Feature processing method, device, equipment and medium based on medical images

Abstract

The present application provides a feature processing method, device, equipment and medium based on medical images. The method includes: determining the initial texture features of each lesion sub-region extracted from the lesion area of ​​the medical image, and the representative texture features corresponding to the lesion area; determining the connectivity relationship between each lesion sub-region based on the texture feature value of each lesion sub-region facing the representative texture features; based on the connectivity relationship, combining the initial texture features of each lesion sub-region to obtain the connectivity texture features of the lesion area. The present application implements the correlation analysis of the internal features of the lesion area, ensures the comprehensive effectiveness of the texture features in the lesion area, solves the problem of insufficient comprehensiveness of the texture feature extraction in the lesion area in the prior art, and further improves the accuracy of lesion classification in medical images.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of image processing technology, and specifically to a feature processing method, device, equipment and medium based on medical images. Background Art

[0002] In recent years, when a lesion is identified in a medical image, in order to accurately analyze the degree of deterioration of the lesion, the corresponding texture features are usually extracted from the lesion area of ​​the medical image to quantitatively analyze the tissue structural heterogeneity in the lesion area, thereby discovering the degree of deterioration of the lesion area.

[0003] At present, for the lesion area in medical images, the multi-scale stationary wavelet transform (SWT) algorithm is usually used to decompose and reconstruct the lesion area at different scales to obtain multi-resolution and multi-level texture feature maps in the horizontal and vertical directions. However, the existing feature processing mainly extracts texture features from the entire lesion area, but lacks correlation analysis between the internal features of the lesion area, and cannot fully and effectively extract the texture features in the lesion area, which leads to a decrease in the accuracy of analyzing the degree of lesion deterioration. Summary of the invention

[0004] The present application provides a feature processing method, device, equipment and storage medium based on medical images, which realizes the correlation analysis of the internal features of the lesion area according to the connectivity between each lesion sub-area in the lesion area, ensures the comprehensive effectiveness of the texture features in the lesion area, and thus improves the accuracy of lesion classification in medical images.

[0005] In a first aspect, an embodiment of the present application provides a feature processing method based on medical images, the method comprising:

[0006] Determining initial texture features of each lesion sub-region extracted from the lesion region of the medical image, and representative texture features corresponding to the lesion region;

[0007] Determining the connectivity relationship between the lesion sub-regions according to the texture feature value of each lesion sub-region facing the representative texture feature;

[0008] Based on the connectivity relationship, the initial texture features of each of the lesion sub-regions are combined to obtain the connectivity texture features of the lesion region.

[0009] In a second aspect, an embodiment of the present application provides a feature processing device based on medical images, the device comprising:

[0010] A texture feature analysis module, used to determine the initial texture features of each lesion sub-region extracted from the lesion region of the medical image, and the representative texture features corresponding to the lesion region;

[0011] A connectivity determination module, configured to determine connectivity relationships between the lesion sub-regions according to a texture feature value of each lesion sub-region facing the representative texture feature;

[0012] A connectivity feature processing module is used to combine the initial texture features of each of the lesion sub-regions based on the connectivity relationship to obtain the connectivity texture features of the lesion region.

[0013] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising:

[0014] A processor and a memory, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute the feature processing method based on medical images provided in the first aspect of the present application.

[0015] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium for storing a computer program, wherein the computer program enables a computer to execute the medical image-based feature processing method provided in the first aspect of the present application.

[0016] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program / instructions, characterized in that when the computer program / instructions are executed by a processor, a feature processing method based on medical images as provided in the first aspect of the present application is implemented.

[0017] The embodiments of the present application provide a feature processing method, apparatus, device and medium based on medical images. First, each lesion sub-region is extracted from the lesion area of ​​the medical image, and the initial texture features of each lesion sub-region and the representative texture features corresponding to the lesion area are determined. Then, according to the texture feature values ​​of each lesion sub-region facing the representative texture features, the connectivity relationship between the lesion sub-regions is determined, and based on the connectivity relationship, the initial texture features of each lesion sub-region are combined to obtain the connectivity texture features of the lesion area, thereby realizing the correlation analysis of the internal features of the lesion area, ensuring the comprehensive effectiveness of the texture features in the lesion area, solving the problem of insufficient comprehensiveness in the extraction of texture features in the lesion area in the prior art, and further improving the accuracy of lesion classification in medical images. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 A flowchart of a feature processing method based on medical images shown in an embodiment of the present application;

[0020] Figure 2 This is a schematic diagram of extracting a lesion sub-region within a lesion region shown in an embodiment of the present application;

[0021] Figure 3 A schematic diagram showing the connectivity of each lesion sub-region in the lesion region shown in an embodiment of the present application;

[0022] Figure 4 A flowchart of another feature processing method based on medical images shown in an embodiment of the present application;

[0023] Figure 5 A schematic diagram of an initial texture feature extraction process of a lesion sub-region shown in an embodiment of the present application;

[0024] Figure 6 A schematic diagram of a representative texture feature determination process of a lesion area shown in an embodiment of the present application;

[0025] Figure 7 A principle block diagram of a feature processing device based on medical images shown in an embodiment of the present application;

[0026] Figure 8 It is a schematic block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0029] The present application is applicable to various medical images that use texture analysis to assist in mining potential features inside lesions. Considering that the existing feature processing in medical images mainly extracts texture features from the entire lesion area in the medical image, there is a lack of correlation analysis between the internal features of the lesion area, which leads to the texture features of the lesion area being not comprehensive enough. Therefore, the present application designs a method for comprehensively extracting texture features of the lesion area in a medical image, by dividing the lesion area into multiple lesion sub-areas, and then performing correlation analysis on the texture features between the lesion sub-areas to combine and obtain the texture features of the lesion area, thereby ensuring the comprehensive effectiveness of the texture features in the lesion area and further improving the accuracy of lesion classification in medical images.

