Chest DR image quality evaluation method based on radiomics characteristics

Abstract

The invention discloses a chest DR image quality evaluation method based on radiomics characteristics. The method comprises the following steps: firstly, collecting a chest DR image with a quality grade label and a region-of-interest mask label, and constructing a training data set; performing semantic segmentation to obtain a region of interest of the chest DR image, performing feature extraction to obtain image omics features of the chest DR image, and constructing an image omics feature set of the training data set; performing feature selection by using a variance analysis method and a recursive feature elimination method with cross validation to obtain an optimal feature subset of the training data set; and finally, constructing a chest DR image quality automatic evaluation model based on an integrated learning gradient boosting decision tree classifier, and training by using the optimal feature subset of the training data set to realize quality grade evaluation of the chest DR image. The subjective visual perception of the expert is quantitatively described by using the image omics characteristics, and the segmentation precision and the evaluation performance are improved through the segmentation network.

Description

Technical Field

[0001] The present invention belongs to the technical field of image evaluation and deep learning, and specifically relates to a chest DR image quality evaluation method based on radiomics features. Background Art

[0002] Chest DR (Digital Radiograph) image quality assessment plays an important role in ensuring diagnostic accuracy, patient safety, and standardized medical processes. For example, high-quality chest X-ray images help radiologists carefully observe anatomical structures and accurately diagnose diseases, thereby reducing the risk of misdiagnosis and missed diagnosis. Traditional chest DR image quality assessment mainly relies on the radiology quality control team to subjectively evaluate the images according to established standards. However, with the increase in the amount of shooting, this highly repetitive and high-profile evaluation work requires a lot of time and energy and cannot be carried out on a regular basis. In addition, the evaluation process needs to comprehensively consider multiple quality standard points closely related to diagnostic information, such as photographic position, image clarity, and contrast. The evaluation results are easily affected by the experience of the evaluator, resulting in poor evaluation consistency. As science and technology continue to advance towards digitalization and intelligence, the design of objective and efficient automatic evaluation methods for chest DR images has also received increasing attention.

[0003] There are two main types of existing chest DR image quality assessment methods: one is to use an end-to-end deep learning model to extract deep learning features to evaluate image quality; the other is to use a two-step cascade strategy: first use a deep learning model to obtain the region of interest in the image, and then use the existing prior knowledge to extract relevant features to build a quality assessment model. Among them, the first type of method mainly considers using a deep learning model to directly obtain deep learning features in the image for quantitative description of quality problems in the chest DR imaging process; this type of method usually uses a transfer learning strategy to reduce the risk of overfitting caused by the limited number of images, and uses an end-to-end method to train the deep learning model to automatically obtain image quality assessment results; for example, the literature (Jin J, Chen X, Meng Y, et al. Deep Blind Chest X-Ray Image Quality Assessment with ROI-Guided Attention [J]. IEEE MultiMedia, 2023.) adds a dual attention mechanism guided by the region of interest to the model to strengthen the consideration of important areas when evaluating image quality. However, the deep learning features extracted by this method have poor interpretability, which is a significant disadvantage for tasks in the medical field that require high interpretability. Although there are some technologies that provide powerful tools to improve interpretability, such as class activation maps (CAMs), these tools often only provide coarse-grained explanations. In addition, since the differences between images of different quality levels in the chest DR image quality assessment task are small, obtaining highly expressive deep learning features often requires a large amount of data to train the deep learning model. Although transfer learning can partially alleviate the risk of overfitting, the model may still overfit the training data when the data volume is insufficient or the data set is poorly representative.

[0004] The second type of method uses a two-step cascade strategy. This type of method usually first obtains the region of interest related to the quality problems of chest DR images, and then uses existing prior knowledge and digital image processing technology to obtain manual features to quantitatively characterize the quality problems in the chest DR images, and builds a machine learning model for image quality assessment, or directly evaluates the corresponding local quality problems in the image after obtaining the manual features; for example, the literature (Meng Y, Ruan J, Yang B, et al. Automated quality assessment of chest radiographs based on deep learning and linear regression cascade algorithms [J]. European Radiology, 2022, 32 (11): 7680-7690.) uses an automatic chest radiograph quality assessment method based on deep learning and linear regression cascade algorithms. First, the deep learning model is used to segment the chest and scapula areas and detect 16 key points. This information is used to construct 10 quantitative indicators, and these indicators are input into a multivariate linear regression model to predict the image quality assessment scores of the chest radiograph in terms of position and layout. However, there are still the following shortcomings: 1) This type of method has a relatively single perspective on the quality assessment problem: there are many quality standard points closely related to the diagnostic quality of chest DR images, including photography position, image clarity, contrast, artifact impact and other quality standard points closely related to diagnostic information, and this type of method only starts from the image position and layout related to the photography position, and fails to achieve full evaluation. 2) This type of method often needs to segment the region of interest first, but because the adjacent regions with similar density are not distinguished in the X-ray image and some segmentation targets are small and have various shapes, the segmentation accuracy is not high when using the U-Net network with weak context information capture ability and small target positioning and detection ability for semantic segmentation, which affects the subsequent quality assessment. 3) Although the linear regression model used in this type of method is simple and easy to use, it assumes that the relationship between variables is linear, which has limitations in capturing the relationship between quantitative features and image quality levels, and the linear regression model cannot fully utilize the interaction and complex association between quantitative indicators, resulting in a decrease in the accuracy of the assessment. Therefore, chest DR image quality assessment still faces some technical difficulties, which restricts its application and implementation. Summary of the invention

