minimum value min of the pixels in the expanded region, linearly mapping the corrected oral-nasal region pixel values to the [min, max] interval to form an enhanced oral-nasal region image.

[0011] According to one embodiment of the present application, the enhanced oral-nasal region image and the original image of the lateral cephalogram are weightedly fused to form a soft tissue enhanced image of the lateral cephalogram, specifically comprising: expanding the region boundary of the binary mask of the oral-nasal region through morphological dilation operation, wherein the dilation radius is adaptively adjusted according to the image resolution; performing anisotropic Gaussian blur processing based on the local texture features of the image on the expanded mask to generate a weight distribution map α with directional transition characteristics; based on the generated weight distribution map α, the enhanced oral-nasal region image and the original image of the lateral cephalogram are weightedly fused to form a soft tissue enhanced image of the lateral cephalogram; wherein the following formula is used for weighted fusion: ; wherein, is the soft tissue enhanced image of the lateral cephalogram, and α is the weight distribution map, is the enhanced oral-nasal region image, is the original image of the lateral cephalogram.

[0012] The second aspect embodiment of the present application proposes a soft tissue adaptive enhancement device for a lateral cephalogram, comprising: an extraction module, which is used to extract key anatomical points related to the oral-nasal region in the lateral cephalogram by using a UNet-based feature point recognition model; an enhancement module, which is used to extract the oral-nasal region in the lateral cephalogram according to the key anatomical points and perform local enhancement on the extracted oral-nasal region; and a fusion module, which is used to weightedly fuse the enhanced oral-nasal region image and the original image of the lateral cephalogram to form a soft tissue enhanced image of the lateral cephalogram.

[0013] The soft tissue adaptive enhancement device for a lateral cephalogram described above further comprises the following additional technical features: According to one embodiment of the present application, the UNet-based feature point recognition model is generated in the following manner: collect lateral cephalogram samples and perform expert annotation of key anatomical points on the samples; pre-process and data enhance the annotated samples; convert each annotated key anatomical point in the sample into a Gaussian heat map form; divide the samples into a training set, a validation set and a test set; construct a UNet-based feature point recognition model, and sample the training set, the validation set and the test set to train the feature point recognition model, wherein a binary cross-entropy loss combined with a Sigmoid activation is used as the objective function during training, an Adam optimizer is selected, and a cosine annealing strategy is used to dynamically adjust the learning rate.

[0014] According to one embodiment of the present application, the enhancement module is specifically used for: connecting the lateral contour points and the medial contour points in the key anatomical points in sequence respectively according to the anatomical order to form an inner contour line and an outer contour line; extending the end points at the two ends of the inner contour line and the outer contour line to the image boundary along the vertical direction respectively to construct a closed region on the image plane; taking the midpoint between the starting points of the inner contour line and the outer contour line as a seed point, performing a filling operation on the closed region to generate an initial position template; performing a morphological dilation processing on the initial position template; taking the dilated initial position template as a spatial mask and applying it to the original lateral cephalogram to realize extraction of the oronasal region.

[0015] According to one embodiment of the present application, the enhancement module is further used for: normalizing the gray values of all pixels in the extracted oronasal region to the interval [0, 1]; performing dynamic Gamma correction on the normalized gray values; expanding the oronasal region mask by a specific radius, counting the maximum value max and the minimum value min of the pixels in the expanded region, linearly mapping the corrected oronasal region pixel values to the interval [min, max] to form an enhanced oronasal region image.

[0016] According to one embodiment of the present application, the fusion module is specifically used for: expanding the region boundary of the binary mask of the oronasal region through a morphological dilation operation, wherein the dilation radius is adaptively adjusted according to the image resolution; performing an anisotropic Gaussian blur processing on the expanded mask based on the local texture features of the image to generate a weight distribution graph alpha with directional transition characteristics; based on the generated weight distribution graph alpha, performing weighted fusion on the enhanced oronasal region image and the original image of the lateral cephalogram to form a soft tissue enhanced image of the lateral cephalogram; wherein the following formula is used for weighted fusion: ; wherein, is the soft tissue enhanced image of the lateral cephalogram, alpha is the weight distribution graph, is the enhanced oronasal region image, is the original image of the lateral cephalogram.

