Method and apparatus for evaluating activity imaging of common abdominal solid tumors in children and media

Abstract

The application provides a kind of common solid tumor activity image evaluation method and medium in children abdomen, utilize 3D convolution neural network model, capture the change information of lesion on the image of plain scan and enhancement after intravenous period, carry out intelligentized silhouette generation lesion in the intensification area and intensification degree after chemotherapy, finally comprehensively all information is intelligently diagnosed, judge tumor activity, form abdominal solid tumor artificial intelligence auxiliary activity evaluation scheme after chemotherapy;The application utilizes the plain scan and enhanced CT image, realizes the automatic intensification degree quantitative evaluation of children hepatoblastoma, nephroblastoma and neuroblastoma by tumor automatic segmentation and registration subtraction image calculation, so as to extract lesion characteristics from multiple angles, fully excavate image information, which can significantly improve the accuracy and sensitivity of activity discrimination of children abdominal solid tumor after chemotherapy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and in particular to a method and device for evaluating activity images of common abdominal solid tumors in children and a computer readable storage medium. BACKGROUND

[0002] Childhood abdominal solid tumors account for about 40% of all childhood malignant tumors, and are most commonly found in the abdomen, among which neuroblastoma, nephroblastoma and hepatoblastoma are the most common, seriously endangering the health of children in China. Surgical resection combined with neoadjuvant chemotherapy and postoperative radiotherapy and chemotherapy has become the standard treatment for current common abdominal solid tumors in children, and accurate imaging evaluation is an important method for judging the efficacy of neoadjuvant chemotherapy, determining the timing of surgery and monitoring postoperative recurrence.

[0003] Long-term multidisciplinary comprehensive diagnosis and treatment (MDT) and clinical drug experience show that after neoadjuvant chemotherapy, tumor reduction is often asymmetric, and some only show a decrease in the range of intratumoral enhancement or a decrease in the degree of enhancement in the early stage of chemotherapy. In addition, common abdominal solid tumors in children are usually large in size, irregular in shape, easy to push, wrap and invade surrounding organs and large blood vessels, which poses a challenge to accurate imaging evaluation.

[0004] Currently, the double-diameter method or the three-diameter method is still commonly used to evaluate tumor size in clinical practice, which has poor accuracy. The three-dimensional quantitative technique of drawing and measuring tumor volume layer by layer can obtain the actual size of the tumor and avoid errors caused by asymmetric reduction. However, manual drawing of the tumor is time-consuming and laborious, and for large tumors, the manual measurement time can be more than 1 hour. Moreover, the boundaries of some tumors are irregular, and the contrast with the surrounding structures is low, so pure manual drawing depends on the judgment of the clinician's experience, and the results vary greatly between measurers and have poor repeatability, making it difficult to apply. In addition, the evaluation of tumor function and activity is also an important aspect of tumor imaging evaluation. Currently, the most common method in clinical practice is CT / MRI enhancement examination or MRI diffusion-weighted imaging. CT / MRI enhancement examination can indirectly reflect the degree of tumor activity by observing the current blood supply of the tumor, and can reflect the efficacy of the tumor earlier than the change in tumor volume. However, for large anatomical span of children's solid tumors, the assessment of the range and degree of enhancement is more experience-dependent. When the volume change is not obvious, young doctors and even primary doctors may draw opposite conclusions from experienced experts, which seriously affects the choice of subsequent treatment plan and prognosis of the child. Therefore, there is an urgent need for an intelligent evaluation tool that can balance the accuracy of tumor volume measurement and functional analysis, has high accuracy and strong timeliness, to achieve accurate imaging evaluation of common abdominal solid tumors in children. SUMMARY

[0005] The present invention aims to provide a method, apparatus and computer-readable storage medium for evaluating the activity of common abdominal solid tumors in children by images that overcomes or at least partially solves the above problems.

[0006] To achieve the above objectives, the technical solution of the present invention is specifically implemented as follows:

[0007] One aspect of the present invention provides a method for assessing the activity of common abdominal solid tumors in children using imaging techniques, comprising:

[0008] A tumor segmentation model is determined, wherein the tumor segmentation model is the UNETR model;

[0009] The input image is resampled according to a preset standard interval;

[0010] The resampled data is block-sampled according to the preset block size;

[0011] The training samples are randomly rotated and augmented using time-of-test (TOT) augmentation techniques.

