Cancer image contrast enhancement method and system based on deep learning

Abstract

The invention discloses a cancer image contrast enhancement method and system based on deep learning, belongs to the technical field of medical image processing, and adopts an adaptive histogram equalization algorithm for preprocessing to enhance the contrast of a local region. Texture, shape and edge features are extracted through multi-scale analysis, and feature descriptors are constructed. And based on the feature descriptor, segmenting the image by using a region growing segmentation algorithm to obtain a focus region boundary. A self-adaptive contrast enhancement technology is adopted for the segmented focus area, and the focus display effect is highlighted. In consideration of noise interference, a denoising method based on wavelet transform is introduced, noise is suppressed in a high-frequency sub-band, and details are reserved in a low-frequency sub-band. Through the organic combination of the above technical schemes, the contrast and definition of the cancer image can be effectively enhanced, the focus area is highlighted, and noise interference is inhibited at the same time.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image processing, and in particular relates to a cancer image contrast enhancement method and system based on deep learning. Background Art

[0002] Cancer image contrast enhancement is a challenging task that requires consideration of many factors. First, cancer images of different types and stages vary greatly in imaging quality, resolution, and contrast, which makes image preprocessing and feature extraction difficult. Second, the lesion area in the cancer image is often close to the grayscale value of the surrounding tissue, with blurred boundaries and low contrast. During the enhancement process, it is necessary to avoid introducing too much noise and artifacts that affect the image clarity. Furthermore, the complexity and diversity of cancer images make it difficult for general image enhancement methods to achieve ideal results, and targeted algorithms and models need to be designed. Finally, in practical applications, the cancer image contrast enhancement system needs to take into account both processing speed and performance to achieve fast, accurate, and stable enhancement effects. At the same time, it must be closely integrated with the medical workflow to facilitate doctors' use. Therefore, how to balance various factors and develop efficient and reliable cancer image contrast enhancement methods and systems is a technical problem that needs to be solved urgently. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a cancer image contrast enhancement method and system based on deep learning to solve the problems existing in the above-mentioned prior art.

[0004] In a first aspect, to achieve the above-mentioned object, the present invention provides a method for contrast enhancement of cancer images based on deep learning, comprising the following steps:

[0005] Acquire a cancer image, and preprocess the cancer image based on an adaptive histogram equalization algorithm;

[0006] For the preprocessed cancer image, based on the multi-scale analysis method, features are extracted at different scales, and a feature descriptor is constructed by fusing the multi-scale features, wherein the features include texture features, shape features and edge features;

[0007] Based on the feature descriptor, a region growing segmentation algorithm is used to segment the cancer image, the grayscale and texture features of the lesion area are used as growth criteria, the segmentation area is adaptively expanded, and segmentation noise and artifacts are eliminated through morphological processing;

[0008] For the segmented lesion area, based on the adaptive contrast enhancement technology, the enhancement parameters are automatically adjusted according to the grayscale statistical characteristics of the lesion area to enhance the contrast of the lesion area;

[0009] For the enhanced image, a denoising method based on wavelet transform is adopted. Through multi-scale wavelet decomposition, soft threshold processing is performed on the wavelet coefficients in the high-frequency sub-band to suppress high-frequency noise, while retaining the original coefficients in the low-frequency sub-band, and reconstructing the image through inverse transform to obtain the final image with contrast enhancement.

[0010] Optionally, the process of preprocessing the cancer image using an adaptive histogram equalization algorithm includes:

[0011] Obtain a cancer image dataset and convert the images into grayscale;

[0012] Perform block processing on the grayscale image, divide the image into several local areas, and calculate the grayscale histogram of each area;

[0013] For areas with small grayscale differences, an adaptive histogram equalization algorithm is used to adjust the grayscale distribution and enhance the local contrast;

[0014] All local regions are combined to obtain the complete cancer image after preprocessing.

[0015] Optionally, the multi-scale analysis method includes: performing multi-scale decomposition on the pre-processed cancer image using wavelet transform to extract texture features, shape features and edge features;

[0016] The texture features, shape features and edge features extracted at different scales are fused, and a feature descriptor is constructed through feature fusion. The feature descriptor is then screened using the Fisher discriminant criterion to obtain the screened feature descriptor.

[0017] Optionally, the process of segmenting the cancer image includes:

[0018] The region growing segmentation algorithm expands the segmentation region by dynamically adjusting the thresholds of grayscale and texture features, and optimizes the segmentation result in combination with morphological processing to obtain the boundary of the lesion region.

[0019] Optionally, based on the adaptive contrast enhancement technology, the enhancement parameters are automatically adjusted according to the grayscale statistical characteristics of the lesion area. The process of enhancing the contrast of the lesion area includes:

[0020] Obtaining segmented lesion region image data, performing grayscale statistical analysis on the lesion region image, and obtaining grayscale statistical characteristic parameters of the lesion region;

[0021] Determine the target parameter range of adaptive contrast enhancement according to the obtained grayscale statistical characteristic parameters of the lesion area;

[0022] Adopting adaptive threshold segmentation algorithm, the lesion area image is divided into multiple sub-areas. For each sub-area, the contrast enhancement parameters are automatically adjusted within the target parameter range according to its grayscale statistical characteristics.

[0023] Contrast enhancement is performed on each sub-region image, and all enhanced sub-region images are fused to obtain an overall enhanced lesion region image.

[0024] Optionally, the wavelet transform-based denoising method suppresses noise by performing soft threshold processing on wavelet coefficients of high-frequency subbands, while retaining original coefficients of low-frequency subbands to preserve image details, and obtains a denoised image through inverse transformation.

[0025] In a second aspect, the present invention further provides a cancer image contrast enhancement system based on deep learning, which is used to implement a cancer image contrast enhancement method based on deep learning, and the system comprises:

[0026] A preprocessing module, used for acquiring a cancer image, and preprocessing the cancer image based on an adaptive histogram equalization algorithm;

[0027] A multi-scale feature extraction module, for extracting features at different scales for the pre-processed cancer image based on a multi-scale analysis method, and constructing a feature descriptor by fusing the multi-scale features, wherein the features include texture features, shape features and edge features;

[0028] A region growing segmentation module is used to segment the cancer image based on the feature descriptor using a region growing segmentation algorithm, taking the grayscale and texture features of the lesion area as growth criteria, adaptively expanding the segmentation area, and eliminating segmentation noise and artifacts through morphological processing;

[0029] The contrast enhancement module is used to automatically adjust the enhancement parameters and enhance the contrast of the lesion area based on the adaptive contrast enhancement technology according to the grayscale statistical characteristics of the lesion area after segmentation;

[0030] The wavelet denoising module is used to denoise the enhanced image by using a wavelet transform-based denoising method. Through multi-scale wavelet decomposition, soft threshold processing is performed on the wavelet coefficients in the high-frequency sub-band to suppress high-frequency noise, while retaining the original coefficients in the low-frequency sub-band, and reconstructing the image through inverse transformation to obtain the final image with contrast enhancement.