[0017] The present application has the following advantages: The present application realizes automatic and accurate positioning of key feature points through a feature point recognition model, overcoming the drawbacks of low efficiency, strong subjectivity and easy variation of traditional manual labeling; By performing independent image enhancement processing on the segmented oronasal region, the contrast, clarity and detail visibility of the key region can be significantly improved without affecting other regions of the image; The fusion technology based on adaptive weight graph ensures smooth transition of the enhanced region and the original image at the boundary, ensuring the overall consistency and visual authenticity of the image.

[0018] The mouth and nose regions are automatically and accurately segmented by constructing a closed region and generating a position template using the inner and outer contour points, thereby laying a reliable foundation for subsequent targeted image processing. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a flowchart of a soft tissue adaptive enhancement method for a lateral cephalogram according to an embodiment of the present application; Figure 2 is a schematic diagram of outer contour points according to an embodiment of the present application; Figure 3 is a schematic diagram of inner contour points according to an embodiment of the present application; Figure 4 is a block schematic diagram of a soft tissue adaptive enhancement device for a lateral cephalogram according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the scope of the present application.

[0021] Figure 1 is a flowchart of a soft tissue adaptive enhancement method for a lateral cephalogram according to an embodiment of the present application, as shown in Figure 1 , the method comprises the following steps: S1, a feature point recognition model based on UNet is used to extract key anatomical points related to the mouth and nose regions in the lateral cephalogram.

[0022] In the embodiments of the present application, the mouth and nose regions are accurately extracted, and the following 16 key anatomical points can be located as feature points, and a region segmentation boundary is constructed by connecting lines, as shown in Figure 2 , Figure 3 The key anatomical points are divided into two categories: inner contour points and outer contour points. The outer contour points include the glabella point (G), the soft tissue nasi root point (Ns), the nasal tip point (Prn), the subnasale point (Sn), the upper lip protuberance point (UL), the stomion inferior point (Stoms), the lower lip protuberance point (LL), the lower alveolar ridge point (B), the soft tissue menton point (Pos), the soft tissue gnathion point (Gs), and the soft tissue mentomeatal point (Mes). The inner contour points include the nasi root point (N), the anterior nasal spine point (ANS), the upper alveolar ridge point (Pr), the gnathion point (Gn), and the cervical point (C).

[0023] In an embodiment of the present application, the following method can be used to generate a UNet-based feature point recognition model (steps S101-S105): S101, collect head lateral film samples, and perform expert annotation of key anatomical points on the samples.

[0024] Specifically, 1000 sample head lateral films can be collected, and two experienced clinical experts can independently annotate 16 key anatomical points according to the definitions shown in Figure 2 、 3 To reduce subjective differences in annotation, the average coordinates of the annotation results of the two experts are used as the final true annotation values of each key anatomical point, thereby ensuring the consistency and reliability of the training labels.

[0025] S102, pre-process and data enhance the annotated samples.

[0026] Specifically, the original image is converted to a grayscale image for pre-processing. Due to the limited amount of medical image data, various data enhancement strategies can be used to expand the training samples, thereby improving the generalization ability of the model and reducing the risk of overfitting. The enhancement methods include: random cropping of the image, with a cropping ratio of 0% to 15% in the up, down, left and right directions; adding a black border to the left and right sides of the image, with an expansion ratio of 0% to 100%; random rotation of the image, with an angle range of -5° to 5°; random Gamma transformation, with Gamma values selected from the set {0.2, 0.4, 0.67, 1, 1.5, 2.5, 5.0}, and ensuring that the selection probability of Gamma value 1 (i.e. no transformation) is 50%.