[0005] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and provide a method for evaluating the quality of chest DR images based on radiomics features.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] On the one hand, a method for evaluating chest DR image quality based on radiomics features is provided, comprising the following steps:

[0008] Collect chest DR images with quality grade labels and region of interest mask labels, and construct a training data set; the quality grades include excellent, good, fair, and poor;

[0009] Using the trained fully supervised multi-label semantic segmentation model to perform semantic segmentation on the chest DR images in the training data set to obtain the region of interest of the chest DR images; the region of interest includes the lung field area, the overlapping area of ​​the scapula and the lung field, the clavicle area, and the external metal artifact area;

[0010] The imaging omics features of the chest DR image are obtained by extracting features of the region of interest, and the imaging omics feature set of the training data set is constructed, including: extracting features of the lung field area to obtain the first-order statistical features and texture features of the chest DR image; capturing semantic posture features from the lung field area, the overlapping area of ​​the scapula and the lung field, the clavicle area, and the external metal artifact area to obtain the morphological features of the chest DR image; the imaging omics feature set includes a first-order statistical feature set, a texture feature set, and a morphological feature set;

[0011] The feature selection of the radiomics feature set of the training dataset was performed using the variance analysis method and the recursive feature elimination method with cross-validation to obtain the optimal feature subset of the training dataset;

[0012] An automatic chest DR image quality assessment model was constructed based on the gradient boosting decision tree classifier based on ensemble learning, and the optimal feature subset of the training data set was used for training to achieve quality grade assessment of chest DR images.

[0013] As a preferred technical solution, the fully supervised multi-label semantic segmentation model is a semantic segmentation model with receptive field enhancement with dilated convolution built on the basis of U-Net architecture, including an encoder and a decoder;

[0014] The encoder comprises a plurality of downsampling convolution blocks, each of which is followed by a receptive field enhancement module comprising a dilated convolution;

[0015] The decoder comprises a plurality of upsampling convolution blocks; the decoder is followed by a Sigmoid activation function;

[0016] The number of the downsampling convolution blocks is the same as the number of the upsampling convolution blocks.

[0017] As a preferred technical solution, the downsampling convolution block consists of two convolution blocks including 3×3 convolution, ReLU activation function and batch normalization, and a maximum pooling layer with a stride of 2.

[0018] As an optimal technical solution, the receptive field enhancement module including dilated convolution consists of two parts: a multi-branch structure using dilated convolution and an aggregated weighted layer. The multi-branch structure consists of four parallel dilated convolution branches with shared weights but inconsistent dilated convolution expansion rates, and each parallel dilated convolution branch introduces a residual connection. The aggregated weighted layer performs weighted summation on the information collected by different parallel dilated convolution branches.

[0019] As a preferred technical solution, the first-order statistical features and texture features of the chest DR image are extracted by using local binary patterns, Laplacian of Gaussian images and wavelet filter transforms on the lung field area.

[0020] As a preferred technical solution, the semantic posture features are constructed and evaluated based on the photographic position, including 7 semantic posture morphological features of the lung field, 6 semantic posture morphological features of the overlap between the scapula and the lung field, 4 semantic posture morphological features of the clavicle, and 2 semantic posture morphological features of the artifact;

[0021] Based on the constructed semantic posture morphological features, lung field morphological features, overlapping morphological features, clavicle morphological features and artifact morphological features are extracted from the lung field area, the overlapping area of ​​scapula and lung field, the clavicle area and the external metal artifact area respectively.

[0022] As a preferred technical solution, the feature selection is specifically:

[0023] Normalize the radiomics feature set of the training dataset;

[0024] The normalized radiomics feature set was preliminarily screened using the variance analysis method. The intra-group variance and inter-group variance of each feature in different chest DR images were compared and analyzed to evaluate the importance of the feature, and the features with an inter-group significance level of less than 0.05 were screened out.

[0025] The recursive feature elimination method with cross-validation was used to recursively remove features from the radiomics feature set after preliminary screening:

[0026] Determine a base model for recursive feature removal;

[0027] A cross-validation strategy was adopted to train the base model using the radiomics features of current chest DR images, and the prediction accuracy of the base model was evaluated;

[0028] Calculate the importance score of each feature in the radiomics features of the current chest DR image according to the trained base model, and remove the feature with the lowest importance score in the radiomics features of the current chest DR image from the removed radiomics feature set of the training data;

[0029] The radiomics feature training base model of the next chest DR image is obtained, and the importance score is calculated to remove the feature with the lowest importance score among the radiomics features of the chest DR image. The radiomics feature set is updated until the radiomics features are reduced to the set number, and the optimal feature subset of the training data set is obtained.