[0031] Optionally, the region growing segmentation module optimizes the contour of the segmented region by combining dynamic threshold adjustment with morphological processing.

[0032] In a third aspect, the present invention further provides a computer terminal device, comprising:

[0033] one or more processors;

[0034] A memory, coupled to the processor, for storing one or more programs;

[0035] When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of a cancer image contrast enhancement method based on deep learning.

[0036] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a cancer image contrast enhancement method based on deep learning.

[0037] Compared with the prior art, the present invention has the following advantages and technical effects:

[0038] The present invention provides a cancer image contrast enhancement method and system based on deep learning. Aiming at the problem of insufficient contrast of cancer images, an adaptive histogram equalization algorithm is used for preprocessing to enhance the contrast of local areas. Texture, shape and edge features are extracted through multi-scale analysis to construct a feature descriptor. Based on the feature descriptor, the image is segmented using a region growing segmentation algorithm to obtain the boundary of the lesion area. An adaptive contrast enhancement technique is used for the segmented lesion area to highlight the lesion display effect. Taking into account noise interference, a denoising method based on wavelet transform is introduced to suppress noise in high-frequency sub-bands and retain details in low-frequency sub-bands. Through the organic combination of the above technical solutions, the present invention can effectively enhance the contrast and clarity of cancer images, highlight the lesion area, and suppress noise interference. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0040] Figure 1 is a flowchart of a method for contrast enhancement of cancer images based on deep learning according to an embodiment of the present invention;

[0041] Figure 2 4 is a system schematic diagram of a cancer image contrast enhancement system based on deep learning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0042] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0043] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0044] Embodiment 1

[0045] like Figure 1 As shown, this embodiment provides a cancer image contrast enhancement method based on deep learning, including:

[0046] Acquire a cancer image, and preprocess the cancer image based on an adaptive histogram equalization algorithm;

[0047] For the preprocessed cancer image, based on the multi-scale analysis method, features are extracted at different scales, and a feature descriptor is constructed by fusing the multi-scale features, wherein the features include texture features, shape features and edge features;

[0048] Based on the feature descriptor, a region growing segmentation algorithm is used to segment the cancer image, the grayscale and texture features of the lesion area are used as growth criteria, the segmentation area is adaptively expanded, and segmentation noise and artifacts are eliminated through morphological processing;

[0049] For the segmented lesion area, based on the adaptive contrast enhancement technology, the enhancement parameters are automatically adjusted according to the grayscale statistical characteristics of the lesion area to enhance the contrast of the lesion area;

[0050] For the enhanced image, a denoising method based on wavelet transform is adopted. Through multi-scale wavelet decomposition, soft threshold processing is performed on the wavelet coefficients in the high-frequency sub-band to suppress high-frequency noise, while retaining the original coefficients in the low-frequency sub-band, and reconstructing the image through inverse transform to obtain the final image with contrast enhancement.

[0051] As an implementation method of this embodiment, the process of preprocessing the cancer image using an adaptive histogram equalization algorithm includes:

[0052] Obtain a cancer image dataset and convert the images into grayscale;

[0053] Perform block processing on the grayscale image, divide the image into several local areas, and calculate the grayscale histogram of each area;

[0054] For areas with small grayscale differences, an adaptive histogram equalization algorithm is used to adjust the grayscale distribution and enhance the local contrast;

[0055] All local regions are combined to obtain the complete cancer image after preprocessing.

[0056] As an optional implementation in this embodiment, an adaptive histogram equalization algorithm is used for preprocessing based on the contrast characteristics of the cancer image. By adjusting the grayscale distribution of the cancer image, the contrast of the local area is enhanced, making the grayscale difference between the lesion area and the surrounding tissue more obvious.

[0057] As an implementation method that can be added in this embodiment, a cancer image data set is obtained, the image is grayed, and the color image is converted into a gray image to reduce the computational complexity. The gray image is processed in blocks, and the image is divided into multiple local areas, and each area is independently subjected to a histogram equalization operation. For each local area, the histogram distribution of the grayscale values ​​of the pixels in the area is calculated to obtain a grayscale histogram. According to the grayscale histogram, the grayscale distribution in the area is determined to determine whether histogram equalization is required. If the grayscale distribution in the area is uneven and the grayscale difference is small, an adaptive histogram equalization algorithm is used to process the area. The grayscale distribution in the area is adjusted by an adaptive histogram equalization algorithm to enhance the local contrast and make the grayscale difference between the lesion area and the surrounding tissue more obvious. After all local areas are subjected to adaptive histogram equalization, the processed local areas are recombined into a complete cancer image to obtain a contrast-enhanced cancer image, highlighting the characteristics of the lesion area and providing clearer image data.

[0058] Specifically, obtaining cancer image datasets is an important starting point for medical image processing. Taking breast cancer X-ray images as an example, the original images usually contain complex tissue structures and tiny calcifications. Converting color images to grayscale images can reduce computational complexity while retaining key structural information. For example, a 1024x1024 pixel color breast X-ray image may have a 24-bit color depth. After converting to an 8-bit grayscale image, the amount of data is reduced by two-thirds, but the key information of tissue density and calcifications is still retained. Block processing is an effective way to improve image enhancement. Taking a 512x512 pixel grayscale breast X-ray image as an example, it can be divided into 64 64x64 pixel sub-blocks. Each sub-block is independently histogram analyzed and equalized, which can better adapt to the grayscale characteristics of the local area. For example, in an area containing microcalcifications, the improvement of local contrast can make these tiny lesions more obvious. Calculating the grayscale histogram is a key step in evaluating image contrast. Taking a 64x64 pixel sub-block as an example, assume that its grayscale level is 256. Statistics show that the grayscale values ​​of this sub-block are mainly concentrated between 100-150, indicating that the contrast in this area is low. In this case, the application of histogram equalization can significantly improve the readability of the image. Adaptive histogram equalization (AHE) is an efficient local contrast enhancement method. For the aforementioned low-contrast sub-blocks, the AHE algorithm redistributes the grayscale values ​​so that they are more evenly distributed in the range of 0-255. This can highlight the details that were originally difficult to distinguish. For example, two adjacent pixels with grayscale values ​​of 120 and 125 may become 100 and 150 after AHE processing, greatly improving the contrast between them. The reassembly of the processed sub-blocks into a complete image is the last key step. In order to avoid abrupt transitions at the boundaries of sub-blocks, overlapping areas and smooth transitions are usually used. For example, adjacent sub-blocks can overlap by 8 pixels, and the grayscale values ​​are fused in the overlapping area by weighted averaging to ensure the overall continuity of the image. The advantage of this block-based adaptive histogram equalization method is that it can flexibly adjust the contrast according to local features. For breast cancer X-ray images, it can simultaneously enhance the overall contrast of large areas of tissue and the visibility of local tiny calcification points.