[0030] Figure 1 This is a flowchart of a feature processing method based on medical images shown in an embodiment of the present application. Figure 1 , the method may specifically include the following steps:

[0031] S110, determining initial texture features of each lesion sub-region extracted from the lesion region of the medical image, and representative texture features corresponding to the lesion region.

[0032] According to medical experiments and medical theoretical foundations related to various lesions, different types of lesions will have different texture analysis values ​​in the corresponding lesion areas. Therefore, when analyzing the texture features of lesions in any medical image, the corresponding lesion area will first be segmented from the medical image according to the lesion annotation in the medical image.

[0033] Exemplarily, the medical image in the present application can be a two-dimensional image or a three-dimensional image, which is not limited. If the medical image is a two-dimensional image, then the subsequent lesion feature analysis process can be directly performed. However, if the medical image is a three-dimensional image, then it is first necessary to perform mean processing on the pixel features in a certain specific dimensional direction according to the various dimensional directions of the medical image, so as to perform two-dimensional processing on the medical image, so that the medical image is converted into a two-dimensional image, and then the corresponding lesion feature analysis process is performed on the converted two-dimensional image. At this time, for any two-dimensional medical image, the corresponding lesion annotation can be performed in the medical image through the lesion analysis process. Then, the medical image after the lesion annotation is binarized to accurately highlight the lesion contour in the medical image, and obtain the binarized annotated image. Finally, the medical image and the binarized annotated image are subjected to pixel grayscale processing, and the corresponding lesion area can be extracted from the medical image.

[0034] As an optional implementation scheme in the present application, in order to accurately analyze the correlation between the internal features of the lesion area, a pre-set regional division standard with texture analysis value in the lesion area can be determined based on the medical clinical experiments and medical theoretical basis of the relevant lesions. Furthermore, according to the regional division standard set in the lesion area, multiple lesion sub-areas with texture analysis value can be further extracted from the lesion area, so that the texture correlation between the internal features of the lesion area can be analyzed by subsequently performing texture feature analysis on each lesion sub-area. Figure 2 As shown, for any lesion area m in a medical image i , can be in m i k regions of interest with texture analysis value are extracted from the image, that is, the lesion sub-region R={r i1 ,r i2 ,L,r ik Then, by performing texture feature analysis on each lesion sub-region, the initial texture feature of each lesion sub-region can be obtained.

[0035] At the same time, considering that there are certain commonalities in the factors causing the same type of lesions, the lesion area will also have some texture features that can represent the significant differences between the lesion area and other areas, which are the representative texture features in this application.

[0036] In this application, a large number of medical image datasets P = {p1, p2, L, p nThen, the same texture feature extraction method as that for each lesion sub-region in the lesion region in the present application is adopted to analyze the lesion region of each medical image in the data set P respectively, and the corresponding initial texture features are extracted from each lesion sub-region in the lesion region. Furthermore, by performing texture feature significance analysis on the initial texture features of each lesion sub-region in the lesion region of each medical image in the data set P, the first few initial texture features that can best represent the texture difference between the lesion region and other regions are screened out as representative texture features of the lesion region.

[0037] S120, determining the connectivity relationship between the lesion sub-regions according to the texture feature value of each lesion sub-region facing the representative texture feature.

[0038] After determining the initial texture features of each lesion sub-region in the lesion region of the medical image, it is considered that the representative texture features of the lesion region can highlight the significant differences between the lesion region and other regions. Therefore, in order to accurately measure the correlation between the internal features of the lesion region, the application can perform feature analysis on the initial texture features of each lesion sub-region and the representative texture features, so as to calculate the texture feature value of each lesion sub-region facing the representative texture feature, and the texture feature value can represent the information level of the initial texture features of each lesion sub-region on the representative texture feature. At this time, according to the actual position of each lesion sub-region in the lesion region, the texture feature values ​​of each lesion sub-region are combined to obtain the eigenvalue matrix corresponding to the lesion region.

[0039] Then, in the eigenvalue matrix corresponding to the lesion area, by judging whether each lesion sub-area is connected in the eigenvalue matrix and whether the texture eigenvalues ​​of each lesion sub-area are the same, the connectivity relationship between each lesion sub-area can be determined. Among them, if any two lesion sub-areas have a common edge or a common angle in the eigenvalue matrix, then it can be determined that the two lesion sub-areas are connected. For example, if the texture eigenvalue of any lesion sub-area in the eigenvalue matrix corresponding to the lesion area is the same as the texture eigenvalue of a lesion sub-area in the 8 other lesion sub-areas connected around it, then it is determined that the two are connected.

