An abdominal tuberculosis image processing method

Abstract

The application provides an abdominal tuberculosis image processing method, relates to the technical field of image processing and medical image analysis, and specifically comprises the following steps: collecting abdominal tuberculosis images and making a data set, image preprocessing, a multi-scale convolution module, a multi-scale pooling module, an optimized feature fusion module, a classification prediction module, constructing an abdominal tuberculosis image processing model, and training and detecting the abdominal tuberculosis image processing model. The method pre-processes image data sets by introducing an adaptive enhancement mechanism and a noise suppression strategy, designs a multi-scale convolution module and a pooling module to realize multi-scale feature extraction and fusion, adopts low-frequency high-frequency information fusion and an adaptive weighting mechanism to optimize image feature processing, and multiple modules work together, so that the processing efficiency and classification accuracy of the abdominal tuberculosis image are significantly improved, and the method exhibits strong robustness and adaptability in image analysis of different tuberculosis lesion stages.

Description

Technical Field

[0001] The present invention belongs to the technical field of signal processing, and in particular relates to an abdominal tuberculosis image processing method. Background Art

[0002] Abdominal tuberculosis is a common type of digestive system tuberculosis. Abdominal tuberculosis images are usually acquired through medical imaging equipment such as CT, X-ray, or MRI. However, these abdominal tuberculosis image processing technologies still face many challenges. First, the outline of the lesion area in the abdominal tuberculosis image is blurred, especially in the early stages. The contrast between the lesion area and normal tissue is low and it is easy to be overlooked. Second, the quality of abdominal tuberculosis images is affected by external factors such as noise and artifacts. The low signal-to-noise ratio of abdominal tuberculosis images makes it difficult to accurately locate the tuberculosis area. Traditional image processing technologies cannot effectively distinguish lesions from background, resulting in a decrease in diagnostic accuracy. In addition, existing enhancement technologies have limited ability to enhance weak signals, especially when processing images with small lesions and weak signals. It is difficult to effectively expand the dynamic range of the image and cannot fully improve the image clarity and details.

[0003] In terms of image feature extraction, existing methods mostly focus on the extraction of single features such as grayscale value and contrast, lack comprehensive analysis and processing of multi-scale and multi-angle information, and are unable to fully capture the spatiotemporal characteristics of tuberculosis lesions; at the same time, traditional image enhancement methods lack adaptability when weighting features of different regions, and are unable to dynamically adjust the enhancement intensity of different regions according to the image content, resulting in local details being easily overlooked. For the processing of multi-level and complex structures of abdominal tuberculosis images, existing technologies are insufficient in the accuracy of automated analysis and recognition, especially when faced with scenarios with large differences in image quality. The enhancement effect cannot meet actual needs. There are also problems of high-dimensional calculation and large data volume in image processing. The existing methods perform poorly in real-time and efficiency, and cannot adapt to the real-time and efficient application requirements in clinical practice. Summary of the Invention

[0004] The present invention provides an abdominal tuberculosis image processing method, which aims to improve the visualization and diagnostic effect of tuberculosis lesion areas through multi-level abdominal tuberculosis image enhancement and analysis technology. During the abdominal tuberculosis image preprocessing process, the abdominal tuberculosis image is processed by an adaptive dynamic filtering and noise suppression strategy, and the image brightness and contrast are adjusted through an adaptive enhancement control mechanism to optimize the image quality of the lesion area; a multi-scale convolution and pooling module is designed to extract multidimensional features by combining spatial features, frequency domain features and local context information to accurately locate the lesion area; an adaptive weighting mechanism is used to fuse different features, and low-frequency and high-frequency components are combined to achieve a balance between details and global information; an abdominal tuberculosis image quality comparison and inference module is used to realize the similarity evaluation and result output of two abdominal tuberculosis images, and multiple modules collaborate to complete the efficient processing and analysis of abdominal tuberculosis images.

[0005] In order to achieve the above object, the present invention provides the following technical solution: a method for processing abdominal tuberculosis images, comprising the following specific steps:

[0006] S1. Collect abdominal tuberculosis images using abdominal CT, annotate them using the original pixel intensity and local structure information of the abdominal tuberculosis images, and create an abdominal tuberculosis image dataset.

[0007] S2. Introducing quality assessment and degradation perception, the quality of abdominal tuberculosis images is assessed by the original pixel intensity, local contrast, and local gradient features of the abdominal tuberculosis images. The local texture intensity and global contrast changes of the abdominal tuberculosis images are calculated for degradation perception.