[0059] As an implementation in this embodiment, the multi-scale analysis method includes: using wavelet transform to perform multi-scale decomposition on the pre-processed cancer image to extract texture features, shape features and edge features;

[0060] The texture features, shape features and edge features extracted at different scales are fused, and a feature descriptor is constructed through feature fusion. The feature descriptor is then screened using the Fisher discriminant criterion to obtain the screened feature descriptor.

[0061] As an optional implementation in this embodiment, a preprocessed cancer image is obtained, and a multi-scale analysis method is used to extract texture, shape and edge features of the cancer image at different scales. A feature descriptor is constructed by fusing multi-scale features to improve the discriminative ability of feature representation.

[0062] As an additional implementation method in this embodiment, according to the pre-processed cancer image, the wavelet transform method is used to perform multi-scale decomposition on the image to obtain sub-images at different scales. For each sub-image at each scale, the texture features are extracted, and the gray level co-occurrence matrix method is used to calculate the texture feature parameters such as contrast, correlation, energy and entropy of the image. For each sub-image at each scale, the shape features are extracted, the contour of the image is obtained by the contour extraction algorithm, and the shape feature parameters such as the perimeter, area, and circularity of the contour are calculated. For each sub-image at each scale, the edge features are extracted, the Canny operator is used to detect the edge of the image, and the edge feature parameters such as the direction and intensity of the edge are calculated. The texture features, shape features and edge features extracted at different scales are fused to construct a multi-scale feature descriptor, and each feature parameter is combined into a feature vector in a series manner. The Fisher discriminant criterion is used to perform feature selection on the multi-scale feature descriptor, screen out a feature subset with strong discriminant ability, and reduce the feature dimension. The screened multi-scale feature descriptor is input into the support vector machine classifier for training and testing to obtain the classification result of the cancer image.

[0063] Specifically, wavelet transform is a multi-scale analysis method that can decompose an image into sub-images of different frequencies and spatial resolutions. For example, for a 512x512 pixel cancer image, 10 sub-images can be obtained through 3-level wavelet decomposition, including 1 low-frequency approximate sub-image and 9 high-frequency detail sub-images. This multi-scale decomposition helps capture the characteristics of the image at different scales. Gray-level co-occurrence matrix is ​​an effective method to describe texture features. Taking contrast features as an example, it reflects the clarity of the image and the depth of the grooves of the texture. For a 4x4 image block, its gray-level co-occurrence matrix in the 0-degree direction can be calculated, and then the contrast value can be obtained. A higher contrast value usually indicates that there is a significant grayscale change in the image, which may be related to the boundary of the cancer lesion. Shape features are very important for describing the geometric characteristics of cancer lesions. Taking circularity as an example, it measures the similarity of the target shape to a circle. Assuming that the outline of a cancer lesion is extracted, its circumference is 100 pixels and its area is 500 square pixels, its circularity can be calculated. A circularity close to 1 indicates that the lesion has a regular shape and may be a benign tumor; while a circularity far less than 1 may indicate the irregular growth of a malignant tumor. The Canny edge detection operator can effectively extract edge information from an image. For cancer images, edge features can reflect the boundary between the lesion and the surrounding normal tissue. For example, after processing with the Canny operator, a series of edge pixels can be obtained, each with its direction and intensity information. Areas with high edge intensity may correspond to the boundaries of cancer lesions, while sudden changes in edge direction may indicate irregular growth of lesions. Feature fusion is the combination of different types of features to obtain a more comprehensive image description. For example, the texture, shape, and edge features extracted from 10 sub-images can be concatenated into a high-dimensional vector. Assuming that 20 feature parameters are extracted from each sub-image, the final feature vector dimension will reach 200 dimensions. This high-dimensional feature vector contains multi-scale and multi-angle information of the image. The Fisher discriminant criterion is an effective feature selection method that selects the most discriminative features by maximizing the inter-class distance and minimizing the intra-class distance. For a 200-dimensional feature vector, only 50 of them may be truly useful for cancer analysis. Through the Fisher discriminant criterion, these key features can be screened out, which not only reduces the computational complexity but also improves the accuracy of classification. Support vector machine is a powerful classifier that is particularly suitable for processing high-dimensional feature spaces. In cancer analysis, a 50-dimensional feature vector after feature selection can be used as input to train a binary support vector machine model to distinguish between benign and malignant tumors. By training and validating on a large amount of labeled data, the support vector machine can learn an optimal classification hyperplane, thereby accurately classifying new cancer images.

[0064] As an implementation method in this embodiment, the process of segmenting the cancer image includes:

[0065] The region growing segmentation algorithm expands the segmentation region by dynamically adjusting the thresholds of grayscale and texture features, and optimizes the segmentation result in combination with morphological processing to obtain the boundary of the lesion region.

[0066] As an optional implementation in this embodiment, the cancer image is segmented using a region growing segmentation algorithm based on the extracted feature descriptors, the grayscale and texture features of the lesion area are used as growth criteria, the segmented area is adaptively expanded, and morphological processing is introduced to eliminate noise and artifacts generated during the segmentation process to obtain the lesion area boundary.

[0067] As an implementation method that can be added in this embodiment, the cancer image is preprocessed according to the extracted feature descriptor to enhance the contrast and clarity of the image and highlight the grayscale and texture features of the lesion area. The regional growth segmentation algorithm is adopted, and the grayscale and texture features of the lesion area are used as the growth criteria. Starting from the seed point, the pixels with similar grayscale and texture features are classified into the same area. In the process of regional growth, according to the changes in the grayscale and texture features of the lesion area, the threshold of the growth criterion is adaptively adjusted to dynamically expand the scope of the segmented area. The segmentation result is morphologically processed, and the noise and artifacts generated in the segmentation process are eliminated through corrosion and expansion operations, and the edge of the lesion area is smoothed. According to the segmentation result after morphological processing, the boundary of the lesion area is extracted to obtain an accurate outline of the lesion area. The extracted boundary of the lesion area is feature analyzed, and the morphological features such as the area, perimeter, and roundness of the lesion area are calculated to quantify the size and shape of the lesion area.