[0030] On the other hand, a chest DR image quality assessment system based on radiomics features is provided, comprising a data acquisition module, a multi-label semantic segmentation module, a feature extraction module, a feature selection module and a quality assessment module;

[0031] The data acquisition module is used to collect chest DR images with quality grade labels and region of interest mask labels, and construct a training data set; the quality grades include excellent, good, fair and poor;

[0032] The multi-label semantic segmentation module is used to use the trained multi-label semantic segmentation model based on full supervision to perform semantic segmentation on the chest DR images in the training data set to obtain the region of interest of the chest DR images; the region of interest includes the lung field area, the overlapping area of ​​the scapula and the lung field, the clavicle area and the external metal artifact area;

[0033] The feature extraction module is used to extract features from the region of interest to obtain radiomic features of the chest DR image, and to construct a radiomic feature set of the training data set, including: extracting features from the lung field region to obtain first-order statistical features and texture features of the chest DR image; capturing semantic posture features from the lung field region, the overlapping region of the scapula and the lung field, the clavicle region, and the external metal artifact region to obtain morphological features of the chest DR image; the radiomic feature set includes a first-order statistical feature set, a texture feature set, and a morphological feature set;

[0034] The feature selection module is used to perform feature selection on the radiomics feature set of the training data set using a variance analysis method and a recursive feature elimination method with cross-validation to obtain an optimal feature subset of the training data set;

[0035] The quality assessment module is used to construct an automatic chest DR image quality assessment model based on an integrated learning gradient boosting decision tree classifier, and use the optimal feature subset of the training data set for training to achieve quality level assessment of chest DR images.

[0036] In another aspect, an electronic device is provided, comprising:

[0037] at least one processor; and a memory communicatively coupled to the at least one processor;

[0038] The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the above-mentioned chest DR image quality assessment method based on radiomics features.

[0039] On the other hand, a computer-readable storage medium is provided, which stores a program, and when the program is executed by a processor, the above-mentioned chest DR image quality assessment method based on radiomics features is implemented.

[0040] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0041] 1. Enables more comprehensive image quality assessment

[0042] The existing two-step cascade chest DR image quality assessment method considers a relatively single angle (often only considering the photographic position as a quality control standard point), and fails to achieve a full assessment. Considering that radiomics features can describe information such as shape, texture, and intensity distribution in medical images, and have shown their superiority in quantifying experts' subjective visual perception and capturing subtle differences in images in applications of disease diagnosis and survival prediction, the present invention comprehensively considers multiple quality control standard points such as photographic position, image clarity, contrast, and artifact influence, which are more in line with the doctor's clinical evaluation process, and uses radiomics features to quantitatively describe the expert's subjective visual perception; in addition, the present invention also constructs semantic posture morphological features that are more consistent with the chest DR image quality assessment task based on the quality control standard point of photographic position to replace the original shape features and capture the differences between images of different levels.

[0043] 2. Improve the segmentation accuracy of the region of interest

[0044] Existing methods often use U-Net networks to segment regions of interest in medical images, but the encoder part of the U-Net network needs to use continuous convolution and pooling operations to gradually reduce the resolution of the feature space. This operation process makes the receptive field of the convolution relatively small, making it difficult to cover a larger range of contextual information. In addition, during the downsampling process, the reduction in spatial size easily leads to the loss of detail information and insufficient extraction of edge information, resulting in poor segmentation performance. The present invention constructs a semantic segmentation model with receptive field enhancement with dilated convolution based on the U-Net architecture, and introduces a receptive field enhancement module containing dilated convolution in the encoder part, so that when the segmentation model obtains feature information in the image through the encoder, it can help improve the model's processing ability for multi-scale information without increasing the computational complexity through dilated convolutions with different expansion rates, so as to effectively improve the segmentation accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0046] Figure 1 4 is an overall flow chart of a method for evaluating chest DR image quality based on radiomics features in an embodiment of the present invention.

[0047] Figure 2 (a) is the schematic diagram of the structure of RFEU-Net, and (b) is the schematic diagram of the structure of the RFEM module.

[0048] Figure 3 Schematic diagram of measuring semantic posture morphological features in an embodiment of the present invention.

[0049] Figure 4 Schematic diagram of the structure of a chest DR image quality assessment system based on radiomics features in an embodiment of the present invention.

[0050] Figure 5 Schematic diagram of the structure of a computer-readable storage medium in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0052] Reference to "embodiments" in this application means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described in this application may be combined with other embodiments.