[0008] S3. Perform pre-enhancement adjustment on the abdominal tuberculosis image, using the low-frequency and high-frequency components of the abdominal tuberculosis image to enhance the global information and local details of the abdominal tuberculosis image, especially for the detail enhancement of the tuberculosis lesion area, and design a pre-enhancement module;

[0009] S4. Using adaptive enhancement control, setting adaptive enhancement control parameters, and using local contrast, local texture intensity, and local gradient change as adaptive enhancement control parameters to adjust the brightness and contrast of the abdominal tuberculosis image;

[0010] S5. Introducing local multi-dimensional feature adaptive weighting to extract spatial features, frequency domain features, and context features of the abdominal tuberculosis image, calculating the weighting coefficient of each local area, and combining the spatial weighting and frequency domain weighting coefficients to obtain the final enhancement value of each pixel;

[0011] S6. Perform quality comparison reasoning on the abdominal tuberculosis images before and after enhancement, compare the similarity between the two abdominal tuberculosis images, and calculate the similarity of the original abdominal tuberculosis image feature design based on the structural similarity, gradient difference and brightness change of the abdominal tuberculosis images.

[0012] Preferably, in step S1, for the preparation of an abdominal tuberculosis image dataset, medical imaging equipment is used to collect abdominal tuberculosis images. The collected abdominal tuberculosis images contain different tuberculosis lesion areas, including pathological types of nodules, cavities, and spots. In order to ensure the breadth and representativeness of the dataset, a variety of different scanning angles are used in the acquisition process to ensure that the characteristics of tuberculosis lesions can be captured from different perspectives, especially the changes and morphological manifestations of the lesion areas at different scanning angles. Each abdominal tuberculosis image covers various lesion areas from early to late stages of tuberculosis, covering tuberculosis lesions of different sizes and shapes, and strives to show the different development stages and clinical manifestations of tuberculosis lesions through abdominal tuberculosis image data, covering tuberculosis lesion areas from early to late stages. In order to improve the quality and diversity of the dataset, individual differences of different patients are also taken into account during the acquisition process, including factors such as age, gender, and tuberculosis type, to ensure that the dataset has sufficient diversity and breadth for subsequent training and analysis.

[0013] Preferably, in step S2, the local texture intensity of the abdominal tuberculosis image is calculated to measure the detail preservation of the abdominal tuberculosis image, degradation perception is performed through the gradient and second-order gradient of the abdominal tuberculosis image, and the degraded area in the abdominal tuberculosis image is identified. The specific steps are:

[0014] S21. Calculate the gradient of the abdominal tuberculosis image, obtain the local texture intensity of the abdominal tuberculosis image, normalize it according to the global contrast, and obtain a quality score of the abdominal tuberculosis image; the mathematical model of the abdominal tuberculosis image quality score QS(I) is:

[0015] ;

[0016] Where, is the pixel value of the abdominal tuberculosis image I at position (i, j), is the gradient of the abdominal tuberculosis image I at position (i, j), and are the global mean and standard deviation of abdominal tuberculosis image I, respectively;

[0017] S22. Analyze the image using a second-order gradient method and determine the degree of degradation of the abdominal tuberculosis image by calculating the gradient change in the local area. The mathematical model of the degradation perception task DM(I) is:

[0018] ;

[0019] Where, is the second-order gradient of the abdominal tuberculosis image I at position (i, j), is the global mean of the second-order gradient.

[0020] Preferably, in step S2, the present invention first pre-processes the abdominal tuberculosis image by combining adaptive dynamic filtering and image enhancement strategy to significantly improve the quality of the abdominal tuberculosis image; in the process of pre-processing the abdominal tuberculosis image, by introducing an adaptive enhancement mechanism, the brightness and contrast of the abdominal tuberculosis image are dynamically adjusted according to the contrast difference between the tuberculosis lesion area and the background, so that the details of the lesion area are fully enhanced; specifically, the relatively blurred lesion area in the abdominal tuberculosis image will be more clearly displayed by increasing the contrast, thereby providing more accurate information for subsequent feature extraction; in order to cope with external noise interference, noise suppression technology is adopted, especially in the background of the abdominal tuberculosis image, to reduce unnecessary noise interference and improve the overall signal-to-noise ratio. By performing differentiated enhancement processing on different types of tuberculosis lesions, the performance of the abdominal tuberculosis image in various lesion areas is improved, ensuring that both early and late lesions can be effectively captured and displayed. The design of this module not only enhances the quality of the abdominal tuberculosis image, but also provides support for subsequent feature extraction.

[0021] Preferably, in step S3, a pilot enhancement adjustment is introduced, and the specific steps are:

[0022] An adaptive adjustment mechanism is introduced to adjust the coefficient of the control enhancement strength. The mathematical model is:

[0023] ;

[0024] Where, is the global average value of abdominal tuberculosis image I, is the gradient of abdominal tuberculosis image I at position (i, j);

[0025] The abdominal tuberculosis image is subjected to Fourier transform to separate the low-frequency and high-frequency parts, an enhancement operation is applied to the high-frequency part, and an inverse Fourier transform is used to convert the enhanced frequency-domain abdominal tuberculosis image back to the spatial domain to obtain an enhanced abdominal tuberculosis image. The mathematical model of the pilot enhancement adjustment is:

[0026] ;

[0027] Where, is the Fourier transform, is the inverse Fourier transform, is the frequency filter, is the coefficient for controlling the enhanced strength.