[0068] Specifically, cancer image preprocessing is a key step to improve image clarity. By enhancing contrast and clarity, the features of the lesion area can be highlighted. For example, histogram equalization of lung CT images can make the grayscale distribution of lung nodules more uniform and the edges clearer. This helps with subsequent segmentation and feature extraction. The region growing segmentation algorithm is a commonly used medical image segmentation method. Taking breast cancer ultrasound images as an example, the center point of the mass can be selected as the seed point, and the grayscale value and texture features can be used as growth criteria. Assuming that the grayscale value of the seed point is 150 and the texture feature is roughness 0.8, the pixels with grayscale values ​​between 140-160 and roughness between 0.7-0.9 in the adjacent pixels are classified into the same region. Adaptive adjustment of the growth criterion threshold can improve the accuracy of segmentation. For example, when segmenting tumors in liver CT images, as the region expands, edge regions with gradually changing grayscale values ​​may be encountered. At this time, the threshold range can be dynamically adjusted according to the average grayscale value of the segmented region, so that the threshold changes as the region grows, thereby more accurately capturing the tumor boundary. Morphological processing can optimize the segmentation results. Taking brain MRI image segmentation as an example, the brain tumor area obtained by the initial segmentation is corroded to remove some small misclassified areas; then the expansion operation is performed to fill the cavities inside the tumor and obtain a smoother and continuous tumor contour. Extracting the boundary of the lesion area is the basis for quantitative analysis. For skin cancer images, the edge tracking algorithm can be used to extract the contour of melanoma. By calculating the perimeter, area and other features of the contour, the following quantitative indicators can be obtained: the area is 500 square millimeters, the perimeter is 80 millimeters, and the roundness is 0.75. These indicators can help doctors assess the size and irregularity of the tumor. Morphological feature analysis provides an important basis for improving image clarity. For example, in ultrasound images of thyroid nodules, benign nodules are usually round or elliptical with regular boundaries; while malignant nodules are often irregular in shape and have blurred edges. By calculating the roundness, edge sharpness and other features of the nodules, an objective basis can be provided for the identification of benign and malignant nodules. Assuming that the roundness of a nodule is 0.9 and the edge sharpness is 0.8, this indicates that the nodule is more likely to be benign. Using segmentation results and morphological features as auxiliary decision-making basis can significantly improve image clarity. For example, in the process of analyzing breast cancer images, a scoring system can be constructed by combining multiple indicators such as the size, shape, and edge characteristics of the mass. Assuming that a patient's mass area is 200 square millimeters, the roundness is 0.6, and the edge irregularity is 0.7, the total score obtained according to the scoring system is 8 points, which exceeds the threshold of 6 points for distinguishing benign from malignant tumors, indicating that the mass may be malignant. Quantitative analysis of the images with improved clarity can provide strong support for doctors' analysis decisions and improve the accuracy and consistency of the analysis.

[0069] As an implementation method of this embodiment, based on the adaptive contrast enhancement technology, the enhancement parameters are automatically adjusted according to the grayscale statistical characteristics of the lesion area. The process of enhancing the contrast of the lesion area includes:

[0070] Obtaining segmented lesion region image data, performing grayscale statistical analysis on the lesion region image, and obtaining grayscale statistical characteristic parameters of the lesion region;

[0071] Determine the target parameter range of adaptive contrast enhancement according to the obtained grayscale statistical characteristic parameters of the lesion area;

[0072] Adopting adaptive threshold segmentation algorithm, the lesion area image is divided into multiple sub-areas. For each sub-area, the contrast enhancement parameters are automatically adjusted within the target parameter range according to its grayscale statistical characteristics.

[0073] Contrast enhancement is performed on each sub-region image, and all enhanced sub-region images are fused to obtain an overall enhanced lesion region image.

[0074] As an optional implementation in this embodiment, an adaptive contrast enhancement technology is used for the segmented lesion area. According to the grayscale statistical characteristics of the lesion area, the enhancement parameters are automatically adjusted to enhance the contrast of the lesion area in a targeted manner and highlight the display effect of the lesion.

[0075] As an additional implementation method in this embodiment, the segmented lesion area image data is obtained, and grayscale statistical analysis is performed on the lesion area image to obtain grayscale statistical characteristic parameters of the lesion area; the target parameter range of adaptive contrast enhancement is determined according to the obtained grayscale statistical characteristic parameters of the lesion area; the lesion area image is divided into multiple sub-areas by an adaptive threshold segmentation algorithm, and the contrast enhancement parameters are automatically adjusted within the target parameter range for each sub-area according to its grayscale statistical characteristics; contrast enhancement processing is performed on each sub-area image, and the grayscale difference of the enhanced sub-area image is greater, and the local contrast is improved; all enhanced sub-area images are fused to obtain an overall enhanced lesion area image, and compared with the original lesion area image, the contrast between the lesion and the surrounding tissue is significantly improved; the overall enhanced lesion area image is smoothed and denoised by a morphological processing algorithm to eliminate possible noise points and small holes, so that the lesion area contour is clearer; the processed lesion area image is fused with the original medical image, and the lesion area is highlighted in color or highlight to facilitate analysis by doctors.

[0076] Specifically, after obtaining the segmented lesion area image data, grayscale statistical analysis is first performed. The purpose of this step is to understand the grayscale distribution characteristics of the lesion area and provide a basis for subsequent processing. For example, by calculating the average grayscale value, standard deviation, grayscale histogram and other statistics of the lesion area, the grayscale range, central tendency and dispersion degree of the area can be obtained. Assume that the analysis results show that the average grayscale value of the lesion area is 120 and the standard deviation is 30. These data reflect the overall brightness level and grayscale change degree of the lesion area. According to the grayscale statistical characteristic parameters, the target parameter range of adaptive contrast enhancement is determined. This step is intended to set reasonable goals for subsequent contrast enhancement processing to avoid over-enhancement or under-enhancement. For example, the target parameter range can be set to ±20% of the average grayscale value, that is, between 96 and 144. Such a setting can not only improve the contrast of the lesion area, but also maintain the naturalness of the image. The lesion area is divided into multiple sub-areas using an adaptive threshold segmentation algorithm, which can better adapt to local grayscale changes. For example, the Otsu algorithm or the local adaptive threshold method can be used for segmentation. Assume that the lesion area is divided into 5×5 grids, and each grid is segmented using an independent threshold, so that the structural differences inside the lesion area can be captured more finely. For each sub-region, the contrast enhancement parameters are automatically adjusted within the target parameter range according to its grayscale statistical characteristics. This adaptive method can better handle the characteristics of different sub-regions. For example, for sub-regions with lower grayscale, a larger enhancement parameter can be used, while for sub-regions with higher grayscale, a smaller enhancement parameter can be used. Assuming that the average grayscale value of a sub-region is 80, which is lower than the overall average value, its enhancement parameter can be set to 1.5 times the baseline value. Contrast enhancement is performed on each sub-region image. Common methods include histogram equalization and gamma correction. For example, for the sub-region with an average grayscale value of 80, gamma correction can be applied, and a parameter with a gamma value of 0.7 can be used for enhancement, which can effectively improve the visibility of dark details. All enhanced sub-region images are fused to obtain an overall enhanced lesion region image. During the fusion process, attention should be paid to the smooth transition of the boundary to avoid obvious splicing marks. Weighted average or gradient-based fusion methods can be used to ensure that the grayscale changes between adjacent sub-regions are continuous and natural. The overall enhanced lesion area image is smoothed and denoised by morphological processing algorithms. Common morphological operations include opening and closing operations. For example, an erosion operation with a radius of 2 pixels can be performed first to remove small noise points, and then a dilation operation with a radius of 2 pixels can be performed to fill small holes, and finally a median filter can be performed to further smooth the image. These operations can effectively eliminate the noise that may be introduced during the contrast enhancement process while maintaining the integrity of the lesion area contour.Finally, the processed lesion area image is fused with the original medical image, and the lesion area is highlighted in color or highlight. For example, the enhanced lesion area can be marked in red and superimposed on the original grayscale image. Or the grayscale value of the lesion area can be mapped to the heat map color space using pseudo-color technology, making the lesion area more visually prominent. This visualization method can help doctors quickly locate and evaluate lesions. Through the above steps, the lesion features that may not be easily perceived are enhanced and highlighted, providing doctors with clearer and more valuable image information. This adaptive image enhancement and visualization method not only improves the sensitivity of lesion detection, but also maintains the overall naturalness of the image, avoiding the illusions that may be caused by over-processing.