[0040] For example, Figure 3 From the eigenvalue matrix shown in , it can be determined that the lesion sub-regions connected at 4 positions with a texture eigenvalue of 5 are connected, and the lesion sub-regions connected at 2 positions with a texture eigenvalue of 3 are connected.

[0041] S130, combining the initial texture features of each lesion sub-region based on the connectivity relationship to obtain a connectivity texture feature of the lesion region.

[0042] After determining the connectivity between the lesion sub-regions in the lesion region, considering that there are certain commonalities between the texture features of the connected lesion sub-regions, Figure 3 As shown, the present application can merge the initial texture features of the connected corresponding lesion sub-regions in sequence to obtain the connected texture features of the lesion region. At this time, within the connected texture features, the initial texture features of each lesion sub-region with common texture features can be closely arranged together to comprehensively and effectively represent the correlation between the internal features of the lesion region, thereby ensuring the accuracy of the connected texture features of the lesion region.

[0043] In addition, in order to accurately classify the level of the lesion area, the present application can obtain a large number of historical medical images that match the texture analysis value of the lesion area in the medical image in the present application as corresponding training samples. Then, using the texture feature processing method of the connected texture features of the above-mentioned lesion area, the corresponding sample lesion area can be extracted from each training sample, and the connected texture features of the sample lesion area in each training sample can be determined. Furthermore, on the initial deep residual network (Deep residual network, referred to as ResNet), the ResNet network is trained using the connected texture features of the sample lesion area in each training sample, and the lesion classification model in the present application can be constructed for identifying the level of the lesion area in any medical image.

[0044] Among them, the ResNet network of the present application adds residual units through a short-circuit mechanism to form residual learning. Because of the scalability of features and the explosion of feature dimensions when extracting features based on different parameters, the increase in network parameters will cause gradient explosion and disappearance, resulting in the failure of the model to converge. Therefore, the ResNet network allows it to be deepened as much as possible, but it directly sends part of the input data to the output without passing through the convolutional network, retaining part of the original information. This structure effectively prevents the gradient diffusion problem during back propagation, so that the network depth can reach hundreds of layers or even deeper, and also solves the problem of feature dimension explosion. Because the lesion sub-areas and classifications are far less than the ResNet network, the ResNet network is used as the basic architecture, the network depth is reduced, and the convolution kernel size is reduced. At this time, the parameter settings of the convolution layer can be 16@3×3, 16@2×2, and 32@2×2, respectively, and the step size is 1, and the linear correction unit (Rectified Liner Units, referred to as ReLu) is used as the activation function. The convolutional neural network combines the local perception area of ​​the two-dimensional connected texture features, shares weights, and fully utilizes the connected locality of the data itself in space. Through the convolution operation, the various local features of the two-dimensional connected texture features and the connected edge and contour features of the connected texture features can be extracted. The relationship between the internal and external regional features of the common features can be effectively extracted, providing more favorable information for distinguishing the grade of the lesion. The pooling layer uses a maximum pooling of 2×2 with a step size of 2. The role of the pooling layer is to merge similar features semantically, mainly extracting the texture and edges of the two-dimensional feature connection. The pooling layer makes the features have good suppression of noise and abnormal feature values ​​through the pooling operation. The information extracted by the bottom layer of the pooling layer is transmitted to the next layer through neuron connections, and gradually transmitted to the upper layer, thereby extracting the most essential structural information of the connected texture features of the lesion area. The number of nodes of the two fully connected layers is 64 and 32 respectively, and the dropout layer is used during training to randomly ignore some neurons, which can also avoid overfitting. Finally, the logistic regression (softmax) algorithm is used to output the classification results. Each neuron in the fully connected layer is fully connected to the previous layer. This layer combines the various local features extracted in the early stage, and finally obtains the posterior probability of each category through the output layer.

[0045] Therefore, after constructing the lesion classification model in the above manner, the connected texture features of the lesion area in the medical image in this application can be input into the constructed lesion classification model, and the connected texture features of the lesion area can be analyzed accordingly through the lesion classification model, so as to output the lesion classification results in the medical image, thereby further improving the accuracy of lesion grade analysis.

[0046] The technical solution provided in the embodiment of the present application first extracts each lesion sub-region from the lesion region of the medical image, and determines the initial texture features of each lesion sub-region and the representative texture features corresponding to the lesion region, then determines the connectivity relationship between the lesion sub-regions based on the texture feature values ​​of each lesion sub-region facing the representative texture features, and based on the connectivity relationship, combines the initial texture features of each lesion sub-region to obtain the connectivity texture features of the lesion region, thereby realizing the correlation analysis of the internal features of the lesion region, ensuring the comprehensive effectiveness of the texture features in the lesion region, solving the problem of insufficient comprehensiveness in the extraction of texture features in the lesion region in the prior art, and further improving the accuracy of lesion classification in medical images.