[0028] Preferably, in step S3, for the multi-scale convolution module, convolution kernels of different sizes can be used to process images, and different levels of information of abdominal tuberculosis images can be extracted from the global perspective and the local perspective respectively, especially in the extraction of smaller lesions, which can better capture the detail features. The pooling process effectively retains the important features of the abdominal tuberculosis images by pooling the multi-scale resolution of the abdominal tuberculosis images. Combined with the addition of the attention mechanism, the model can automatically focus on important areas in the abdominal tuberculosis images, improve the ability to focus on the tuberculosis lesion area, and effectively enhance the feature expression of the lesion area. The design of the local feature extraction strategy can enhance the details of the abdominal tuberculosis images and ensure that the multi-dimensional information of the abdominal tuberculosis images can be fully mined. Through the collaboration of this module, the extraction and fusion of multi-scale information in the abdominal tuberculosis images can be realized, the performance of the lesion area can be optimized, and the accuracy of subsequent classification and recognition can be improved.

[0029] Preferably, in step S4, adaptive enhanced control is performed, specifically in the following steps:

[0030] S41, adaptive enhancement control parameters To achieve dynamic adaptive adjustment, the control parameters The mathematical model is:

[0031] ;

[0032] Where, is the global average value of abdominal tuberculosis image I, To represent the local contrast of the abdominal tuberculosis image at position (x, y), is the gradient of abdominal tuberculosis image I at position (i, j);

[0033] S42. Calculate the enhancement coefficient at each location based on the local contrast and texture intensity of the abdominal tuberculosis image. Combined with the local gradient of the abdominal tuberculosis image, dynamically adjust the enhancement intensity to meticulously enhance the lesion area. The mathematical model for calculating the enhancement coefficient at each location is:

[0034] , ;

[0035] Where, is the gradient of the abdominal tuberculosis image I at position (i, j), is the global average value of abdominal tuberculosis image I, represents the local contrast of the image at position (x, y) for abdominal tuberculosis, is the global mean of the local contrast of abdominal tuberculosis, Adaptive enhancement of control parameters for abdominal tuberculosis, dynamically adjusted based on local image features, and are the maximum pixel value and minimum pixel value of the local window near the position (i, j) of the abdominal tuberculosis image I, respectively.

[0036] Preferably, in step S4, for the multi-scale pooling module, the module dynamically adjusts the feature weighting coefficient through an adaptive weighting mechanism, and assigns different weights to different features according to the comparison between the detail information of the lesion area and the background area. The local texture features and contrast features of the tuberculosis lesion area will be given higher weights, while the features of the background area will be less weighted. In this way, excessive weighting of features in the lesion area can be avoided, which leads to interference from background noise. The global structure and local details of the image are optimized through the weighted fusion of low-frequency and high-frequency components. The low-frequency component represents the global information of the image and can maintain the overall structure of the image, while the high-frequency component can enhance the detail performance of the image. Through adaptive weighting and the fusion of low-frequency and high-frequency information, the details of the image after enhancement and the overall consistency are ensured, especially the enhancement of the lesion area, so that the final image is clearer and easier to process subsequently.

[0037] Preferably, in step S5, local multi-dimensional feature adaptive weighting is introduced, and the specific steps are:

[0038] S51, extracting spatial features, frequency domain features, and context features of the abdominal tuberculosis image to obtain multidimensional features of the abdominal tuberculosis image; the mathematical model for calculating the multidimensional features of the image is:

[0039] , ;

[0040] Where, is the spatial weighting coefficient of abdominal tuberculosis image, is the local contrast of abdominal tuberculosis image, is the frequency domain weighting coefficient, is the Fourier transform;

[0041] S52. Calculate the weighting coefficient of each local area based on the spatial characteristics and frequency domain characteristics of the abdominal tuberculosis image. The mathematical model for calculating the weighting coefficient of each local area is:

[0042] ;

[0043] Where, To enhance the pixel value of posterior abdominal tuberculosis image, is the pixel value of the original abdominal tuberculosis image.