[0012] The tumor segmentation model is trained using the training set with the segmentation annotation results of the tumor region annotated by the physician as the target, and the trained tumor segmentation model is obtained.

[0013] The trained tumor segmentation model was validated and its parameters were tuned using a validation set.

[0014] The trained tumor segmentation model was evaluated using a test set.

[0015] The trained tumor segmentation model is used to segment the image to be segmented, and the tumor segmentation result is obtained and visualized.

[0016] Acquire CT images of children and remove artifacts from the CT images of children;

[0017] Artifact-removed plain CT images of children were registered with arterial and venous phase images.

[0018] The tumor region in the tumor segmentation result is directly mapped onto the plain scan and arterial phase images after registration and transformation. The subtraction images of the arterial phase, venous phase and plain scan are calculated respectively to obtain the absolute enhancement value images of the tumor's artery and vein. The absolute enhancement value images are divided by the plain scan images to obtain the normalized enhancement images.

[0019] The degree of enhancement is evaluated using the absolute enhancement value image and the normalized enhancement image.

[0020] Optionally, the step of training the tumor segmentation model using the training set with the segmentation annotation results of the tumor region annotated by the physician as the target, to obtain the trained model includes:

[0021] Using the backpropagation algorithm and stochastic gradient descent method, the weights of each layer are iteratively updated based on the magnitude of the loss function between the predicted segmentation result and the labeled result of the forward propagation. The training of the model is stopped when the loss function between the predicted segmentation result and the labeled result of the model tends to converge, and the trained model is obtained.

[0022] Optionally, the step of validating and tuning the trained tumor segmentation model using a validation set includes:

[0023] By monitoring the Dice and IoU coefficients, the optimal model parameters and hyperparameter configurations can be selected.

[0024] Optionally, the preset standard spacing is 1.0×1.0×1.0 mm; the preset block size is 128×128×128 voxels.

[0025] Optionally, acquiring pediatric CT images and removing artifacts from the pediatric CT images includes:

[0026] The pediatric CT images were decomposed into different frequency sub-bands using wavelet transform;

[0027] In the decomposed subbands, textures exhibiting high-frequency noise or abnormalities are detected, and the detected artifacts are thresholded.

[0028] The sub-bands after thresholding are reconstructed back into the image space using inverse wavelet transform to obtain the CT image after artifact removal.

[0029] Optionally, the registration of the artifact-removed plain CT images of children with arterial and venous phase images includes:

[0030] The tumor segmentation results and abdominal high-density bone were used to perform auxiliary feature point registration, wherein the registration evaluation criterion was based on mutual information (MI).

[0031] Optionally, the auxiliary feature point registration using the tumor segmentation results and abdominal high-density bone includes:

[0032] Standardization of pediatric CT plain scan images after artifact removal was performed using Z-scores before registration.

[0033] High-density bone was segmented using the Otsu adaptive threshold segmentation method, and voids were filled using a closing operation. Edge points and corner points were extracted from the obtained tumor and high-density bone regions, respectively. The edge points were obtained by edge detection using the Canny edge detection algorithm and by selecting points at equal intervals at the edges. The corner points were obtained by the Harris corner detection algorithm.

[0034] The nearest neighbor matching algorithm was used to find the corresponding feature point pairs in the artifact-removed pediatric CT plain scan image, the arterial phase image, and the venous phase image;

[0035] Registration is performed using affine transformation, and the transformation parameters are estimated by minimizing the mean square error of the matching point pairs.

[0036] A similarity metric is defined using mutual information and feature point matching error;

[0037] Maximize the similarity measure using gradient descent;

[0038] The iteration continues until the similarity metric no longer increases significantly and the feature point matching error is less than a preset value, at which point the iteration stops and the registration is complete.