[0053] In recent years, radiomics features have demonstrated their superiority in the diagnosis and prognosis prediction of diseases such as lung diseases and cardiovascular diseases by describing information such as shape, texture, and intensity distribution in medical images, quantifying the subjective visual perception of experts, and capturing subtle differences in images. Therefore, the present invention comprehensively considers multiple quality control standard points related to diagnostic quality and proposes a chest DR image quality assessment method based on radiomics features, which provides a new way to evaluate the diagnostic quality of chest DR images. Figure 1 As shown, this embodiment provides a method for evaluating chest DR image quality based on radiomics features, comprising the following steps:

[0054] S1. Collect chest DR images with quality grade labels and region of interest mask labels, and construct a training dataset.

[0055] The present invention focuses on the evaluation of the overall diagnostic quality of chest DR images. Therefore, the quality level of chest DR images divided in the literature (Jia Xiaoqian, Zhang Xiangli, Liu Zhe, et al. The application value of convolutional neural networks in assisting chest digital X-ray image quality control [J]. Journal of Xi'an Jiaotong University: Medical Edition, 2019, 40(5):5. DOI:10.7652 / jdyxb201905023.) is adopted, including excellent, good, general and poor.

[0056] S2. Use the trained fully supervised multi-label semantic segmentation model to perform semantic segmentation on the chest DR images in the training dataset to obtain the region of interest of the chest DR images; the region of interest includes the lung field area, the overlapping area between the scapula and the lung field, the clavicle area, and the external metal artifact area.

[0057] Since the adjacent areas with similar density are not distinguishable in X-ray images, and the metal artifact segmentation target is small and has various shapes, the segmentation effect of the traditional U-Net network on some images when segmenting the region of interest is inconsistent with the anatomical structure, and there is a phenomenon of missed segmentation for metal artifacts. Therefore, the multi-label semantic segmentation model based on full supervision adopts a semantic segmentation model with a receptive field enhancement with a hole convolution built on the U-Net architecture, named RFEU-Net, to enhance the receptive field during the convolution operation of the U-Net network encoder area, thereby improving the accuracy of the model semantic segmentation. RFEU-Net keeps the overall architecture of the encoder and decoder shaped like the English letter "U" unchanged, wherein the encoder part includes multiple downsampling convolution blocks, and the feature information of the target area is retained to the maximum extent by the method of step-by-step downsampling; because the encoder has limited ability to capture long-distance feature clues under such a contraction path and is prone to ignoring detail information, RFEU-Net adds a receptive field enhancement module (Receptive Field Enhancement Model, RFEM) containing a hole convolution after each downsampling convolution block. The decoder part includes multiple upsampling convolution blocks; since X-ray imaging maps three-dimensional human features into two-dimensional space, an X-ray may pass through multiple human tissues to reach the detector, that is, a single pixel may belong to multiple semantic categories. In order to be applicable to the chest DR image segmentation task of the present invention, the Sigmoid activation function is used at the end of the decoder to replace the original Softmax activation function to ensure that each pixel has independent output in the four categories of the overlapping area of ​​scapula and lung field, lung field area, clavicle area and external metal artifact area. The number of upsampling convolution blocks is the same as that of downsampling convolution blocks, which are set according to requirements. Figure 2 As shown in (a), this implementation provides a specific structure of RFEU-Net, in which the number of upsampling convolution blocks, downsampling convolution blocks and RFEM is four.

[0058] More specifically, the downsampling convolution block consists of two convolution blocks containing 3×3 convolution, ReLU activation function and batch normalization (Batch Norm), and a maximum pooling layer with a stride of 2. When the feature map enters the convolution block, it undergoes two convolution operations to extract feature information, and the number of channel dimensions doubles; after the maximum pooling downsampling operation, the feature map is reduced by half in size in the spatial dimension. Figure 2As shown in (b), the RFEM module is constructed with reference to the literature (Yu Z, Huang H, Chen W, et al. Yolo-facev2: Ascale and occlusion aware face detector [J]. Pattern Recognition, 2024: 110714.), and is mainly composed of a multi-branch structure using dilated convolution and an aggregated weighted layer. The multi-branch structure uses four parallel dilated convolution branches with weight sharing but inconsistent dilated convolution expansion rates to capture multi-scale information, and each parallel dilated convolution branch introduces a residual connection. The use of dilated convolutions with different expansion rates can make full use of the information in each sample and has the ability to capture long-range dependencies. The weight sharing between branches reduces the number of parameters and the risk of potential model overfitting. The introduction of residual connections prevents potential gradient explosion and gradient vanishing problems in the training process. The aggregated weighted layer is mainly responsible for weighted summation of the information collected by different branches to balance the contribution values ​​between different branches.

[0059] S3. Extract features from the region of interest to obtain the radiomics features of the chest DR image, and construct the radiomics feature set of the training data set, including: extract features from the lung field area to obtain the first-order statistical features and texture features of the chest DR image; capture semantic posture features from the lung field area, the overlapping area of ​​the scapula and lung field, the clavicle area, and the external metal artifact area to obtain the morphological features of the chest DR image. The constructed radiomics feature set includes a first-order statistical feature set, a texture feature set, and a morphological feature set.