[0044] Preferably, in step S5, for the group attention module, the model can dynamically adjust the classification strategy according to the different features of the image. The change of the classification factor enables the model to adaptively adjust the weight distribution when facing different types of tuberculosis lesions, thereby improving the model's recognition ability of different lesion types; the introduced regularization constraint effectively prevents the model from overfitting during the training process, improves the generalization ability and stability of the model, and enables the model to maintain a high accuracy when facing new data. By combining the feature integration of the time dimension and the spatial dimension, the model can fully consider the spatiotemporal relationship of the image during classification prediction, thereby improving the accuracy and stability of the classification. The design of this module enhances the effect of image classification and provides reliable technical support for the accurate classification and prediction of tuberculosis lesions.

[0045] Preferably, in step S6, quality comparison reasoning is introduced, and the specific steps are:

[0046] Based on the structural similarity, gradient difference and brightness change of the abdominal tuberculosis image, the similarity calculation is designed based on the original abdominal tuberculosis image features; the similarity The mathematical model for calculation is:

[0047] , , ;

[0048] Where, and Abdominal tuberculosis images and abdominal tuberculosis images The pixel value at position (i,j), To adjust the coefficient of brightness difference contribution to similarity calculation, To adjust the coefficients of the influence of gradient difference and brightness difference on similarity calculation, is the standard deviation of abdominal tuberculosis image I.

[0049] Preferably, in step S6, for the classification prediction module, the frequency upper and lower limits of the abdominal tuberculosis image are dynamically adjusted to optimize the adaptability of the abdominal tuberculosis image under different frequency components, so that areas with weak signals or blurred details can be effectively enhanced. By dynamically adjusting the low-frequency and high-frequency components, the weak signals and details in the abdominal tuberculosis image are more accurately enhanced. After the abdominal tuberculosis image is enhanced, the abdominal tuberculosis image is segmented in a sliding window manner to form standardized data blocks, thereby improving the extraction accuracy of the abdominal tuberculosis image features and the accuracy of classification. After the standardized processing, each data block can better structure the extraction of abdominal tuberculosis image features, thereby improving the resolution and classification accuracy of the abdominal tuberculosis image. The overall design strengthens the flexibility and robustness of the abdominal tuberculosis image processing, provides a reliable data basis for subsequent abdominal tuberculosis image modeling and analysis, and ensures the effectiveness of the abdominal tuberculosis image in various practical application scenarios.

[0050] Compared with the existing technology, the present invention has the following technical effects: the technical solution provided by the present invention improves the traditional image processing method through an adaptive enhancement mechanism and a noise suppression strategy, thereby improving the quality and processing accuracy of abdominal tuberculosis images; by designing a multi-scale convolution module and a pooling module, combined with an attention mechanism and a local feature extraction strategy, multi-scale feature extraction and fusion of the image are realized, which can more accurately capture the detailed information of the tuberculosis lesion area; at the same time, the adaptive weighting mechanism and the fusion of low-frequency and high-frequency information are adopted to effectively improve the detail retention ability of the image and avoid excessive enhancement of the background area; in the classification and prediction stage, adaptive classification factors and regularization constraints are introduced, combined with the integration of spatial and temporal dimension features, to optimize the classification accuracy and stability of tuberculosis lesions; the collaborative work of multiple modules makes the processing of abdominal tuberculosis images more efficient and accurate, greatly improves the classification and detection effects of the image, and meets the requirements of image processing accuracy and real-time performance in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The figure is a flow chart of the abdominal tuberculosis image processing method provided by the present invention.

[0052] Figure 2 This is a structural diagram of the abdominal tuberculosis image processing preprocessing provided by the present invention.

[0053] Figure 3 This is a structural diagram of the local multi-dimensional feature adaptive weighting provided by the present invention.

[0054] Figure 4 This is a diagram showing the enhancement effect of the abdominal tuberculosis image processing model provided by the present invention on the abdominal tuberculosis image. Figure 4 (a) is the image of abdominal tuberculosis after enhancement processing. Figure 4Middle (b) is the abdominal tuberculosis image before enhancement processing. DETAILED DESCRIPTION

[0055] The present invention provides an abdominal tuberculosis image processing method, which aims to improve the processing accuracy and efficiency of abdominal tuberculosis images through the collaborative work of multiple modules; abdominal tuberculosis images of different scanning angles are collected by medical imaging equipment, covering nodules, cavities, and spots, ensuring the comprehensiveness and representativeness of the data set; after the abdominal tuberculosis images are collected, the abdominal tuberculosis images are preprocessed using an adaptive enhancement mechanism and a noise suppression strategy to optimize the brightness and contrast of the abdominal tuberculosis images and improve the detail expression of the lesion area; a multi-scale convolution and pooling module is used to accurately extract the abdominal tuberculosis images by combining the attention mechanism and local feature extraction. The multi-level information of the abdominal tuberculosis image is taken into consideration, especially the key features of the lesion area; in the process of weighting and fusion of abdominal tuberculosis image features, an adaptive weighting mechanism is used, combined with the fusion of low-frequency and high-frequency information, to ensure the balance between the details and global information of the abdominal tuberculosis image and avoid excessive enhancement of the background area; the classification and prediction module introduces adaptive classification factors and regularization constraints, combines the integration of time and space dimension features, optimizes the classification accuracy, and improves the stability and accuracy of abdominal tuberculosis image classification; multiple modules work together to ensure efficient enhancement and detection of abdominal tuberculosis images, and provide high-quality data support for subsequent abdominal tuberculosis image analysis and diagnosis.