[0047] As an optional implementation scheme in the embodiment of the present application, the present application can provide a detailed exemplary description of each of the above steps. Figure 4 FIG. 1 is a flowchart of another feature processing method based on medical images shown in an embodiment of the present application. Figure 4 As shown, the method may specifically include the following steps:

[0048] S410, extracting each lesion sub-region from the lesion region based on the texture analysis value of the lesion region in the medical image; performing a multi-level stationary wavelet transform on each lesion sub-region to obtain an initial texture feature of the lesion sub-region.

[0049] When performing texture feature analysis on each lesion sub-region in the lesion region, the present application can first determine the regional division criteria of the lesion region with texture analysis value pre-set by experienced medical experimenters based on the medical clinical experiments and medical theoretical basis of the relevant lesions. Then, according to the regional division criteria set in the lesion region, multiple lesion sub-regions with texture analysis value can be further extracted from the lesion region.

[0050] Then, considering that the multi-scale stationary wavelet transform (SWT) algorithm can perform multi-resolution and multi-level feature decomposition of the input image at different scales in the horizontal and vertical directions, thereby decomposing and peeling off the image information layer by layer, it not only conforms to the receptive field cell structure of the visual system, but also can obtain the best joint resolution of the deep frequency domain and the spatial domain, which is very suitable for the representation and discrimination of image texture. Therefore, after extracting each lesion sub-region from the lesion area, the present application can extract the initial texture features of each lesion sub-region by multi-level stationary wavelet transform of each lesion sub-region.

[0051] For example, Figure 5 As shown, for each lesion sub-region r ij(1≤j≤k), n-level two-dimensional SWT decomposition can be performed. At this time, at each decomposition level, the lesion sub-region can be decomposed into four frequency bands: low-frequency component LL, horizontal high-frequency component LH, vertical high-frequency component HL, and diagonal high-frequency component HH. Considering that the high-frequency components decomposed in the horizontal, vertical, and diagonal directions represented by LH, HL, and HH can express the significant information that there are obvious differences in texture features between the lesion sub-region and other regions, when continuing to perform SWT decomposition on the next level at each decomposition level, only the LL low-frequency component at the decomposition level can continue to be decomposed by SWT, and the cycle is repeated until the n-level SWT decomposition of the lesion sub-region is completed. Then, by extracting the radiomics texture features of the lesion sub-region at each frequency component at each decomposition level after decomposition, the initial texture features of the lesion sub-region are obtained, and the initial texture features are a texture vector at this time.

[0052] The above steps are performed on each lesion sub-region to obtain the initial texture features of each lesion sub-region.

[0053] S420, obtaining historical medical images that match the texture analysis value of the lesion area in the medical image as corresponding training samples; extracting corresponding sample lesion sub-regions from the sample lesion area of ​​each training sample, and determining the initial texture features of each sample lesion sub-region; performing feature screening on the initial texture features of each sample lesion sub-region to obtain representative texture features corresponding to the lesion area.

[0054] In order to accurately analyze the texture features in the lesion area that can represent the significant difference between the lesion area and other areas, the present application can collect a large number of historical medical images with the same texture analysis value as the lesion area in the medical image, such as a medical image dataset P = {p1, p2, L, p n}, as the training sample in this application. Then, from each training sample p i Extract the corresponding sample lesion area m i , and from the sample lesion area m i Extract multiple sample lesion sub-regions r ij Then, by adopting the same steps of extracting the initial texture features of each lesion sub-region in the lesion region of the medical image, the texture features of each sample lesion sub-region r can be determined. ij The initial texture feature t ij .

[0055] Then, in order to analyze the correlation between the initial texture features of each sample lesion sub-region in each training sample, the present application can analyze the initial texture features of each sample lesion sub-region in each training sample. i Each sample lesion sub-region r ijThe horizontal correlation and position relationship, or the vertical correlation and position relationship, are analyzed, and each sample lesion area m i Each sample lesion sub-region r ij The initial texture features of are uniformly arranged into corresponding two-dimensional texture features, that is,

[0056] Furthermore, by analyzing The similarity between any two initial texture features within is used to analyze the lesion sub-regions r of each sample ij Then, according to the correlation between the initial texture features of each sample lesion sub-region, multiple initial texture features with large feature differences from other texture features can be screened out from the initial texture features of each sample lesion sub-region as representative texture features corresponding to the lesion region in this application.

[0057] As an optional implementation scheme in the present application, feature screening is performed on the initial texture features of each sample lesion sub-region to obtain representative texture features corresponding to the lesion region. Specifically, for the sample lesion region of each training sample, based on the texture feature similarity between every two sample lesion sub-regions in the sample lesion region, the difference texture features corresponding to the training sample are screened out from the initial texture features between each sample lesion sub-region in the sample lesion region; based on the total number of times each difference texture feature is screened under each training sample, the representative texture features corresponding to the lesion region are determined.

[0058] That is to say, for each training sample, the sample lesion area m i In this application, the degree of correlation between the initial texture features of any two sample lesion sub-regions can be measured by comparing the absolute value of the Pearson correlation coefficient between the initial texture features of the two sample lesion sub-regions.