[0077] The overall enhanced image of the lesion area is smoothed and denoised through a morphological processing algorithm to eliminate possible noise points and tiny holes, making the outline of the lesion area clearer.

[0078] The enhanced image of the lesion area is obtained and used as the input image of the morphological processing algorithm. For the input image, the morphological opening operation is used to eliminate the noise and small holes in the image by corroding and dilating the image. For the image processed by the opening operation, the morphological closing operation is further used to smooth the contour of the lesion area by dilating and corroding the image. According to the size and shape parameters of the preset structural elements, the parameters of the morphological operation are adaptively adjusted to obtain the optimal smoothing and denoising effect. The contour of the lesion area is extracted by edge detection of the processed image. The extracted contour of the lesion area is further morphologically processed, such as contour smoothing, contour refinement, etc., to make the contour line clearer and smoother. The lesion area image after a series of morphological processing is used as the output to obtain a smooth, denoised, and clear-contour lesion area image, which is ready for subsequent feature extraction and analysis.

[0079] Specifically, morphological processing is an image processing method based on mathematical morphology theory, which is mainly used for image smoothing, denoising and contour extraction. In the processing of lesion area images, morphological operations can effectively eliminate noise, fill holes and smooth contours. First, the enhanced lesion area image is processed by opening operation. The opening operation consists of an erosion operation followed by a dilation operation. The erosion operation can remove small objects and protrusions in the image, while the dilation operation can fill small holes and disconnected areas. For example, for a lung nodule image with noise, a circular structure element of appropriate size (such as a radius of 3 pixels) is selected, and an erosion operation is first performed to remove noise smaller than the structure element, and then a dilation operation is performed to restore the main structure, thereby achieving a denoising effect. Next, a closing operation is performed on the image after the opening operation. The closing operation consists of a dilation operation followed by an erosion operation, which can fill small holes inside the object and smooth the contour. For example, for a liver tumor image with an uneven contour, a square structure element of 5×5 pixels is used for closing operation, which can effectively fill the depression at the edge of the tumor and make the contour smoother. The parameter selection of the morphological operation is crucial to the processing effect. The size and shape of the structural element should be adaptively adjusted according to the characteristics of the lesion. For example, for lung nodules of different sizes, an adaptive algorithm can be designed to dynamically adjust the radius of the structural element according to the size of the nodule area to achieve more accurate morphological processing. After morphological processing, the contour of the lesion area needs to be extracted. Common edge detection algorithms include Sobel and Canny. Taking the Canny algorithm as an example, it can effectively extract the edge contour of the lesion through steps such as Gaussian filtering, gradient calculation, non-maximum suppression and double threshold detection. For a breast mass image that has been morphologically processed, the Canny algorithm can accurately depict the boundary of the mass. The extracted contour may still have some irregularities and need further processing. Contour smoothing can use methods such as B-spline curve fitting to connect discrete contour points into a smooth curve. Contour thinning can use skeleton extraction algorithms, such as the Zhang-Suen thinning algorithm, to thin the rough contour lines into lines with a single pixel width. Through this series of morphological processing, an image of the lesion area with reduced noise and clear contours can be obtained. This processing not only improves the visual quality of the image, but also lays the foundation for subsequent feature extraction and analysis. For example, in lung nodule detection, morphologically processed nodule images can more accurately calculate morphological features such as diameter and circumference, which helps to judge whether it is benign or malignant. In brain tumor segmentation, clear tumor contours can improve the accuracy of volume measurement and provide a reliable basis for the formulation of treatment plans.

[0080] As an implementation method in this embodiment, the wavelet transform-based denoising method suppresses noise by performing soft threshold processing on the wavelet coefficients of the high-frequency sub-band, while retaining the original coefficients of the low-frequency sub-band to preserve image details, and obtains the denoised image through inverse transformation.

[0081] As an optional implementation in this embodiment, taking into account the noise interference of cancer images, a denoising method based on wavelet transform is introduced into the enhanced image. Through multi-scale wavelet decomposition, soft threshold processing is performed on the wavelet coefficients in the high-frequency sub-band to suppress high-frequency noise. At the same time, the original coefficients are retained in the low-frequency sub-band to retain the detail information of the lesion area. The image is reconstructed through inverse transformation to improve the signal-to-noise ratio and clarity of the image.

[0082] As an additional implementation method in this embodiment, step 1: obtain a cancer image, and use a denoising method based on wavelet transform to process the noise interference in the image. Step 2: Perform multi-scale wavelet decomposition on the cancer image to obtain wavelet coefficients at different frequency scales. Step 3: According to the preset soft threshold, the wavelet coefficients of the high-frequency sub-band are soft-thresholded. If the absolute value of the wavelet coefficient is less than the soft threshold, the coefficient is set to zero, otherwise the coefficient is retained. This step can effectively suppress high-frequency noise. Step 4: In the low-frequency sub-band, the original wavelet coefficients are retained without any processing. The low-frequency sub-band contains the main energy and detail information of the image, and retaining this information helps to retain the detail features of the lesion area. Step 5: According to the processed high-frequency sub-band wavelet coefficients and the original coefficients of the low-frequency sub-band, the image is reconstructed by wavelet inverse transform to obtain a denoised cancer image. Step 6: Calculate the signal-to-noise ratio of the cancer image before and after denoising, and judge the denoising effect by comparing the degree of improvement of the signal-to-noise ratio. If the signal-to-noise ratio is significantly improved, it indicates that the denoising method is effective. Otherwise, the soft threshold needs to be adjusted and steps 3-5 are repeated. Step 7: Evaluate the clarity of the denoised cancer image and determine the clarity of the image by calculating the gradient information of the image. If the clarity meets the preset conditions, the final denoised and enhanced cancer image is output. Otherwise, return to step 1 to reacquire the image and process it.