[0060] Among them, the first-order statistical features are mainly used to measure the uniformity, symmetry and local intensity distribution changes of pixels; the texture features can reflect the spatial arrangement relationship between the grayscale of image pixels. From the perspective of diagnostic quality evaluation, doctors are more concerned about the density distribution, exposure conditions and clarity and sharpness of the texture structure of the lung field in the lung area. Therefore, the present invention firstly extracts only the first-order statistical features and texture features of the lung field area, specifically:

[0061] The lung field region is processed by local binary pattern (LBP), Gaussian Laplace (LoG) image and wavelet filter transform (Wavelet) to extract the first-order statistical features and texture features of the image before and after the transformation. In this embodiment, 651 first-order statistical features and texture features are finally obtained from the lung field region image.

[0062] Since morphological features are often used for geometric measurement of tumor morphology, information such as position and angle may also be considered when evaluating the quality of chest X-rays. Therefore, the present invention constructs 19 semantic posture morphological features that are more in line with the quality evaluation of chest DR images based on the quality control standard point of photographic position to replace morphological features, including 7 semantic posture morphological features of lung fields, 6 semantic posture morphological features of shoulder blade and lung field overlap, 4 semantic posture morphological features of clavicle, and 2 semantic posture morphological features of artifacts. Then, based on the constructed semantic posture morphological features, lung field morphological features, overlapping morphological features, clavicle morphological features, and artifact morphological features are extracted from the lung field area, the overlapping area of ​​scapula and lung field, the clavicle area, and the external metal artifact area, respectively.

[0063] Figure 3 The figure is a schematic diagram of the measurement of semantic pose morphological features, which defines examples of capturing key semantic pose morphological features from four regions of interest. The detailed semantic pose morphological feature measurement information is shown in Table 1 below:

[0064] Table 1 Semantic pose morphological feature measurement information

[0065]

[0066]

[0067] The lung field area contains 7 features, which are mainly used to measure the relationship between the lung field area and the edge of the camera and the size of the field of view; the overlapping area between the scapula and the lung field has 6 features, which use the area and the maximum intercept to measure the overlap between the scapula and the lung; the clavicle area uses 4 features to measure the flatness and symmetry of the left and right clavicles. clavi_l The distance between the intersection of the left clavicle and ribs and the upper edge of the clavicle area, D clavi_r It represents the distance between the intersection of the right clavicle and ribs and the upper edge of the clavicle area, D clavi_l_r It means D clavi_l and D clavi_r The difference between the two is used to reflect the flatness and symmetry of the clavicles on both sides; while the external metal artifact area is measured using only two features, one is the area of ​​the artifact, and the other is the area where the artifact is located. When the metal artifact is located inside the lung field and the larger the area covered, the greater the impact on the doctor's diagnosis.

[0068] S4. Use variance analysis method and recursive feature elimination method with cross-validation to perform feature selection on the radiomics feature set of the training data set to obtain the optimal feature subset of the training data set.

[0069] Since the extracted high-throughput imaging genomics feature set contains redundant and strongly correlated features, this often increases the difficulty of model training and the performance of model prediction. In order to obtain a high-quality feature subset to improve the performance of the final model, the present invention adopts a step-by-step strategy for feature selection: first, preliminary screening is performed using the analysis of variance (ANOVA) method, and then fine screening is performed using the recursive feature elimination method with cross-validation (RFECV). Specifically:

[0070] Before the preliminary screening, in order to eliminate the influence of different feature dimensions, the radiomics feature set of the training data set was normalized to its maximum and minimum values, so that the feature size was between 0 and 1.

[0071] Analysis of variance (ANOVA) is an effective filtering selection strategy. During the initial screening, the intra-group variance and inter-group variance of each feature in different chest DR images are compared and analyzed to evaluate the importance of the feature, and the features with inter-group significance level lower than 0.05 are screened out.

[0072] Recursive feature elimination with cross validation (RFECV) is a wrapper selection strategy that recursively removes the least important features and then uses cross validation to evaluate model performance to select the optimal feature subset. The specific implementation steps are as follows:

[0073] First, a model for feature selection is determined. Since the automatic chest DR image quality assessment model in this application is constructed using a gradient boosting decision tree classifier (GBDT) based on ensemble learning, GBDT is also used as the base model to evaluate the performance of feature subsets during feature selection.

[0074] Secondly, a cross-validation strategy was adopted to train the base model using the radiomics features of current chest DR images, and the prediction accuracy of the base model was evaluated.

[0075] Then, the importance score of each feature in the radiomics features of the current chest DR image is calculated according to the trained base model, and the feature with the lowest importance score in the radiomics features of the current chest DR image is removed from the removed radiomics feature set of the training data.

[0076] Finally, the radiomics features of the next chest DR image are obtained and the above steps are repeated to train the base model and calculate the importance score. The feature with the lowest importance score among the radiomics features of the chest DR image is removed, and the radiomics feature set is updated until the number of radiomics features is reduced to the set number, thus obtaining the optimal feature subset of the training data set.