[0027] After enhancement, the sample images are uniformly scaled to 512x512 pixels to adapt to the network input size.

[0028] S103, convert each annotated key anatomical point in the sample to a Gaussian heat map form.

[0029] Specifically, to convert the original coordinate regression problem into a more stable probability distribution prediction task, the present application converts each annotated feature point into a Gaussian heat map form. For any pixel point coordinate in the sample image, its heat map response value is calculated by a two-dimensional Gaussian function:

[0030] wherein, represents the response value at coordinate in the heat map, represents the actual coordinate of the key anatomical point, is the target scale adaptive variance, which controls the distribution range of the Gaussian kernel in the heat map. Since there are 16 key anatomical points in each sample lateral skull radiograph, the sample image is converted into a 16x512x512 heat map label.

[0031] S104, the sample is divided into a training set, a validation set and a test set.

[0032] Specifically, the pre-processed sample lateral skull radiograph can be divided into 700 training sets, 200 validation sets and 100 test sets.

[0033] S105, a feature point recognition model based on UNet is constructed, and the training set, the validation set and the test set are sampled to train the feature point recognition model, wherein a binary cross entropy loss combined with Sigmoid activation is used as the objective function during training, an Adam optimizer is selected, and the learning rate is dynamically adjusted using a cosine annealing strategy.

[0034] Specifically, the network of the feature point recognition model based on UNet includes five layers of encoders and five layers of decoders. Each level of the encoder is composed of a down-sampling module and a convolution module for extracting multi-scale features; each level of the decoder is composed of an up-sampling module and a convolution module for restoring spatial resolution. The encoder and the decoder are connected by a skip connection between the corresponding levels to fuse shallow details and deep semantic information.

[0035] During the training process, the training data is input in batches with a batch size of 6. A binary cross entropy loss combined with Sigmoid activation is used as the objective function: ; wherein LOSS represents the loss function, y is the true heat map value, is the model prediction value. The Adam optimizer is selected for optimization, and the learning rate is dynamically adjusted using a cosine annealing strategy: ; wherein, and represent the maximum and minimum values of the learning rate, is the current training period, is the total training period. The training is performed for 1000 iterations, and the model parameters with the minimum loss on the validation set are saved as the final recognition model.

[0036] Through the above trained model, the coordinates of the corresponding key anatomical points can be output by inputting the lateral skull radiograph.

[0037] In summary, first, enough head lateral radiographs are collected as samples and key anatomical points are labeled to construct a labeled dataset; then, the samples in the dataset are preprocessed and data augmented to improve the generalization ability of the model; then, the coordinate values of the key anatomical points are converted into heat map form, so as to convert the coordinate regression task into a probability distribution prediction task; then, the processed dataset is divided into a training set, a validation set and a test set, and a deep learning model is constructed and trained based on the divided dataset, so that the model learns the mapping relationship from the input image to the key anatomical point heat map; finally, the head lateral radiograph to be detected is input into the trained model, and the probability heat map corresponding to each anatomical point is output and decoded into coordinate position, so as to realize the accurate positioning of the key anatomical point.

[0038] S2, extracting the oral-nasal region in the head lateral radiograph according to the key anatomical point, and performing local enhancement on the extracted oral-nasal region.

[0039] In an embodiment of the present application, the oral-nasal region in the head lateral radiograph is extracted according to the key anatomical point, specifically including the following S201-S205: S201, connecting the lateral contour points and the medial contour points in the key anatomical points in sequence according to the anatomical order to form an inner contour line and an outer contour line.

[0040] S202, extending the end points of the inner contour line and the outer contour line at the two ends to the image boundary along the vertical direction, so as to construct a closed region on the image plane.

[0041] S203, taking the midpoint between the starting points of the inner contour line and the outer contour line as a seed point to perform a filling operation on the closed region to generate an initial position template.