[0056] See Figure 1 As shown, a method for processing abdominal tuberculosis images in an embodiment of the present application has the following specific steps.

[0057] S1. Abdominal TB images are collected using abdominal CT. The original pixel intensity and local structure information of abdominal TB images are used for annotation to create an abdominal TB image dataset.

[0058] Furthermore, in step S1, for the production of the abdominal tuberculosis image dataset, abdominal tuberculosis image data containing various types of tuberculosis lesions are collected, covering abdominal tuberculosis images of different lesion areas and development stages, including tuberculosis lesions of nodule, cavity, and spot pathological types. Abdominal tuberculosis images are collected by medical imaging equipment to ensure that the abdominal tuberculosis image data covers different tuberculosis lesion areas from early to late stages, including lesions of different sizes and shapes. During the abdominal tuberculosis image collection process, multiple scanning angles are used to ensure that the different manifestations and morphological changes of tuberculosis lesions are captured. Each abdominal tuberculosis image is annotated, including information such as lesion type, lesion location, lesion size, scanning angle, and abdominal tuberculosis image acquisition time. The collected abdominal tuberculosis images and annotation information are organized into a unified dataset format to ensure data standardization, and finally an abdominal tuberculosis image dataset is formed to provide reliable data support for subsequent abdominal tuberculosis image processing, feature extraction, and classification tasks.

[0059] S2. Quality assessment and degradation perception are introduced. The quality of abdominal tuberculosis images is evaluated through the original pixel intensity, local contrast, and local gradient features of abdominal tuberculosis images. The local texture intensity and global contrast changes of abdominal tuberculosis images are calculated for degradation perception.

[0060] Furthermore, in step S2, a local sliding window is constructed at each position of the image, and the local contrast value is calculated based on the grayscale mean and standard deviation of the pixels in the window, thereby reflecting the degree of grayscale difference between the region and the surrounding pixels. The higher the local contrast, the clearer the image edge and texture. Taking local contrast as one of the core features of image quality assessment is helpful to identify blurred and degraded areas of the image, and provide basic support for subsequent image enhancement and adaptive adjustment. Local texture intensity is used to measure the intensity of grayscale fluctuations in the local area of ​​the image, which can reflect the integrity and clarity of structures such as edges, contours, and details in the region. The higher the texture intensity, the more drastic the grayscale changes in the image in the region and the richer the details. For abdominal tuberculosis images, the lesion area often contains subtle structural differences, so local texture intensity can be used as a measure. It is an important basis for measuring lesion clarity and image quality, calculating the local texture intensity of the abdominal tuberculosis image, measuring the detail retention of the abdominal tuberculosis image, performing degradation perception through the gradient and second-order gradient of the abdominal tuberculosis image, and identifying the degraded area in the abdominal tuberculosis image. The specific steps are S21 to S22. In degradation perception, the present invention determines whether the current image has global degradation features by performing difference comparison with the contrast reference value of the standard image set or the image before enhancement. If it is found that the contrast change of the image is too small, it means that the visual hierarchy of the image is blurred and the details are compressed, and an enhancement operation needs to be performed; if the change is abnormally large, it indicates that over-enhancement or structural distortion may have occurred. The dynamic change of global contrast serves as a global judgment basis for evaluating the quality of abdominal tuberculosis images, and effectively supports the adaptive control of subsequent enhancement adjustment and image quality optimization strategies.

[0061] S21. Calculate the gradient of the abdominal tuberculosis image, obtain the local texture intensity of the abdominal tuberculosis image, normalize it according to the global contrast, and obtain a quality score of the abdominal tuberculosis image; the mathematical model of the abdominal tuberculosis image quality score QS(I) is:

[0062] ;

[0063] Where, is the pixel value of the abdominal tuberculosis image I at position (i, j), is the gradient of the abdominal tuberculosis image I at position (i, j). The Sobel operator is used to calculate the gradient of the abdominal tuberculosis image in the implementation process. The value range is 0 to 255. and are the global mean and standard deviation of the abdominal tuberculosis image I, representing the overall brightness of the abdominal tuberculosis image and measuring the distribution of pixel values ​​in the abdominal tuberculosis image. During the implementation process, the initial value is set to 1, and the value range is 0 to 255.