[0077] In this embodiment, after feature selection, a feature set of 30 is obtained as the optimal feature subset to train the automatic chest DR image quality assessment model.

[0078] S5. An automatic chest DR image quality assessment model is constructed based on the gradient boosting decision tree classifier based on ensemble learning, and the optimal feature subset of the training data set is used for training to achieve quality level assessment of chest DR images.

[0079] After selecting the optimal feature subset, it is a common feature analysis method to use a machine learning model to extract the effective information contained in the feature subset and realize the mapping of feature vectors and prediction labels. In this application, a gradient boosting decision tree (GBDT) classifier based on ensemble learning is used to construct an automatic assessment model for chest DR image quality. The GBDT method uses a decision tree model as the base learner of the boosting strategy. In each iteration, GBDT adds a decision tree to try to fit the negative gradient of the residual after the previous iteration, so as to gradually reduce the global loss, and finally integrate all decision trees to form a powerful prediction model. By iteratively learning the residuals and gradually improving the model accuracy, GBDT can efficiently capture the complex relationship between features and ultimately achieve the assessment of the quality level of chest DR images.

[0080] In summary, the present application applies the radiomics features that quantitatively describe the tumor status in medical images to the task of chest DR image quality assessment, solving the problems of single quality control standard points, strong subjectivity, and poor segmentation accuracy considered in the current chest DR image quality assessment. On the one hand, the present invention hopes to consider multiple quality control standard points for quality assessment; traditional methods often only obtain deep learning features from the entire image or only consider a single quality control standard point, and when evaluating the overall diagnostic quality level of chest DR images, considering multiple quality control standard points such as photographic position, image clarity, contrast, and artifact influence is more in line with the doctor's clinical evaluation process. On the other hand, the method proposed in the present invention hopes to quantify the subjective visual perception of experts and capture subtle differences in images; use radiomics features to quantitatively describe the above-mentioned quality control standard points, and design semantic posture morphological features to replace shape features to more effectively represent the photographic position problem in quality control, thereby improving the evaluation results. On the other hand, the segmentation network proposed in the present invention can improve the segmentation performance of the segmentation model on X-ray images; by introducing a receptive field enhancement module to enhance the receptive field during convolution operations, the segmentation model's ability to capture contextual information is improved, thereby improving segmentation accuracy.

[0081] It should be noted that, for the sake of convenience, the aforementioned method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that the present invention is not limited to the described order of actions, because according to the present invention, certain steps can be performed in other orders or simultaneously.

[0082] Based on the same idea as the chest DR image quality assessment method based on radiomics features in the above-mentioned embodiment, the present invention also provides a chest DR image quality assessment system based on radiomics features, which can be used to execute the above-mentioned chest DR image quality assessment method based on radiomics features. For ease of explanation, the structural schematic diagram of the embodiment of the chest DR image quality assessment system based on radiomics features only shows the parts related to the embodiment of the present invention. Those skilled in the art can understand that the illustrated structure does not constitute a limitation on the device, and may include more or fewer components than shown in the diagram, or combine certain components, or arrange the components differently.

[0083] like Figure 4 As shown, another embodiment of the present invention provides a chest DR image quality assessment system based on radiomics features, including a data acquisition module, a multi-label semantic segmentation module, a feature extraction module, a feature selection module and a quality assessment module;

[0084] The data acquisition module is used to collect chest DR images with quality grade labels and region of interest mask labels and construct a training data set; the quality grades include excellent, good, fair, and poor;

[0085] The multi-label semantic segmentation module is used to use the trained fully supervised multi-label semantic segmentation model to perform semantic segmentation on the chest DR images in the training data set to obtain the regions of interest of the chest DR images; the regions of interest include the lung field area, the overlapping area between the scapula and the lung field, the clavicle area, and the extracorporeal metal artifact area;

[0086] The feature extraction module is used to extract features from the region of interest to obtain the radiomics features of the chest DR image and construct the radiomics feature set of the training data set, including: extracting features from the lung field area to obtain the first-order statistical features and texture features of the chest DR image; capturing semantic posture features from the lung field area, the overlapping area of ​​the scapula and lung field, the clavicle area, and the external metal artifact area to obtain the morphological features of the chest DR image; the radiomics feature set includes a first-order statistical feature set, a texture feature set, and a morphological feature set;

[0087] The feature selection module is used to select the radiomics feature set of the training data set using the variance analysis method and the recursive feature elimination method with cross-validation to obtain the optimal feature subset of the training data set;

[0088] The quality assessment module is used to build an automatic chest DR image quality assessment model based on the gradient boosting decision tree classifier of ensemble learning, and use the optimal feature subset of the training data set for training to achieve quality level assessment of chest DR images.