[0083] Specifically, after obtaining cancer images, the first thing to solve is the noise problem that may exist in the image. This is like listening to someone talking in a noisy environment. You need to eliminate the background noise before you can hear the content of the speech clearly. The denoising method based on wavelet transform is similar to an intelligent "noise reduction headset" that can effectively remove noise from the image. The wavelet transform here can be understood as a special "filter" that can decompose the image into components of different frequencies, just like decomposing a song into high, medium and low notes. For example, suppose there is a lung CT image containing fine noise particles. These noise particles are like "noise points" on the photo, which will affect the doctor's observation of the lesion. Performing multi-scale wavelet decomposition is like sifting sand with sieves of different specifications. Components of different frequencies are like sand grains of different sizes and will be separated into different "sieves". For example, an image can be decomposed into four parts: a low-frequency part, which is similar to the overall outline of the image, like a fuzzy thumbnail; and three high-frequency parts, which represent details in the horizontal, vertical and diagonal directions, similar to the edges and textures of the image. The preset soft threshold can be understood as the noise reduction strength of the "noise reduction headphones". The soft threshold processes the wavelet coefficients of the high-frequency subband, just like removing the high-frequency noise in the song. If the intensity of a high-frequency component (the absolute value of the wavelet coefficient) is lower than the preset "noise standard" (soft threshold), it is considered to be noise and eliminated (set to zero); if it is higher than this standard, it is considered to be the real details of the image and needs to be retained. For example, a soft threshold can be set to 0.2. If the absolute value of a wavelet coefficient is 0.15, which is less than 0.2, it is set to zero; if the absolute value of another wavelet coefficient is 0.3, which is greater than 0.2, the coefficient is retained. The advantage of this is that it can effectively remove noise while retaining the details of the image as much as possible. The low-frequency subband, like the main melody in the song, contains the main information of the image. It is not processed to avoid damage to the main structure of the image and retain the main information of the image and the detailed features of the lesion area, such as the shape and size of the lesion. This is like when listening to a song, you need to remove noise while ensuring that the main melody is clear and discernible. Inverse wavelet transform can be understood as recombining components of different frequencies into the original image, just like recombining high, mid and low frequencies into a complete song. Through inverse wavelet transform, the processed high-frequency subband wavelet coefficients and low-frequency subband original coefficients can be recombined into a complete image, which is the denoised cancer image. The signal-to-noise ratio is an important indicator for evaluating the denoising effect, just like using a "clarity score" to measure the effect of noise-cancelling headphones. The higher the signal-to-noise ratio, the less noise in the image and the clearer the image. For example, the signal-to-noise ratio of the image before denoising is 20, and it increases to 30 after denoising, indicating that the denoising method is effective. If the signal-to-noise ratio is not significantly improved, it is necessary to adjust the soft threshold and reprocess it.Adjusting the soft threshold is like adjusting the noise reduction strength of noise reduction headphones. It needs to be fine-tuned according to the actual situation to achieve the best denoising effect. Clarity assessment is achieved by calculating the gradient information of the image. The gradient can be understood as the rate of change of the pixel value in the image. The larger the gradient, the clearer the image. For example, the difference between each pixel in the image and its surrounding pixels can be calculated, and then these differences can be statistically analyzed to obtain the clarity index of the image. If the clarity meets the preset conditions, such as the clarity index is greater than 0.8, the denoised image is considered to be clear enough for subsequent analysis. Otherwise, the image needs to be reacquired and processed until an image with qualified clarity is obtained. In this way, layer by layer, one link after another, a high-quality denoised and enhanced cancer image is finally obtained, providing a reliable basis for the doctor's analysis.

[0084] According to the processed high-frequency subband wavelet coefficients and the low-frequency subband original coefficients, the image is reconstructed by inverse wavelet transform to obtain the denoised cancer image.

[0085] Obtain the cancer image data to be processed, preprocess the image data, including image format conversion, image size adjustment and other operations, and obtain standardized cancer image data. Select a suitable wavelet basis function, perform wavelet transform on the standardized cancer image data, decompose the image into high-frequency sub-bands and low-frequency sub-bands, and obtain the wavelet coefficients of the high-frequency sub-bands and the original coefficients of the low-frequency sub-bands. According to the statistical characteristics of the high-frequency sub-band wavelet coefficients, design a suitable threshold function, perform threshold processing on the high-frequency sub-band wavelet coefficients, remove high-frequency noise, and obtain the processed high-frequency sub-band wavelet coefficients. The processed high-frequency sub-band wavelet coefficients are fused with the original coefficients of the low-frequency sub-band, and the image is reconstructed by inverse wavelet transform to obtain the denoised cancer image. Perform quality assessment on the denoised cancer image, calculate the image signal-to-noise ratio, peak signal-to-noise ratio and other indicators, and judge whether the image denoising effect meets the requirements. If it does not meet the requirements, adjust the wavelet basis function or threshold function parameters and perform image denoising again. The denoised cancer images were compared with the original cancer images, the differences before and after the images were denoised were analyzed, the performance of the denoising algorithm was evaluated, and the algorithm was optimized and improved based on the evaluation results. The denoised cancer images were output and saved for subsequent cancer analysis, and the parameter settings and processing results of the image denoising were recorded to provide a reference for subsequent denoising of similar images.

[0086] Specifically, the first step in cancer image processing is to obtain and preprocess image data. For example, for breast cancer X-ray mammography images, it may be necessary to convert the DICOM format to a common PNG or JPEG format and uniformly adjust the image size to 1024x1024 pixels for subsequent processing. Wavelet transform is a multi-resolution analysis method that can effectively separate different frequency components of an image. In cancer image denoising, commonly used wavelet basis functions include Haar, Daubechies, and Symlets. Taking Daubechies4 (db4) as an example, it can better balance time and frequency positioning and is suitable for medical image processing. After the image is wavelet transformed, the high-frequency subband contains the details and noise information of the image, while the low-frequency subband retains the main structure of the image. For the wavelet coefficients of the high-frequency subband, a soft threshold or hard threshold function can be designed for processing. The soft threshold function is usually smoother and can reduce the pseudo-Gibbs phenomenon in the image. For example, the threshold can be set to 3 times the standard deviation of the high-frequency subband coefficient, and the coefficients with an absolute value less than the threshold are shrunk, while the coefficients greater than the threshold remain unchanged. When reconstructing the image, the processed high-frequency subband coefficients are combined with the retained low-frequency subband coefficients, and the denoised image is obtained by inverse wavelet transform. This method can effectively remove high-frequency noise while retaining important structural information in the image, such as key features such as tumor edges. The evaluation of denoising effect usually adopts a combination of objective indicators and subjective evaluation. Objective indicators include signal-to-noise ratio (SNR) and peak signal-to-noise ratio (PSNR). For example, if the PSNR of the denoised image is improved by more than 3dB, it is generally considered that the denoising effect is significant. If the indicator does not meet the standard, you can try to adjust the wavelet basis function (such as from db4 to db6) or modify the threshold function parameters. By comparing the images before and after denoising, the algorithm performance can be intuitively evaluated. For example, in lung cancer CT images, the edges and internal structures of lung nodules should be clearly displayed after denoising without over-smoothing out important key information. This comparative analysis helps to optimize algorithm parameters, such as adjusting the form or parameter value of the threshold function. Finally, the processed image is saved in a standard format (such as 16-bit TIFF) to retain sufficient grayscale information. At the same time, the parameters such as the wavelet basis function, decomposition layer number, threshold function, etc. used are recorded to provide a reference for subsequent research. These processed images can be used in computer-aided diagnosis (CAD) systems to improve the accuracy of early cancer detection. The purpose of the entire denoising process is to improve the quality of cancer images so that doctors can more accurately identify and locate lesion areas. By removing interfering noise, key features in the image, such as the clarity of the tumor boundary, are enhanced, which in turn helps to improve the accuracy and timeliness of the analysis. However, the key point of this embodiment is to improve the clarity of the image, and improving the accuracy of the analysis is a reasonable assumption.This image preprocessing technology plays an important role in early cancer screening and precise diagnosis and treatment, and lays the foundation for subsequent tasks such as image segmentation, feature extraction and lesion classification.