[0064] S22. Analyze the abdominal tuberculosis image using a second-order gradient, and determine the degree of degradation of the abdominal tuberculosis image by calculating the gradient change in the local area. The mathematical model of the degradation perception task DM(I) is:

[0065] ;

[0066] Where, is the second-order gradient of the abdominal tuberculosis image I at position (i, j), with a value range of 0 to 255, reflecting the acceleration of the change of pixel values ​​in the abdominal tuberculosis image. It is achieved by performing second-order derivative filtering on the abdominal tuberculosis image using the horizontal and vertical filters of the Sobel operator. The global mean of the second-order gradient is used to calculate the average value of the second-order gradient of all pixels in the abdominal tuberculosis image. The value range is 0 to 255, which is used to measure the overall detail change intensity of the abdominal tuberculosis image.

[0067] S3. Perform preliminary enhancement adjustment on the abdominal tuberculosis image, and use the low-frequency and high-frequency components of the abdominal tuberculosis image to enhance the global information and local details of the abdominal tuberculosis image respectively, especially for the detail enhancement of the tuberculosis lesion area, and design a pre-enhancement module.

[0068] Furthermore, in step S3, for the lead enhancement regulation, its structure is as follows Figure 2 As shown, the specific steps are:

[0069] An adaptive adjustment mechanism is introduced to adjust the coefficient of the control enhancement strength. The mathematical model is:

[0070] ;

[0071] Where, is the global average value of the abdominal tuberculosis image I, and the average brightness of all pixels in the abdominal tuberculosis image is calculated, with a value range of 0 to 255. is the gradient of abdominal tuberculosis image I at position (i, j);

[0072] The abdominal tuberculosis image is subjected to Fourier transform to separate the low-frequency and high-frequency parts, an enhancement operation is applied to the high-frequency part, and an inverse Fourier transform is used to convert the enhanced frequency-domain abdominal tuberculosis image back to the spatial domain to obtain an enhanced abdominal tuberculosis image. The mathematical model of the pilot enhancement adjustment is:

[0073] ;

[0074] Where, is the Fourier transform, is the inverse Fourier transform, is a frequency filter with a value range of 0 to 2π. is the coefficient for controlling the enhanced strength.

[0075] S4. Use adaptive enhancement control to set adaptive enhancement control parameters, and use local contrast, local texture intensity, and local gradient change as adaptive enhancement control parameters to adjust the brightness and contrast of the abdominal tuberculosis image.

[0076] Furthermore, in step S4, adaptive enhanced control is performed, specifically steps S41 to S42.

[0077] S41, adaptive enhancement control parameters To achieve dynamic adaptive adjustment, the control parameters The mathematical model is:

[0078] ;

[0079] Where, is the global average value of abdominal tuberculosis image I, To represent the local contrast of the abdominal tuberculosis image at position (x, y), a sliding window is used to calculate a neighborhood of a certain size around each pixel and calculate the mean and standard deviation of the neighborhood. The difference between each pixel and the mean of its local neighborhood is calculated and then divided by the standard deviation to obtain the local contrast. is the gradient of the abdominal tuberculosis image I at position (i, j).

[0080] S42. Calculate the enhancement coefficient at each location based on the local contrast and texture intensity of the abdominal tuberculosis image. Combined with the local gradient of the abdominal tuberculosis image, dynamically adjust the enhancement intensity to meticulously enhance the lesion area. The mathematical model for calculating the enhancement coefficient at each location is:

[0081] , ;

[0082] Where, is the gradient of the abdominal tuberculosis image I at position (i, j), is the global average value of abdominal tuberculosis image I, To represent the local contrast of the abdominal tuberculosis image at position (x, y), is the global mean of local contrast, To adaptively enhance the control parameters, the local features of abdominal tuberculosis images are dynamically adjusted. and are the maximum pixel value and minimum pixel value of the local window near the position (i, j) of the abdominal tuberculosis image I, respectively.

[0083] S5. Introduce local multidimensional feature adaptive weighting to extract the spatial features, frequency domain features, and context features of the abdominal tuberculosis image, calculate the weighting coefficient of each local area, and combine the spatial weighting and frequency domain weighting coefficients to obtain the final enhancement value of each pixel.

[0084] Furthermore, in step S5, local multi-dimensional feature adaptive weighting is introduced, and the specific steps are S51 to S52.

[0085] S51, extracting spatial features, frequency domain features, and context features of the abdominal tuberculosis image to obtain multidimensional features of the abdominal tuberculosis image; the mathematical model for calculating the multidimensional features of the image is:

[0086] , ;

[0087] Where, is the spatial weighting coefficient of abdominal tuberculosis image, ranging from 0 to 1. is the local contrast of abdominal tuberculosis image, is the frequency domain weighting coefficient, ranging from 0 to 1, is the Fourier transform.