[0042] S204, performing morphological dilation processing on the initial position template.

[0043] S205, applying the dilated initial position template as a spatial mask to the original head lateral radiograph to realize extraction of the oral-nasal region.

[0044] Specifically, the extraction of the oral-nasal region first connects the lateral contour points and the medial contour points in sequence according to the anatomical order. Then, the end points of the inner and outer contour lines at the two ends are extended to the image boundary along the vertical direction, so as to construct a closed region on the image plane. Taking the midpoint between the starting points of the inner and outer contour lines as a seed point, a filling operation is performed on the closed region to generate an initial position template. In order to further ensure that the oral-nasal region is completely covered, morphological dilation processing is performed on the template to expand its effective area range. Finally, the dilated position template is applied as a spatial mask to the original head lateral radiograph to realize accurate extraction of the oral-nasal region.

[0045] In one embodiment of the present invention, local enhancement is performed on the extracted nasal and oral region, specifically including the following steps S206-S208: S206, normalize the gray values ​​of all pixels in the extracted mouth and nose area to the [0, 1] interval.

[0046] S207 performs dynamic Gamma correction on the normalized grayscale values.

[0047] Specifically, dynamic gamma correction can solve the problem of uneven enhancement effect of traditional gamma correction on different brightness areas within a region, and introduces a dynamic exponential adjustment strategy based on the pixel's own brightness: ; Where k is the adjustment coefficient, and P is the normalized gray value;

[0048] in, This is the grayscale value after Gamma correction.

[0049] Dynamic gamma correction can be applied to dark areas ( Small) apply stronger enhancement ( Large), while for bright areas ( If it is large, then protective adjustments will be made. (Small), thereby effectively widening the local dynamic range and preventing overexposure.

[0050] S208 expands the mask of the mouth and nose region by a specific radius, counts the maximum and minimum pixel values ​​of the original image within the expanded region, and linearly maps the corrected pixel values ​​of the mouth and nose region to the [min, max] interval to form an enhanced image of the mouth and nose region.

[0051] Specifically, local contrast matching is performed by expanding the mask of the mouth and nose region by a specific radius, and calculating the maximum and minimum pixel values ​​(max and min) of the original image within this expanded region. The enhanced pixel values ​​of the mouth and nose region are then linearly mapped to the calculated [min, max] interval, so that the contrast of the enhanced region blends naturally with that of the surrounding tissue.

[0052] S3, the enhanced image of the mouth and nose region is weighted and fused with the original image of the lateral cephalometric radiograph to form a soft tissue enhanced image of the lateral cephalometric radiograph.

[0053] In a specific embodiment of the present invention, the enhanced image of the mouth and nose region is weighted and fused with the original image of the lateral cephalometric radiograph to form a soft tissue enhanced image of the lateral cephalometric radiograph, specifically including the following steps S301-S303: S301, the binary mask of the mouth-nose region is expanded by a morphological dilation operation, wherein the dilation radius is adaptively adjusted according to the image resolution.

[0054] Specifically, the binary mask of the mouth-nose region is first expanded by a morphological dilation operation, and the dilation radius is adaptively adjusted according to the image resolution, and the calculation formula is: ; Wherein, r is the dilation radius, W and H represent the width and height of the image respectively.

[0055] S302, the expanded mask is processed by an anisotropic Gaussian blur based on the local texture features of the image to generate a weight distribution map alpha with directional transition characteristics.

[0056] Specifically, the expanded mask is processed by an anisotropic Gaussian blur based on the local texture features of the image to generate a weight distribution map alpha with directional transition characteristics, which fully considers the anatomical structure characteristics of the mouth-nose region, so that the weight distribution of the transition region is consistent with the physiological structure of the tissue.