[0087] As an implementation method in this embodiment, a parallel computing framework based on a graphics processor is adopted to assign computing tasks such as image preprocessing, feature extraction, segmentation, contrast enhancement and denoising to the graphics processor for execution. A corresponding parallelization strategy is designed for each algorithm step to fully utilize the parallel computing capabilities of the graphics processor and improve the execution efficiency of the algorithm.

[0088] As an optional implementation in this embodiment, according to the computing task of image processing, a parallel computing framework is adopted to make full use of the parallel computing capability of the graphics processor and design a corresponding parallelization strategy. For the image preprocessing task, the efficiency of image preprocessing is improved by dividing the image into blocks and assigning the divided sub-images to different threads of the graphics processor for parallel processing. For the feature extraction task, a parallelization algorithm based on the graphics processor is adopted to assign the computing task of feature extraction to multiple threads of the graphics processor for parallel execution, thereby accelerating the feature extraction process. For the image segmentation task, the image is divided into multiple sub-regions and the sub-regions are assigned to different threads of the graphics processor for parallel segmentation to improve the efficiency of image segmentation. For the contrast enhancement task, a parallel algorithm based on histogram equalization is adopted to accelerate the contrast enhancement processing by assigning the histogram calculation and pixel mapping to multiple threads of the graphics processor for parallel execution. For the image denoising task, a parallel denoising algorithm based on wavelet transform is adopted to improve the efficiency of image denoising by assigning the computing tasks of wavelet decomposition and reconstruction to multiple threads of the graphics processor for parallel execution. According to different image processing tasks, the parallelization strategy and task allocation method are dynamically adjusted to make full use of the parallel computing resources of the graphics processor and maximize the execution efficiency of the algorithm.

[0089] As an additional implementation method in this embodiment, specifically, in image processing tasks, the use of a parallel computing framework can significantly improve processing efficiency. Taking image preprocessing as an example, a 4K resolution image can be divided into 16 1080p sub-images, and each sub-image is assigned to a thread of the graphics processor for processing. This method can shorten the processing time from the original 1 minute to about 4 seconds, greatly improving efficiency. In feature extraction tasks, a parallelization algorithm based on a graphics processor can be used. For example, when extracting SIFT features, the image can be divided into multiple overlapping local areas, and each area is assigned to a thread for feature point detection and descriptor calculation. This parallel strategy can shorten the feature extraction time from the traditional 10 seconds to less than 1 second, with an acceleration ratio of more than 10 times. Image segmentation tasks can be parallelized by dividing the image into multiple sub-areas. Taking superpixel segmentation as an example, a 4K image can be divided into 16 sub-areas, and each sub-area is independently superpixel segmented by a thread. This method can reduce the segmentation time from the original 30 seconds to about 2 seconds, significantly improving segmentation efficiency. Contrast enhancement tasks can use a parallel algorithm based on histogram equalization. First, the image is divided into multiple blocks, and each block is calculated by a thread to calculate the local histogram. Then, all local histograms are merged to obtain the global histogram. Finally, multiple threads perform pixel mapping in parallel. This parallel strategy can reduce the processing time from the original 5 seconds to less than 0.5 seconds, improving the efficiency by 10 times. The image denoising task can adopt a parallel denoising algorithm based on wavelet transform. For example, for an 8K resolution image, it can be divided into 64 1080p sub-images. Each sub-image is independently subjected to wavelet decomposition, threshold processing and reconstruction by a thread. This parallel method can reduce the denoising time from the original 2 minutes to about 5 seconds, greatly improving the processing efficiency. Dynamic adjustment of the parallelization strategy is crucial to fully utilize the resources of the graphics processor. For example, when processing low-resolution images, the number of sub-regions can be reduced and the workload of each thread can be increased to avoid excessive thread scheduling overhead. On the contrary, for high-resolution images, the number of sub-regions can be increased to fully utilize the parallel capabilities of the graphics processor. Through this dynamic adjustment, the best performance can be obtained in image processing tasks of different scales. In general, the use of a parallel computing framework can significantly improve the efficiency of image processing. Through reasonable task allocation and parallelization strategies, the computing resources of the graphics processor can be fully utilized and the processing time can be greatly shortened. This not only improves the speed of image processing, but also makes real-time image processing and large-scale image data analysis possible.

[0090] Based on this, an embodiment of the present invention provides a cancer image contrast enhancement method based on deep learning. Aiming at the problem of insufficient contrast of cancer images, an adaptive histogram equalization algorithm is used for preprocessing to enhance the contrast of local areas. Texture, shape and edge features are extracted through multi-scale analysis to construct feature descriptors. Based on the feature descriptor, the image is segmented using the regional growth segmentation algorithm to obtain the boundary of the lesion area. An adaptive contrast enhancement technique is used for the segmented lesion area to highlight the lesion display effect. Taking into account noise interference, a denoising method based on wavelet transform is introduced to suppress noise in high-frequency sub-bands and retain details in low-frequency sub-bands. The present invention also adopts a parallel computing framework based on a graphics processor to assign each computing task to a graphics processor for execution, thereby improving the efficiency of the algorithm. Through the organic combination of the above technical solutions, the present invention can effectively enhance the contrast and clarity of cancer images, highlight the lesion area, and suppress noise interference.

[0091] Embodiment 2

[0092] In this embodiment, an electronic device is provided, including a memory and a processor. The memory stores a computer program, and the processor is configured to run the computer program to execute the method in the above embodiment.

[0093] The above program can be run in the processor, or it can be stored in the memory (or computer-readable medium), which includes permanent and non-permanent, removable and non-removable media. Information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0094] These computer programs can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps of the functions specified in one or more blocks can be implemented by different modules corresponding to different steps.

[0095] like Figure 2 As shown, this embodiment provides such a system, which is called a cancer image contrast enhancement system based on deep learning, including:

[0096] A preprocessing module, used for acquiring a cancer image, and preprocessing the cancer image based on an adaptive histogram equalization algorithm;

[0097] A multi-scale feature extraction module, for extracting features at different scales for the pre-processed cancer image based on a multi-scale analysis method, and constructing a feature descriptor by fusing the multi-scale features, wherein the features include texture features, shape features and edge features;

[0098] A region growing segmentation module is used to segment the cancer image based on the feature descriptor using a region growing segmentation algorithm, taking the grayscale and texture features of the lesion area as growth criteria, adaptively expanding the segmentation area, and eliminating segmentation noise and artifacts through morphological processing;

[0099] The contrast enhancement module is used to automatically adjust the enhancement parameters and enhance the contrast of the lesion area based on the adaptive contrast enhancement technology according to the grayscale statistical characteristics of the lesion area after segmentation;

[0100] The wavelet denoising module is used to denoise the enhanced image by using a wavelet transform-based denoising method. Through multi-scale wavelet decomposition, soft threshold processing is performed on the wavelet coefficients in the high-frequency sub-band to suppress high-frequency noise, while retaining the original coefficients in the low-frequency sub-band, and reconstructing the image through inverse transformation to obtain the final image with contrast enhancement.