[0059] For example, for each training sample, the sample lesion area m i Overall texture features after internal merging and arrangement Any two initial texture features can be calculated by the following formula: and The Pearson correlation coefficient between:

[0060]

[0061] Then, in calculating Any two initial texture features and After the Pearson correlation coefficients between them, we can calculate the Pearson correlation coefficients (ρ i12 ,ρ i13,L,ρ ik-1,k ) are arranged in descending order to indicate the degree of correlation between the two initial texture features. Then, the initial texture features with the leading Pearson correlation coefficient greater than the preset threshold H are excluded, and the remaining initial texture features with a smaller degree of correlation are retained as the difference texture features of each sample lesion sub-region with feature differences between the sample lesion region and other sample lesion sub-regions in this application, which can be recorded as Where c is the feature dimension after preliminary feature screening of the initial texture features of each sample lesion sub-region, so that the initial texture features with internal structural differences can be screened out in each sample lesion sub-region and the dimension of the texture features can be reduced to a relatively small scale.

[0062] Optionally, in order to ensure the accuracy of the difference texture features, after using the Pearson correlation coefficient to perform preliminary feature screening on the initial texture features of each sample lesion sub-region, the present application will further use the minimum absolute shrinkage and selection operator lasso to perform secondary feature screening on the difference texture features obtained after the preliminary screening. The loss function used at this time is: Among them, y is the sample label, α is the regularization coefficient, and the L1 norm can promote the sparsity of features and avoid overfitting. At this time, the difference texture features of each sample lesion sub-region after secondary feature screening can be recorded as: Where r is the feature dimension after secondary feature screening of the initial texture features of each sample lesion sub-region.

[0063] Finally, for the training sample dataset P = {p1, p2, L, p n Each sample lesion sub-region r in the sample lesion region of each training sample in} ij , perform the above-mentioned feature screening steps. Then, calculate the total number of occurrences of each difference texture feature remaining after feature screening in each training sample, as the total number of times each difference texture feature is screened in this application. Furthermore, according to the total number of times each remaining difference texture feature is screened, the first few difference texture features with the highest total number of times can be screened as the representative texture features corresponding to the lesion area, which are used to represent the texture features with the best significance in each lesion sub-area within the lesion area. For example, Figure 6 As shown, the present application can be obtained from the difference texture features after secondary feature screening The top five difference texture features with the highest total number of screening times are screened out as the representative texture features F = {f1, f2, f3, f4, f5} of the lesion area in this application.

[0064] It should be noted that in the present application, there is no specific execution order between S410 and S420. They can be executed simultaneously or sequentially, and the present application does not limit this.

[0065] S430, using a preset interval feature threshold, converting the texture feature value of each lesion sub-region oriented toward the representative texture feature to obtain the converted texture feature value of each lesion sub-region.

[0066] As an optional implementation scheme in the present application, since there are multiple representative texture features of the lesion area, and the significance of the texture difference between the lesion area and other areas represented by each representative texture feature is different. Therefore, in the screening process of representative texture features, the present application has sorted each representative texture feature according to the order of the significance of each representative texture feature for the texture features between the lesion area and other areas. Moreover, in the screening process of representative texture features, the texture feature value of the initial texture feature of each lesion sub-area under the current representative texture feature has been calculated. At this time, the texture feature value of each lesion sub-area under the current representative texture feature can be directly obtained, so that the feature value can be converted to it later using a preset interval feature threshold.

[0067] Since the texture feature values ​​of each lesion sub-region are relatively scattered, in order to ensure the connectivity accuracy of the lesion region, the present application can pre-set a corresponding interval feature threshold for the texture feature value of each lesion sub-region. For example, if the texture feature value of each lesion sub-region is a value between 0 and 100, then the texture feature value in the interval (0, 10) can be converted to the feature threshold 1 of the region, and the texture feature value in the interval (11, 20) can be converted to the feature threshold 2 of the region, and so on, so that the texture feature values ​​in each interval can be converted to a unified feature threshold of the region, so as to facilitate the subsequent connectivity judgment of each lesion sub-region according to its texture feature value.

[0068] In the present application, when preliminarily determining the texture feature value of each lesion sub-region facing a certain representative texture feature, the texture feature value of the lesion sub-region can be converted into a feature threshold set in a pre-set interval by judging in which interval the texture feature value of each lesion sub-region is located, thereby obtaining the converted texture feature value of the lesion sub-region.

[0069] S440, determining a connected region within the lesion region according to the actual position of each lesion sub-region within the lesion region and the converted texture feature value of each lesion sub-region.

[0070] By analyzing whether there is a common edge or common angle between the actual positions of any two lesion sub-regions in each lesion sub-region, it is determined whether the two lesion sub-regions are connected. At the same time, whether the texture feature values ​​of the two lesion sub-regions after conversion are the same is combined to determine whether the two lesion sub-regions are connected. At this time, if the actual positions of any two lesion sub-regions in the lesion area are connected, and the texture feature values ​​of the two lesion sub-regions after conversion are the same, then it can be determined that the two lesion sub-regions are connected. According to the above steps, the connected area composed of the corresponding lesion sub-regions in the lesion area can be determined. For example, Figure 3 The lesion sub-regions connected at 4 positions with a texture eigenvalue of 5 are connected, and the lesion sub-regions connected at 2 positions with a texture eigenvalue of 3 are connected.