[0057] S303, based on the generated weight distribution map alpha, the enhanced mouth-nose region image and the original image of the lateral cephalogram are weighted and fused to form a soft tissue enhanced image of the lateral cephalogram, and the following formula is used for weighted fusion: ; Wherein, is the soft tissue enhanced image of the lateral cephalogram, alpha is the weight distribution map, is the enhanced mouth-nose region image, is the original image of the lateral cephalogram.

[0058] Therefore, the weighted fusion process ensures smooth transition of the enhanced region and the original image at the boundary, realizing seamless connection of the visual effect.

[0059] In summary, according to the soft tissue adaptive enhancement method of the lateral cephalogram of the embodiment of the present application, the automatic and accurate positioning of the key feature points is realized through the feature point recognition model, which overcomes the drawbacks of low efficiency, strong subjectivity and easy variation of traditional manual labeling; through independent image enhancement processing of the segmented mouth-nose region, the contrast, clarity and detail visibility of the key region can be significantly improved without affecting other regions of the image; the fusion technology based on adaptive weight map is adopted to ensure smooth transition of the enhanced region and the original image at the boundary, and the overall consistency and visual authenticity of the image are ensured. The closed region is constructed by using the inner and outer contour points, and the position template is generated, realizing the automatic and accurate segmentation of the mouth-nose region, and laying a reliable foundation for subsequent targeted image processing.

[0060] Corresponding to the soft tissue adaptive enhancement method of the lateral cephalogram, the application also provides a soft tissue adaptive enhancement device of the lateral cephalogram. The device embodiment of the application corresponds to the method embodiment described above. For details not disclosed in the device embodiment, please refer to the method embodiment described above. The application will not be described again.

[0061] Figure 4 The figure is a block diagram of the soft tissue adaptive enhancement device of the lateral cephalogram according to an embodiment of the application. As shown in the figure, the device comprises an extraction module 1, an enhancement module 2 and a fusion module 3. Figure 4

[0062] The extraction module 1 is used to extract key anatomical points related to the oral-nasal region in the lateral cephalogram by using a UNet-based feature point recognition model. The enhancement module 2 is used to extract the oral-nasal region in the lateral cephalogram according to the key anatomical points and perform local enhancement on the extracted oral-nasal region. The fusion module 3 is used to perform weighted fusion on the enhanced oral-nasal region image and the original image of the lateral cephalogram to form a soft tissue enhanced image of the lateral cephalogram.

[0063] According to an embodiment of the application, the UNet-based feature point recognition model is generated in the following manner: collect lateral cephalogram samples and perform expert annotation of key anatomical points on the samples; perform preprocessing and data enhancement processing on the annotated samples; convert each annotated key anatomical point in the samples into a Gaussian heat map; divide the samples into a training set, a validation set and a test set; construct a UNet-based feature point recognition model, and sample the training set, the validation set and the test set to train the feature point recognition model. During the training, the binary cross-entropy loss combined with the Sigmoid activation is used as the objective function, the Adam optimizer is selected, and the cosine annealing strategy is used to dynamically adjust the learning rate.

[0064] According to an embodiment of the application, the enhancement module 2 is specifically configured to: connect the lateral contour points and the medial contour points in the key anatomical points in sequence according to the anatomical order to form an inner contour line and an outer contour line; extend the end points of the inner contour line and the outer contour line at the two ends to the image boundary along the vertical direction, so as to construct a closed region on the image plane; take the midpoint between the starting points of the inner contour line and the outer contour line as a seed point to perform a filling operation on the closed region to generate an initial position template; perform morphological dilation processing on the initial position template; apply the dilated initial position template as a spatial mask to the original lateral cephalogram to realize extraction of the oral-nasal region.

[0065] ​According to one embodiment of the present application, the enhancement module 2 is further configured to normalize the gray values of all pixels in the extracted oronasal region to the interval [0, 1], perform dynamic Gamma correction on the normalized gray values, expand the oronasal region mask by a specific radius, count the maximum value max and the minimum value min of the pixels in the expanded region, linearly map the corrected oronasal region pixel values to the interval [min, max] to form an enhanced oronasal region image.