[0101] As an implementation method in this embodiment, the region growing segmentation module optimizes the contour of the segmented region by combining dynamic threshold adjustment and morphological processing.

[0102] As an implementation method in this embodiment, it also includes a parallel computing module, which is used to adopt a parallel computing framework based on a graphics processor to assign computing tasks such as image preprocessing, feature extraction, segmentation, contrast enhancement and denoising to the graphics processor for execution, and design a corresponding parallelization strategy for each algorithm step to fully utilize the parallel computing capabilities of the graphics processor and improve the execution efficiency of the algorithm.

[0103] The system or device is used to implement the functions of the method in the above-mentioned embodiment. Each module in the system or device corresponds to each step in the method, which has been explained in the method and will not be repeated here.

[0104] Through the above implementation, the problems of differences in imaging quality and low contrast in the lesion area in the related technology are solved, thereby ensuring that the contrast and clarity of the cancer image are effectively enhanced, the lesion area is highlighted, and noise interference is suppressed.

[0105] The present invention also provides a computer terminal device, comprising:

[0106] one or more processors;

[0107] A memory, coupled to the processor, for storing one or more programs;

[0108] When the one or more programs are executed by the one or more processors, the one or more processors implement the steps of a cancer image contrast enhancement method based on deep learning.

[0109] The present invention also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of a method for contrast enhancement of cancer images based on deep learning are implemented.

[0110] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A cancer image contrast enhancement method based on deep learning, characterized in that: The following steps are involved: Acquire a cancer image, and preprocess the cancer image based on an adaptive histogram equalization algorithm; For the preprocessed cancer image, based on the multi-scale analysis method, features are extracted at different scales, and a feature descriptor is constructed by fusing the multi-scale features, wherein the features include texture features, shape features and edge features; Based on the feature descriptor, a region growing segmentation algorithm is used to segment the cancer image, the grayscale and texture features of the lesion area are used as growth criteria, the segmentation area is adaptively expanded, and segmentation noise and artifacts are eliminated through morphological processing; For the segmented lesion area, based on the adaptive contrast enhancement technology, the enhancement parameters are automatically adjusted according to the grayscale statistical characteristics of the lesion area to enhance the contrast of the lesion area; For the enhanced image, a denoising method based on wavelet transform is adopted. Through multi-scale wavelet decomposition, soft threshold processing is performed on the wavelet coefficients in the high-frequency sub-band to suppress high-frequency noise, while retaining the original coefficients in the low-frequency sub-band, and reconstructing the image through inverse transform to obtain the final image with contrast enhancement.

2. The method according to claim 1, characterized in that The process of preprocessing the cancer image using an adaptive histogram equalization algorithm includes: Obtain a cancer image dataset and convert the images into grayscale; Perform block processing on the grayscale image, divide the image into several local areas, and calculate the grayscale histogram of each area; For areas with small grayscale differences, an adaptive histogram equalization algorithm is used to adjust the grayscale distribution and enhance the local contrast; All local regions are combined to obtain the complete cancer image after preprocessing.

3. The method according to claim 1, characterized in that The multi-scale analysis method comprises: performing multi-scale decomposition on the pre-processed cancer image using wavelet transform to extract texture features, shape features and edge features; The texture features, shape features and edge features extracted at different scales are fused, and a feature descriptor is constructed through feature fusion. The feature descriptor is then screened using the Fisher discriminant criterion to obtain the screened feature descriptor.

4. The method according to claim 1, characterized in that: The process of segmenting cancer images includes: The region growing segmentation algorithm expands the segmentation region by dynamically adjusting the thresholds of grayscale and texture features, and optimizes the segmentation result in combination with morphological processing to obtain the boundary of the lesion region.

5. The method according to claim 1, characterized in that Based on the adaptive contrast enhancement technology, the enhancement parameters are automatically adjusted according to the grayscale statistical characteristics of the lesion area. The process of enhancing the contrast of the lesion area includes: Obtaining segmented lesion region image data, performing grayscale statistical analysis on the lesion region image, and obtaining grayscale statistical characteristic parameters of the lesion region; Determine the target parameter range of adaptive contrast enhancement according to the obtained grayscale statistical characteristic parameters of the lesion area; Adopting adaptive threshold segmentation algorithm, the lesion area image is divided into multiple sub-areas. For each sub-area, the contrast enhancement parameters are automatically adjusted within the target parameter range according to its grayscale statistical characteristics. Contrast enhancement is performed on each sub-region image, and all enhanced sub-region images are fused to obtain an overall enhanced lesion region image.

6. The method according to claim 1, characterized in that The denoising method based on wavelet transform suppresses noise by performing soft threshold processing on wavelet coefficients of high-frequency subbands, while retaining original coefficients of low-frequency subbands to retain image details, and obtains a denoised image through inverse transformation.

7. A cancer image contrast enhancement system based on deep learning, characterized in that: The system comprises: A preprocessing module, used for acquiring a cancer image, and preprocessing the cancer image based on an adaptive histogram equalization algorithm; A multi-scale feature extraction module, for extracting features at different scales for the pre-processed cancer image based on a multi-scale analysis method, and constructing a feature descriptor by fusing the multi-scale features, wherein the features include texture features, shape features and edge features; A region growing segmentation module is used to segment the cancer image based on the feature descriptor using a region growing segmentation algorithm, taking the grayscale and texture features of the lesion area as growth criteria, adaptively expanding the segmentation area, and eliminating segmentation noise and artifacts through morphological processing; The contrast enhancement module is used to automatically adjust the enhancement parameters and enhance the contrast of the lesion area based on the adaptive contrast enhancement technology according to the grayscale statistical characteristics of the lesion area after segmentation; The wavelet denoising module is used to denoise the enhanced image by using a wavelet transform-based denoising method. Through multi-scale wavelet decomposition, soft threshold processing is performed on the wavelet coefficients in the high-frequency sub-band to suppress high-frequency noise, while retaining the original coefficients in the low-frequency sub-band, and reconstructing the image through inverse transformation to obtain the final image with contrast enhancement.

8. The system according to claim 7, characterized in that The region growing segmentation module optimizes the contour of the segmented region by combining dynamic threshold adjustment with morphological processing.

9. A computer terminal device, characterized in that: include: one or more processors; A memory, coupled to the processor, for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the various steps of the cancer image contrast enhancement method based on deep learning as described in any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the cancer image contrast enhancement method based on deep learning as described in any one of claims 1 to 6 is implemented.

Images