[0071] S450, for each connected region, combining the initial texture features of each lesion sub-region in the connected region to obtain the connected texture features of the connected region; combining the connected texture features of each connected region in order according to the size of the texture feature values ​​of each connected region to obtain the connected texture features of the lesion region.

[0072] After determining the connected regions within the lesion area, considering that each connected region is composed of at least one lesion sub-region, for each connected region, the initial texture features of the lesion sub-regions in the connected region can be arranged and combined in sequence to obtain the connected texture features of the connected region.

[0073] Then, after the connected texture features of each connected region are obtained, the connected texture features of each connected region can be combined in order according to the size of the texture feature values ​​of each connected region, so as to obtain the connected texture features of the lesion region.

[0074] As an optional implementation scheme in the present application, considering that there are multiple lesion sub-regions with the same texture feature values ​​after conversion, but the actual positions are not connected, resulting in the formation of multiple connected regions with the same texture feature values, at this time, when the connected texture features of each connected region are combined in sequence according to the size of the texture feature values ​​of each connected region, the front and back order of each connected region with the same texture feature value cannot be determined. Therefore, the present application can continue to calculate the texture feature values ​​of each lesion sub-region facing the next representative texture feature for the texture feature values ​​of each lesion sub-region facing the current representative texture feature. Then, for each connected region determined when the texture feature values ​​of each lesion sub-region facing the current representative texture feature are the same, the connected texture features of the two connected regions can be combined in sequence according to the size of the texture feature values ​​of each connected region after conversion under the next representative texture feature.

[0075] Exemplarily, for the representative texture feature F={f1, f2, f3, f4, f5}, the texture feature value of each lesion sub-region after conversion under the representative texture feature f1 can be calculated first, and then it can be judged whether each lesion sub-region is connected, and each connected region in the lesion region is determined. At this time, when the connected texture features of each connected region are combined in order according to the size of the texture feature value of each connected region, if there are multiple connected regions with the same texture feature value, it is necessary to continue to calculate the texture feature value of each lesion sub-region in these connected regions after conversion under the representative texture feature f2, so as to continue to combine the connected texture features of these connected regions in order according to the size of the texture feature value of these connected regions after conversion under the representative texture feature f2. If the texture feature values ​​of these connected regions after conversion under the representative texture feature f2 are still the same, then the texture feature values ​​of each lesion sub-region in these connected regions after conversion under the representative texture feature f3 are continued to be calculated until the initial texture features of each lesion sub-region in each connected region are completely combined into the connected texture features of the lesion region.

[0076] The technical solution provided in the embodiment of the present application first extracts each lesion sub-region from the lesion region of the medical image, and determines the initial texture features of each lesion sub-region and the representative texture features corresponding to the lesion region, then determines the connectivity relationship between the lesion sub-regions based on the texture feature values ​​of each lesion sub-region facing the representative texture features, and based on the connectivity relationship, combines the initial texture features of each lesion sub-region to obtain the connectivity texture features of the lesion region, thereby realizing the correlation analysis of the internal features of the lesion region, ensuring the comprehensive effectiveness of the texture features in the lesion region, solving the problem of insufficient comprehensiveness in the extraction of texture features in the lesion region in the prior art, and further improving the accuracy of lesion classification in medical images.

[0077] Figure 7 FIG. 1 is a block diagram of a feature processing device based on medical images shown in an embodiment of the present application. Figure 7 As shown, the apparatus 700 may include:

[0078] The texture feature analysis module 710 is used to determine the initial texture features of each lesion sub-region extracted from the lesion region of the medical image, and the representative texture features corresponding to the lesion region;

[0079] A connectivity determination module 720, configured to determine connectivity relationships between the lesion sub-regions according to a texture feature value of each lesion sub-region facing the representative texture feature;

[0080] The connectivity feature processing module 730 is used to combine the initial texture features of each of the lesion sub-regions based on the connectivity relationship to obtain the connectivity texture features of the lesion region.

[0081] Furthermore, the texture feature analysis module 710 may include:

[0082] The initial texture analysis unit is used to extract each lesion sub-region from the lesion region based on the texture analysis value of the lesion region in the medical image; perform multi-level stationary wavelet transform on each lesion sub-region to obtain the initial texture features of the lesion sub-region.

[0083] Furthermore, the texture feature analysis module 710 may also include:

[0084] The representative texture analysis unit is used to obtain historical medical images that match the texture analysis value of the lesion area in the medical image as corresponding training samples; extract the corresponding sample lesion sub-area from the sample lesion area of ​​each training sample, and determine the initial texture features of each sample lesion sub-area; perform feature screening on the initial texture features of each sample lesion sub-area to obtain the representative texture features corresponding to the lesion area.

[0085] Furthermore, the representative texture analysis unit may be specifically used for:

[0086] For each sample lesion region of a training sample, based on the texture feature similarity between every two sample lesion sub-regions within the sample lesion region, a differential texture feature corresponding to the training sample is screened out from the initial texture features between the sample lesion sub-regions within the sample lesion region;

[0087] Based on the total number of times each difference texture feature is screened under each of the training samples, a representative texture feature corresponding to the lesion area is determined.