[0066] According to one embodiment of the present application, the fusion module 3 is specifically configured to: for the binary mask of the oronasal region, expand the region boundary through a morphological dilation operation, wherein the dilation radius is adaptively adjusted according to the image resolution; perform anisotropic Gaussian blur processing based on the local texture features of the image on the expanded mask to generate a weight distribution map a with directional transition characteristics; based on the generated weight distribution map a, perform weighted fusion of the enhanced oronasal region image and the original image of the lateral cephalogram to form a soft tissue enhanced image of the lateral cephalogram; wherein the following formula is used for weighted fusion: ; wherein, is the soft tissue enhanced image of the lateral cephalogram, a is the weight distribution map, is the enhanced oronasal region image, is the original image of the lateral cephalogram.

[0067] In summary, according to the soft tissue adaptive enhancement device for lateral cephalogram of the present application, the automatic and accurate positioning of the key feature points is realized through the feature point recognition model, which overcomes the drawbacks of low efficiency, strong subjectivity and easy variation of traditional manual labeling; through independent image enhancement processing of the segmented oronasal region, the contrast, clarity and detail visibility of the key region can be significantly improved without affecting other regions of the image; the fusion technology based on adaptive weight map is adopted to ensure smooth transition of the enhanced region and the original image at the boundary, and to ensure the overall consistency and visual authenticity of the image. The closed region is constructed by using the inner and outer contour points, and the position template is generated, which realizes the automatic and accurate segmentation of the oronasal region and lays a reliable foundation for subsequent targeted image processing.

[0068] In the description of the application, reference to "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" means that a particular feature, structure, material, or characteristic being described is included in at least one embodiment or example of the application. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment or example. Furthermore, the described specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples. Moreover, the usage of the terms "first", "second" or "third" does not limit the quantity or order of the specific features, structures, materials or characteristics, but merely distinguishes one from another. Thus, features, structures, materials or characteristics referred to by the same or different terms in different places of the specification have no significance in terms of hierarchy or importance.

[0069] Furthermore, the terms "first", "second", or the like, merely denote different instances of a similar feature, structure, material or characteristic, and do not imply a relative importance or a specific order. Thus, features, structures, materials or characteristics defined with "first", "second" or the like can include at least one of the features, structures, materials or characteristics. In the description of the application, the meaning of "a plurality" is at least two, for example two, three or more, unless otherwise expressly specified.

[0070] Any process or method descriptions or blocks in flow charts or otherwise described herein represent embodiments which can be managed as one or more modules, segments, or portions of code which include one or more steps for implementing specific logic functions or steps, and the terms in the specification have their ordinary technical and / or preferred definitions that the skilled artisan will be able to determine in context of the relevant art, and alternative implementations are possible. The various embodiments or examples of the application can be further understood by the following examples.

[0071] The logic and / or steps represented in flow diagrams or otherwise described herein, for example, can be considered as a sequence of instructions to implement logic functions, and can be embodied in any computer-readable medium for use by an instruction execution system, apparatus, or device, such as a computer-based system, processor- containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. In the context of this specification, a "computer-readable medium" can be any means that can contain, store, communicate, propagate or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can be a machine-readable storage device (e.g., magnetic, optical or other) a machine-readable storage diskette (e.g., floppy disk, optical disk, etc.), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), optical fibers, and a portable compact disc read-only memory (CDROM). Further, the computer-readable medium can even be paper or other suitable medium upon which the program is printed, as the program can be electronically captured, for example via the optical scanner of a device or other electronic capture device, and then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and stored in a computer memory.