[0089] It should be noted that the chest DR image quality assessment system based on imaging genomics features of the present invention corresponds one-to-one to the chest DR image quality assessment method based on imaging genomics features of the present invention. The technical features and beneficial effects described in the embodiment of the chest DR image quality assessment method based on imaging genomics features are applicable to the embodiment of the chest DR image quality assessment system based on imaging genomics features. For specific contents, please refer to the description in the embodiment of the method of the present invention, which will not be repeated here. This is hereby declared.

[0090] In addition, in the implementation of the chest DR image quality assessment system based on radiomics features in the above-mentioned embodiment, the logical division of each program module is only an example. In actual applications, the above-mentioned functions can be assigned to different program modules as needed, for example, for the convenience of corresponding hardware configuration requirements or software implementation. That is, the internal structure of the chest DR image quality assessment system based on radiomics features is divided into different program modules to complete all or part of the functions described above.

[0091] See also Figure 5 In one embodiment, an electronic device for implementing a method for evaluating chest DR image quality based on radiomics features is provided. The electronic device may include a first processor, a first memory, and a bus, and may also include a computer program stored in the first memory and executable on the first processor, such as a chest DR image quality evaluation program.

[0092] Wherein, the first memory includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the first memory can be an internal storage unit of an electronic device, such as a mobile hard disk of the electronic device. In other embodiments, the first memory can also be an external storage device of an electronic device, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device. Further, the first memory can also include both an internal storage unit of the electronic device and an external storage device. The first memory can not only be used to store application software and various types of data installed in the electronic device, such as the code of the chest DR image quality assessment program, but also can be used to temporarily store data that has been output or is to be output.

[0093] In some embodiments, the first processor may be composed of an integrated circuit, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The first processor is the control core (ControlUnit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes various functions of the electronic device and processes data by running or executing programs or modules (such as chest DR image quality assessment programs, etc.) stored in the first memory, and calling data stored in the first memory.

[0094] Figure 5 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 5 The structure shown does not constitute a limitation on the electronic device, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0095] The chest DR image quality assessment program stored in the first memory of the electronic device is a combination of multiple instructions, and when executed in the first processor, can achieve:

[0096] Collect chest DR images with quality grade labels and region of interest mask labels and construct a training dataset; the quality grades include excellent, good, fair, and poor;

[0097] Use the trained fully supervised multi-label semantic segmentation model to perform semantic segmentation on the chest DR images in the training data set to obtain the regions of interest of the chest DR images; the regions of interest include the lung field area, the overlapping area between the scapula and the lung field, the clavicle area, and the extracorporeal metal artifact area;

[0098] The radiomics features of the chest DR images were obtained by extracting features from the regions of interest, and the radiomics feature set of the training data set was constructed, including: extracting features from the lung field region to obtain the first-order statistical features and texture features of the chest DR images; capturing semantic posture features from the lung field region, the overlapping region of the scapula and lung field, the clavicle region, and the extracorporeal metal artifact region to obtain the morphological features of the chest DR images; the radiomics feature set includes a first-order statistical feature set, a texture feature set, and a morphological feature set;

[0099] The feature selection of the radiomics feature set of the training dataset was performed using the variance analysis method and the recursive feature elimination method with cross-validation to obtain the optimal feature subset of the training dataset;

[0100] An automatic chest DR image quality assessment model was constructed based on the gradient boosting decision tree classifier based on ensemble learning, and the optimal feature subset of the training data set was used for training to achieve quality grade assessment of chest DR images.

[0101] Furthermore, if the module / unit integrated in the electronic device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile computer-readable storage medium. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0102] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0103] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0104] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A method for evaluating chest DR image quality based on radiomics features, characterized in that: The steps include: Collect chest DR images with quality grade labels and region of interest mask labels, and construct a training data set; the quality grades include excellent, good, fair, and poor; Using the trained fully supervised multi-label semantic segmentation model to perform semantic segmentation on the chest DR images in the training data set to obtain the region of interest of the chest DR images; the region of interest includes the lung field area, the overlapping area of ​​the scapula and the lung field, the clavicle area, and the external metal artifact area; The imaging omics features of the chest DR image are obtained by extracting features of the region of interest, and the imaging omics feature set of the training data set is constructed, including: extracting features of the lung field area to obtain the first-order statistical features and texture features of the chest DR image; capturing semantic posture features from the lung field area, the overlapping area of ​​the scapula and the lung field, the clavicle area, and the external metal artifact area to obtain the morphological features of the chest DR image; the imaging omics feature set includes a first-order statistical feature set, a texture feature set, and a morphological feature set; The feature selection of the radiomics feature set of the training dataset was performed using the variance analysis method and the recursive feature elimination method with cross-validation to obtain the optimal feature subset of the training dataset; An automatic chest DR image quality assessment model was constructed based on the gradient boosting decision tree classifier based on ensemble learning, and the optimal feature subset of the training dataset was used for training to achieve quality grade assessment of chest DR images.