[0088] Furthermore, the connectivity determination module 720 may be specifically configured to:

[0089] Using a preset interval feature threshold, converting the texture feature value of each lesion sub-region facing the representative texture feature to obtain the converted texture feature value of each lesion sub-region;

[0090] A connected area within the lesion area is determined according to the actual position of each lesion sub-area within the lesion area and the converted texture feature value of each lesion sub-area.

[0091] Furthermore, the connectivity feature processing module 730 may be specifically used for:

[0092] For each connected region, combining the initial texture features of the lesion sub-regions in the connected region to obtain a connected texture feature of the connected region;

[0093] According to the texture feature values ​​of each of the connected regions, the connected texture features of each of the connected regions are combined in order to obtain the connected texture features of the lesion region.

[0094] Furthermore, the medical image-based feature processing device 700 may further include:

[0095] A lesion classification module, used for inputting the connected texture features of the lesion area into a pre-built lesion classification network to obtain a lesion classification result in the medical image;

[0096] Among them, the training samples of the lesion classification network are historical medical images that match the texture analysis value of the lesion area in the medical image, and the lesion classification network is obtained by training the connected texture features of the sample lesion area in each of the training samples.

[0097] In an embodiment of the present application, each lesion sub-region is first extracted from the lesion region of the medical image, and the initial texture features of each lesion sub-region and the representative texture features corresponding to the lesion region are determined. Then, based on the texture feature values ​​of each lesion sub-region facing the representative texture features, the connectivity relationship between the lesion sub-regions is determined, and based on the connectivity relationship, the initial texture features of each lesion sub-region are combined to obtain the connected texture features of the lesion region, thereby realizing the correlation analysis of the internal features of the lesion region, ensuring the comprehensive effectiveness of the texture features in the lesion region, solving the problem of insufficient comprehensiveness of the texture feature extraction in the lesion region in the prior art, and further improving the accuracy of lesion classification in medical images.

[0098] It should be understood that the device embodiment and the method embodiment may correspond to each other, and similar descriptions may refer to the method embodiment. To avoid repetition, they will not be described here. Specifically, Figure 7 The device 700 shown can execute any method embodiment provided in the present application, and the aforementioned and other operations and / or functions of each module in the device 700 are respectively for implementing the corresponding processes in each method of the embodiment of the present application. For the sake of brevity, they will not be repeated here.

[0099] The above describes the device 700 of the embodiment of the present application from the perspective of the functional module in conjunction with the accompanying drawings. It should be understood that the functional module can be implemented in hardware form, can be implemented by instructions in software form, and can also be implemented by a combination of hardware and software modules. Specifically, the steps of the method embodiment in the embodiment of the present application can be completed by the hardware integrated logic circuit and / or software form instructions in the processor, and the steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to perform, or a combination of hardware and software modules in the decoding processor to perform. Optionally, the software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, an electrically erasable programmable memory, a register, etc. The storage medium is located in a memory, and the processor reads the information in the memory, and completes the steps in the above method embodiment in conjunction with its hardware.

[0100] Figure 8 It is a schematic block diagram of an electronic device 800 provided in an embodiment of the present application.

[0101] like Figure 8 As shown, the electronic device 800 may include:

[0102] The memory 810 and the processor 820, the memory 810 is used to store the computer program and transmit the program code to the processor 820. In other words, the processor 820 can call and run the computer program from the memory 810 to implement the method in the embodiment of the present application.

[0103] For example, the processor 820 may be configured to execute the above method embodiments according to instructions in the computer program.

[0104] In some embodiments of the present application, the processor 820 may include but is not limited to:

[0105] General-purpose processor, digital signal processor (DSP), application-specific integrated circuit (ASIC), field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware components, etc.

[0106] In some embodiments of the present application, the memory 810 includes but is not limited to:

[0107] Volatile memory and / or non-volatile memory. Among them, the non-volatile memory can be read-only memory (ROM), programmable ROM (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM) or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct RAM bus random access memory (DR RAM).

[0108] In some embodiments of the present application, the computer program may be divided into one or more modules, which are stored in the memory 810 and executed by the processor 820 to complete the method provided by the present application. The one or more modules may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device.

[0109] like Figure 8 As shown, the electronic device may also include:

[0110] The transceiver 830 may be connected to the processor 820 or the memory 810 .

[0111] The processor 820 may control the transceiver 830 to communicate with other devices, specifically, to send information or data to other devices, or to receive information or data sent by other devices. The transceiver 830 may include a transmitter and a receiver. The transceiver 830 may further include an antenna, and the number of antennas may be one or more.

[0112] It should be understood that the various components in the electronic device are connected via a bus system, wherein the bus system includes not only a data bus but also a power bus, a control bus and a status signal bus.

[0113] The embodiment of the present application also provides a computer storage medium on which a computer program is stored, and when the computer program is executed by a computer, the computer can perform the method of the above method embodiment. In other words, the embodiment of the present application also provides a computer program product containing instructions, and when the instructions are executed by a computer, the computer can perform the method of the above method embodiment.