[0072] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. As such, in some embodiments, the steps or methods can be implemented in a combination of hardware and software. If implemented in hardware, as in another embodiment, any of the above techniques can be implemented with or without accompanying software or in software / firmware alone. The techniques can be implemented using either or both of the following technologies, which are well known in the art: (a) a combination of custom hard-wired circuitry and custom circuitry both designed to perform the techniques, and (b) a combination of custom hard-wired circuitry and one or more identified

[0073] Those of skill in the art will understand that the steps or methods carried out in the above-described embodiments can be carried out by program instructions stored in a computer-readable storage medium and executed by a processor. The program instructions can be implemented in any of the above technologies, or in any other technology.

[0074] In addition, each function unit in each embodiment of the present application can be integrated in one processing module, or each unit can be physically present separately, or two or more units can be integrated in one module. The integrated module can be realized in the form of hardware or in the form of a software function module. When the integrated module is realized in the form of a software function module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0075] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for soft tissue adaptive enhancement of lateral cephalometric radiographs, characterized in that, Includes the following steps: Key anatomical points related to the oral and nasal regions in the lateral cephalometric radiographs were extracted using a UNet-based feature point recognition model. Based on the key anatomical points, the oral and nasal regions in the lateral cephalometric radiographs were extracted, and the extracted oral and nasal regions were locally enhanced. The enhanced image of the mouth and nose region is weighted and fused with the original image of the lateral cephalometric radiograph to form a soft tissue enhanced image of the lateral cephalometric radiograph.

2. The method for soft tissue adaptive enhancement of lateral cephalometric radiographs according to claim 1, characterized in that, The UNet-based feature point recognition model is generated using the following method: Collect lateral cephalometric radiographs and annotate key anatomical points on the radiographs with expert annotations. Preprocessing and data augmentation are performed on the labeled samples; Convert each labeled key anatomical point in the sample into a Gaussian heatmap. The samples are divided into training set, validation set and test set; A feature point recognition model based on UNet is constructed. The model is trained by sampling the training set, validation set, and test set. During training, a binary cross-entropy loss with Sigmoid activation is used as the objective function, the Adam optimizer is selected, and the learning rate is dynamically adjusted using a cosine annealing strategy.

3. The method for soft tissue adaptive enhancement of lateral cephalometric radiographs according to claim 1, characterized in that, Based on the key anatomical points, the oral and nasal regions in the lateral cephalometric radiographs were extracted, specifically including: Connect the outer contour points and the inner contour points of the key anatomical points in sequence according to the anatomical order to form the inner contour line and the outer contour line. The endpoints of the inner and outer contour lines at both ends are extended vertically to the image boundary to construct a closed region on the image plane. Using the midpoint between the starting points of the inner and outer contour lines as the seed point, the closed region is filled to generate an initial position template. Perform morphological dilation on the initial position template; The expanded initial position template is used as a spatial mask and applied to the original lateral cephalometric radiograph to extract the mouth and nose region.

4. The method for soft tissue adaptive enhancement of lateral cephalometric radiographs according to claim 3, characterized in that, Local enhancement is performed on the extracted nasal and oral regions, specifically including: The gray values ​​of all pixels extracted from the mouth and nose region are normalized to the range of [0, 1]. Dynamic Gamma correction is applied to the normalized gray values; The mask of the mouth and nose region is enlarged by a specific radius. The maximum and minimum pixel values ​​of the original image within the expanded region are statistically analyzed. The corrected pixel values ​​of the mouth and nose region are linearly mapped to the [min, max] interval to form the enhanced image of the mouth and nose region.

5. The method for soft tissue adaptive enhancement of lateral cephalometric radiographs according to claim 1, characterized in that, The enhanced image of the oral and nasal region is weighted and fused with the original lateral cephalometric radiograph to form a soft tissue enhanced image of the lateral cephalometric radiograph, specifically including: For the binary mask of the mouth and nose region, the region boundary is expanded by morphological dilation operation, where the dilation radius is adaptively adjusted according to the image resolution; Anisotropic Gaussian blurring based on local image texture features is applied to the expanded mask to generate a weight distribution map α with directional transition characteristics; Based on the generated weight distribution map α, the enhanced image of the mouth and nose region is fused with the original image of the lateral cephalometric radiograph using a weighted method to form a soft tissue enhanced image of the lateral cephalometric radiograph. Specifically, the weighted fusion is performed using the following formula: ； in, The image is a soft tissue enhancement image of the lateral cephalometric radiograph, and α is the weight distribution map. The enhanced image of the mouth and nose region. This is the original image of the lateral cephalometric radiograph.