[0088] S52. Calculate the weighting coefficient of each local area based on the spatial characteristics and frequency domain characteristics of the abdominal tuberculosis image. The mathematical model for calculating the weighting coefficient of each local area is:

[0089] ;

[0090] Where, To enhance the pixel value of posterior abdominal tuberculosis image, is the pixel value of the original abdominal tuberculosis image.

[0091] S6. Perform quality comparison reasoning on the abdominal tuberculosis images before and after enhancement, compare the similarity between the two abdominal tuberculosis images, and calculate the similarity of the original abdominal tuberculosis image feature design based on the structural similarity, gradient difference and brightness change of the abdominal tuberculosis images.

[0092] Furthermore, in step S6, quality comparison reasoning is introduced, and the specific steps are as follows.

[0093] Based on the structural similarity, gradient difference and brightness change of the abdominal tuberculosis image, the similarity calculation is designed based on the original abdominal tuberculosis image features; the similarity The mathematical model for calculation is:

[0094] , , ;

[0095] Where, and Abdominal tuberculosis images and abdominal tuberculosis images The pixel value at position (i,j), To adjust the coefficient of brightness difference's contribution to similarity calculation, the value range is 0 to 0.1 during implementation. In order to adjust the coefficient of the influence of gradient difference and brightness difference on similarity calculation, the value range is 0 to 0.1 during implementation. is the standard deviation of the abdominal tuberculosis image I, and is set to 0.5 during implementation.

[0096] Furthermore, in step S6, the abdominal tuberculosis image processing model is developed based on the Python language and implemented using the TensorFlow or Pytorch framework. The abdominal tuberculosis image is used as input and trained using the Adam optimizer. The learning rate is set to 0.0001, the training batch size is 64, and the number of iterations is 50. During the training process, the generalization ability of the model is further improved by introducing data augmentation technology and regularization methods. After the training is completed, the abdominal tuberculosis image processing model outputs the final lesion area classification result.

[0097] Furthermore, in step S6, Figure 4 The image comparison of abdominal tuberculosis before and after enhancement is shown. Figure 4 (a) is the image of abdominal tuberculosis after enhancement processing. Figure 4 Middle (b) is the abdominal tuberculosis image before enhancement. The model can effectively improve the visualization of the tuberculosis lesion area, making the lesion edge clearer and the details more prominent. In the abdominal tuberculosis image before enhancement, the lesion area is difficult to identify clearly due to low contrast and fuzziness. However, in the abdominal tuberculosis image after enhancement, the outline of the lesion area is more obvious and the details are enhanced. Especially in the identification of mid-term and late-stage tuberculosis lesions, the accuracy and clarity of the abdominal tuberculosis image are significantly improved. The overall effect shows that the abdominal tuberculosis image enhancement technology significantly improves the quality of abdominal tuberculosis images, especially in terms of details and lesion differentiation, ensuring the efficiency and accuracy of subsequent feature extraction and classification tasks.

[0098] The above are only preferred embodiments of the present invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for processing abdominal tuberculosis images, characterized in that: The specific steps include: S1. Collect tuberculosis images using abdominal CT, annotate them using the original pixel intensity and local structure information of abdominal tuberculosis images, and create an abdominal tuberculosis image dataset. S2. Introducing quality assessment and degradation perception, the quality of abdominal tuberculosis images is assessed by the original pixel intensity, local contrast, and local gradient features of the abdominal tuberculosis images. The local texture intensity and global contrast changes of the abdominal tuberculosis images are calculated for degradation perception. S3. Perform pre-enhancement adjustment on the abdominal tuberculosis image, using the low-frequency and high-frequency components of the abdominal tuberculosis image to enhance the global information and local details of the abdominal tuberculosis image, especially for the detail enhancement of the tuberculosis lesion area, and design a pre-enhancement module; S4. Use adaptive enhancement control to set adaptive enhancement control parameters, and use local contrast, local texture intensity, and local gradient change as adaptive enhancement control parameters to adjust the brightness and contrast of the abdominal tuberculosis image; specifically: S41, adaptive enhancement control parameters To achieve dynamic adaptive adjustment, the control parameters The mathematical model is: ； Where, is the global average value of abdominal tuberculosis image I, To represent the local contrast of the abdominal tuberculosis image at position (x, y), is the gradient of abdominal tuberculosis image I at position (i, j); S42. Calculate the enhancement coefficient at each location based on the local contrast and texture intensity of the abdominal tuberculosis image. Combined with the local gradient of the abdominal tuberculosis image, dynamically adjust the enhancement intensity to meticulously enhance the lesion area. The mathematical model for calculating the enhancement coefficient at each location is: ， ； Where, is the gradient of the abdominal tuberculosis image I at position (i, j), is the global average value of abdominal tuberculosis image I, To represent the local contrast of the abdominal tuberculosis image at position (x, y), is the global mean of local contrast, To adaptively enhance the control parameters, the local features of abdominal tuberculosis images are dynamically adjusted. and are the maximum pixel value and the minimum pixel value of the local window near the position (i, j) in the abdominal tuberculosis image I, respectively; S5. Introduce local multidimensional feature adaptive weighting to extract the spatial features, frequency domain features, and context features of the abdominal tuberculosis image, calculate the weighting coefficient of each local area, and combine the spatial weighting and frequency domain weighting coefficients to obtain the final enhancement value of each pixel, specifically: S51, extracting spatial features, frequency domain features, and context features of the abdominal tuberculosis image to obtain multidimensional features of the abdominal tuberculosis image; S52. Calculating a weighting coefficient for each local area based on the spatial characteristics and frequency domain characteristics of the abdominal tuberculosis image; S6. Perform quality comparison reasoning on the abdominal tuberculosis images before and after enhancement, compare the similarity between the two abdominal tuberculosis images, and calculate the similarity of the original abdominal tuberculosis image feature design based on the structural similarity, gradient difference and brightness change of the abdominal tuberculosis images.