2. The method for evaluating chest DR image quality based on radiomics features according to claim 1, characterized in that: The fully supervised multi-label semantic segmentation model is a semantic segmentation model with receptive field enhancement with dilated convolution built on the basis of U-Net architecture, including an encoder and a decoder; The encoder comprises a plurality of downsampling convolution blocks, each of which is followed by a receptive field enhancement module comprising a dilated convolution; The decoder comprises a plurality of upsampling convolution blocks; the decoder is followed by a Sigmoid activation function; The number of the downsampling convolution blocks is the same as the number of the upsampling convolution blocks.

3. The chest DR image quality assessment method based on radiomics features according to claim 2 is characterized in that: The downsampling convolution block consists of two convolution blocks containing 3×3 convolution, ReLU activation function and batch normalization, and a maximum pooling layer with a stride of 2.

4. The chest DR image quality assessment method based on radiomics features according to claim 2, characterized in that: The receptive field enhancement module including dilated convolution consists of two parts: a multi-branch structure using dilated convolution and an aggregated weighted layer. The multi-branch structure consists of four parallel dilated convolution branches with shared weights but inconsistent dilated convolution expansion rates, and each parallel dilated convolution branch introduces a residual connection. The aggregated weighted layer performs weighted summation on the information collected by different parallel dilated convolution branches.

5. The method for evaluating chest DR image quality based on radiomics features according to claim 1, characterized in that: The first-order statistical features and texture features of the chest DR image are extracted by using local binary patterns, Laplacian of Gaussian images and wavelet filtering transformation on the lung field area.

6. The method for evaluating chest DR image quality based on radiomics features according to claim 1, characterized in that: The semantic posture features are constructed and evaluated based on the photographic position, including 7 semantic posture morphological features of the lung field, 6 semantic posture morphological features of the overlap between the scapula and the lung field, 4 semantic posture morphological features of the clavicle, and 2 semantic posture morphological features of the artifact; Based on the constructed semantic posture morphological features, lung field morphological features, overlapping morphological features, clavicle morphological features and artifact morphological features are extracted from the lung field area, the overlapping area of ​​scapula and lung field, the clavicle area and the external metal artifact area respectively.

7. The method for evaluating chest DR image quality based on radiomics features according to claim 1, characterized in that: The feature selection is specifically as follows: Normalize the radiomics feature set of the training dataset; The normalized radiomics feature set was preliminarily screened using the variance analysis method. The intra-group variance and inter-group variance of each feature in different chest DR images were compared and analyzed to evaluate the importance of the feature, and the features with an inter-group significance level of less than 0.05 were screened out. The recursive feature elimination method with cross-validation was used to recursively remove features from the radiomics feature set after preliminary screening: Determine a base model for recursive feature removal; A cross-validation strategy was adopted to train the base model using the radiomics features of current chest DR images, and the prediction accuracy of the base model was evaluated; Calculate the importance score of each feature in the radiomics features of the current chest DR image according to the trained base model, and remove the feature with the lowest importance score in the radiomics features of the current chest DR image from the removed radiomics feature set of the training data; The radiomics feature training base model of the next chest DR image is obtained, and the importance score is calculated to remove the feature with the lowest importance score among the radiomics features of the chest DR image. The radiomics feature set is updated until the radiomics features are reduced to the set number, and the optimal feature subset of the training data set is obtained.

8. A chest DR image quality assessment system based on radiomics features, characterized in that: It includes data acquisition module, multi-label semantic segmentation module, feature extraction module, feature selection module and quality assessment module; The data acquisition module is used to collect chest DR images with quality grade labels and region of interest mask labels, and construct a training data set; the quality grades include excellent, good, fair and poor; The multi-label semantic segmentation module is used to use the trained multi-label semantic segmentation model based on full supervision to perform semantic segmentation on the chest DR images in the training data set to obtain the region of interest of the chest DR images; the region of interest includes the lung field area, the overlapping area of ​​the scapula and the lung field, the clavicle area and the external metal artifact area; The feature extraction module is used to extract features from the region of interest to obtain radiomic features of the chest DR image, and to construct a radiomic feature set of the training data set, including: extracting features from the lung field region to obtain first-order statistical features and texture features of the chest DR image; capturing semantic posture features from the lung field region, the overlapping region of the scapula and the lung field, the clavicle region, and the external metal artifact region to obtain morphological features of the chest DR image; the radiomic feature set includes a first-order statistical feature set, a texture feature set, and a morphological feature set; The feature selection module is used to perform feature selection on the radiomics feature set of the training data set using a variance analysis method and a recursive feature elimination method with cross-validation to obtain an optimal feature subset of the training data set; The quality assessment module is used to construct an automatic chest DR image quality assessment model based on an integrated learning gradient boosting decision tree classifier, and use the optimal feature subset of the training data set for training to achieve quality level assessment of chest DR images.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively coupled to the at least one processor; Wherein, the memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the chest DR image quality assessment method based on radiomics features as described in any one of claims 1-7.

10. A computer-readable storage medium storing a program, characterized in that: When the program is executed by a processor, the chest DR image quality assessment method based on radiomics features described in any one of claims 1 to 7 is implemented.

Images