6. A soft tissue adaptive enhancement device for lateral cephalometric radiographs, characterized in that, include: The extraction module is used to extract key anatomical points related to the oral and nasal regions in the lateral cephalometric radiograph using a UNet-based feature point recognition model. An enhancement module is used to extract the oral and nasal region from the lateral cephalometric radiograph based on the key anatomical points, and to locally enhance the extracted oral and nasal region. The fusion module is used to perform weighted fusion of the enhanced image of the mouth and nose region with the original image of the lateral cephalometric radiograph to form a soft tissue enhanced image of the lateral cephalometric radiograph.

7. The soft tissue adaptive enhancement device for lateral cephalometric radiographs according to claim 6, characterized in that, The UNet-based feature point recognition model is generated using the following method: Collect lateral cephalometric radiographs and annotate key anatomical points on the radiographs with expert annotations. Preprocessing and data augmentation are performed on the labeled samples; Convert each labeled key anatomical point in the sample into a Gaussian heatmap. The samples are divided into training set, validation set and test set; A feature point recognition model based on UNet is constructed. The model is trained by sampling the training set, validation set, and test set. During training, a binary cross-entropy loss with Sigmoid activation is used as the objective function, the Adam optimizer is selected, and the learning rate is dynamically adjusted using a cosine annealing strategy.

8. The soft tissue adaptive enhancement device for lateral cephalometric radiographs according to claim 6, characterized in that, The enhancement module is specifically used for: Connect the outer contour points and the inner contour points of the key anatomical points in sequence according to the anatomical order to form the inner contour line and the outer contour line. The endpoints of the inner and outer contour lines at both ends are extended vertically to the image boundary to construct a closed region on the image plane. Using the midpoint between the starting points of the inner and outer contour lines as the seed point, the closed region is filled to generate an initial position template. Perform morphological dilation on the initial position template; The expanded initial position template is used as a spatial mask and applied to the original lateral cephalometric radiograph to extract the mouth and nose region.

9. The soft tissue adaptive enhancement device for lateral cephalometric radiographs according to claim 8, characterized in that, The enhancement module is also used for: The gray values ​​of all pixels extracted from the mouth and nose region are normalized to the range of [0, 1]. Dynamic Gamma correction is applied to the normalized gray values; The mask of the mouth and nose region is enlarged by a specific radius. The maximum and minimum pixel values ​​of the original image within the expanded region are statistically analyzed. The corrected pixel values ​​of the mouth and nose region are linearly mapped to the [min, max] interval to form the enhanced image of the mouth and nose region.

10. The soft tissue adaptive enhancement device for lateral cephalometric radiographs according to claim 6, characterized in that, The fusion module is specifically used for: For the binary mask of the mouth and nose region, the region boundary is expanded by morphological dilation operation, where the dilation radius is adaptively adjusted according to the image resolution; Anisotropic Gaussian blurring based on local image texture features is applied to the expanded mask to generate a weight distribution map α with directional transition characteristics; Based on the generated weight distribution map α, the enhanced image of the mouth and nose region is fused with the original image of the lateral cephalometric radiograph using a weighted method to form a soft tissue enhanced image of the lateral cephalometric radiograph. Specifically, the weighted fusion is performed using the following formula: ； in, The image is a soft tissue enhancement image of the lateral cephalometric radiograph, and α is the weight distribution map. The enhanced image of the mouth and nose region. This is the original image of the lateral cephalometric radiograph.

Images