2. The abdominal tuberculosis image processing method according to claim 1, characterized in that: To create an abdominal tuberculosis image dataset, medical imaging equipment was used to collect abdominal tuberculosis images. The collected abdominal tuberculosis images contain different tuberculosis lesion areas, including pathological types such as nodules, cavities, and spots. A variety of scanning angles were used to cover tuberculosis lesion areas from early to late stages.

3. The abdominal tuberculosis image processing method according to claim 2, characterized in that: Calculate the local texture intensity of the abdominal tuberculosis image to measure the detail preservation of the abdominal tuberculosis image. Use the gradient and second-order gradient of the abdominal tuberculosis image to perform degradation perception and identify the degraded areas in the abdominal tuberculosis image. The specific steps are as follows: S21. Calculate the gradient of the abdominal tuberculosis image, obtain the local texture intensity of the abdominal tuberculosis image, normalize it according to the global contrast, and obtain a quality score of the abdominal tuberculosis image; the mathematical model of the abdominal tuberculosis image quality score QS(I) is: ； Where, is the pixel value of the abdominal tuberculosis image I at position (i, j), is the gradient of the abdominal tuberculosis image I at position (i, j), and are the global mean and standard deviation of abdominal tuberculosis image I, respectively; S22. Analyze the abdominal tuberculosis image using a second-order gradient, and determine the degree of degradation of the abdominal tuberculosis image by calculating the gradient change in the local area. The mathematical model of the degradation perception task DM(I) is: ； Where, is the second-order gradient of the abdominal tuberculosis image I at position (i, j), is the global mean of the second-order gradient.

4. The abdominal tuberculosis image processing method according to claim 3, characterized in that: Introduce pilot enhancement regulation, the specific steps are: An adaptive adjustment mechanism is introduced to adjust the coefficient of the control enhancement strength. The mathematical model is: ； Where, is the global average value of abdominal tuberculosis image I, is the gradient of abdominal tuberculosis image I at position (i, j); The abdominal tuberculosis image is subjected to Fourier transform to separate the low-frequency and high-frequency parts, an enhancement operation is applied to the high-frequency part, and an inverse Fourier transform is used to convert the enhanced frequency-domain abdominal tuberculosis image back to the spatial domain to obtain an enhanced abdominal tuberculosis image. The mathematical model of the pilot enhancement adjustment is: ； Where, is the Fourier transform, is the inverse Fourier transform, is the frequency filter, is the coefficient for controlling the enhanced strength.

5. The abdominal tuberculosis image processing method according to claim 4, characterized in that: In step S5, the mathematical model for calculating the multidimensional features of the image is: ， ； Where, is the spatial weighting coefficient of abdominal tuberculosis image, is the local contrast of abdominal tuberculosis image, is the frequency domain weighting coefficient, is the Fourier transform; The mathematical model for calculating the weighted coefficient of each local area is: ； Where, To enhance the pixel value of posterior abdominal tuberculosis image, is the pixel value of the original abdominal tuberculosis image.

6. The abdominal tuberculosis image processing method according to claim 5, characterized in that: Introducing quality comparison reasoning, the specific steps are: Based on the structural similarity, gradient difference and brightness change of the abdominal tuberculosis image, the similarity calculation is designed based on the original abdominal tuberculosis image features; the similarity The mathematical model for calculation is: ， ， ； Where, and Abdominal tuberculosis images and abdominal tuberculosis images The pixel value at position (i,j), To adjust the coefficient of brightness difference contribution to similarity calculation, To adjust the coefficients of the influence of gradient difference and brightness difference on similarity calculation, is the standard deviation of abdominal tuberculosis image I.

Images