[0114] When software is used for implementation, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function according to the embodiment of the present application is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from a website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more available media integration. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a digital video disc (digital video disc, DVD)), or a semiconductor medium (e.g., a solid state drive (solid state disk, SSD)), etc.

[0115] Those of ordinary skill in the art will appreciate that the modules and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0116] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the module is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or modules, which can be electrical, mechanical or other forms.

[0117] The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. For example, each functional module in each embodiment of the present application may be integrated into a processing module, or each module may exist physically separately, or two or more modules may be integrated into one module.

[0118] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which 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 feature processing method based on medical images, characterized in that: include: Determining initial texture features of each lesion sub-region extracted from the lesion region of the medical image, and representative texture features corresponding to the lesion region; Determining the connectivity relationship between the lesion sub-regions according to the texture feature value of each lesion sub-region facing the representative texture feature; Based on the connectivity relationship, combining the initial texture features of each of the lesion sub-regions to obtain a connectivity texture feature of the lesion region; The step of determining the initial texture features of each lesion sub-region extracted from the lesion region of the medical image includes: Extracting each of the lesion sub-regions from the lesion region based on the texture analysis value of the lesion region in the medical image; Performing a multi-level stationary wavelet transform on each of the lesion sub-regions to obtain initial texture features of the lesion sub-region; The determining of the representative texture features corresponding to the lesion area includes: Obtain historical medical images that match the texture analysis value of the lesion area in the medical image as corresponding training samples; Extracting corresponding sample lesion sub-regions from the sample lesion region of each training sample, and determining initial texture features of each sample lesion sub-region; Performing feature screening on the initial texture features of each sample lesion sub-region to obtain representative texture features corresponding to the lesion region; The determining the connectivity relationship between the lesion sub-regions according to the texture feature value of each lesion sub-region facing the representative texture feature includes: Using a preset interval feature threshold, converting the texture feature value of each lesion sub-region facing the representative texture feature to obtain the converted texture feature value of each lesion sub-region; A connected area within the lesion area is determined according to the actual position of each lesion sub-area within the lesion area and the converted texture feature value of each lesion sub-area.

2. The method according to claim 1, characterized in that The feature screening of the initial texture features of each sample lesion sub-region to obtain representative texture features corresponding to the lesion region includes: For each sample lesion region of a training sample, based on the texture feature similarity between every two sample lesion sub-regions within the sample lesion region, a differential texture feature corresponding to the training sample is screened out from the initial texture features between the sample lesion sub-regions within the sample lesion region; Based on the total number of times each difference texture feature is screened under each of the training samples, a representative texture feature corresponding to the lesion area is determined.

3. The method according to claim 1, characterized in that The combining the initial texture features of each lesion sub-region based on the connectivity relationship to obtain the connectivity texture features of the lesion region includes: For each connected region, combining the initial texture features of the lesion sub-regions in the connected region to obtain a connected texture feature of the connected region; According to the texture feature values ​​of each of the connected regions, the connected texture features of each of the connected regions are combined in order to obtain the connected texture features of the lesion region.

4. The method according to claim 1, characterized in that: After combining the initial texture features of the lesion sub-regions based on the connectivity relationship to obtain the connectivity texture features of the lesion region, the method further includes: Inputting the connected texture features of the lesion area into a pre-constructed lesion classification network to obtain a lesion classification result in the medical image; Among them, the training samples of the lesion classification network are historical medical images that match the texture analysis value of the lesion area in the medical image, and the lesion classification network is obtained by training the connected texture features of the sample lesion area in each of the training samples.

5. A feature processing device based on medical images, characterized in that: include: A texture feature analysis module, used to determine the initial texture features of each lesion sub-region extracted from the lesion region of the medical image, and the representative texture features corresponding to the lesion region; A connectivity determination module, configured to determine connectivity relationships between the lesion sub-regions according to a texture feature value of each lesion sub-region facing the representative texture feature; A connectivity feature processing module, used for combining the initial texture features of each of the lesion sub-regions based on the connectivity relationship to obtain a connectivity texture feature of the lesion region; Wherein, the texture feature analysis module is specifically used for: Extracting each of the lesion sub-regions from the lesion region based on the texture analysis value of the lesion region in the medical image; Performing a multi-level stationary wavelet transform on each of the lesion sub-regions to obtain initial texture features of the lesion sub-region; The texture feature analysis module is further specifically used for: Acquire historical medical images that match the texture analysis value of the lesion area in the medical image as corresponding training samples; Extracting corresponding sample lesion sub-regions from the sample lesion region of each training sample, and determining initial texture features of each sample lesion sub-region; Performing feature screening on the initial texture features of each sample lesion sub-region to obtain representative texture features corresponding to the lesion region; The connectivity determination module is specifically used to: Using a preset interval feature threshold, converting the texture feature value of each lesion sub-region facing the representative texture feature to obtain the converted texture feature value of each lesion sub-region; A connected area within the lesion area is determined according to the actual position of each lesion sub-area within the lesion area and the converted texture feature value of each lesion sub-area.

6. An electronic device, characterized in that: include: A processor and a memory, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute the feature processing method based on medical images as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that: Used to store a computer program, wherein the computer program enables a computer to execute the feature processing method based on medical images as described in any one of claims 1 to 4.

Images