[0039] Optionally, the enhancement degree evaluation using the absolute enhancement value image and the normalized enhancement image includes:

[0040] The enhancement distribution information of the patient's tumor region is calculated based on the normalized enhanced image;

[0041] Calculate the percentage of tumor volume in which the enhancement within the tumor exceeds a set threshold.

[0042] Another aspect of the present invention provides an active imaging assessment device for common abdominal solid tumors in children, comprising: a processor and a memory;

[0043] The memory is used to store computer programs;

[0044] The processor is used to execute the above-described method for assessing the activity of common abdominal solid tumors in children by invoking the computer program.

[0045] Another aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the above-described method for evaluating the activity of common abdominal solid tumors in children using imaging.

[0046] Therefore, the method, device, and computer-readable storage medium for assessing the activity of common abdominal solid tumors in children provided by this invention utilize a 3D convolutional neural network model to capture changes in lesions on plain and enhanced venous phase images. Intelligent silhouette generation is then used to determine the enhancement area and degree of enhancement of the lesion after chemotherapy. Finally, all information is integrated for intelligent diagnosis and tumor activity assessment, forming an AI-assisted activity assessment scheme for abdominal solid tumors after chemotherapy. This invention utilizes plain and enhanced CT images, and through automatic tumor segmentation and registration subtraction image calculation, achieves automatic quantitative assessment of the enhancement degree of hepatoblastoma, nephroblastoma, and neuroblastoma in children. This allows for the extraction of lesion features from multiple angles, fully leveraging image information, and significantly improving the accuracy and sensitivity of activity assessment for abdominal solid tumors after chemotherapy in children. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A flowchart of a method for evaluating the activity of common abdominal solid tumors in children, provided in an embodiment of the present invention;

[0049] Figure 2 This is a schematic diagram of the UNETR network structure provided in an embodiment of the present invention. Detailed Implementation

[0050] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0051] To address the problems of existing technologies, this invention proposes a method, device, and computer-readable storage medium for evaluating the activity of common abdominal solid tumors in children based on deep learning. This invention utilizes plain and enhanced CT images, and through automatic tumor segmentation and registration subtraction image calculation, achieves automatic quantitative assessment of the enhancement degree of hepatoblastoma, nephroblastoma, and neuroblastoma in children.

[0052] Figure 1 A flowchart illustrating the method for assessing the activity of common abdominal solid tumors in children according to an embodiment of the present invention is shown. See also... Figure 1The method for assessing the activity of common abdominal solid tumors in children using imaging techniques provided in this invention includes:

[0053] I. Construct a gold standard database and a tumor segmentation model, including:

[0054] Specifically, the tumor segmentation model of this invention is designed as a Transformer-based UNETR network. Deep learning technology has made significant progress in the field of medical image segmentation. UNETR is a Transformer-based segmentation model that introduces a self-attention mechanism into image segmentation tasks to achieve better performance and generalization ability. This invention trains a general-purpose pediatric solid tumor segmentation model for different cancers (NB / WT / HB) on a dataset with tumor-annotated mask information to achieve tumor segmentation.

[0055] S1, determine the tumor segmentation model, wherein the tumor segmentation model is the UNETR model.

[0056] Specifically, this invention selects the UNETR model, such as Figure 2 As shown, this is an advanced model that combines U-Net and Transformer. It not only retains the encoder-decoder structure of U-Net to capture the contextual information of the image, but also enhances the ability to capture global dependencies through the self-attention mechanism of Transformer, thus making it more outstanding in preserving detailed information.

[0057] S2, resample the input image according to the preset standard interval;

[0058] S3, perform block sampling on the resampled data according to the preset block size.

[0059] As an optional embodiment of the present invention, the preset standard spacing is 1.0×1.0×1.0 mm; the preset block size is 128×128×128 voxels.

[0060] Specifically, this step only involves image resampling and voxel unification:

[0061] In practice, resampling the input image to standardize the voxel spacing is a crucial step in ensuring consistency across different images. This is especially true in the UNETR model, which employs a self-attention mechanism; a uniform voxel spacing is essential for maintaining consistent model performance across various image types. This invention selects the median voxel spacing in the dataset, 1.0 × 1.0 × 1.0 mm, as the standard spacing for resampling.

[0062] After resampling, block sampling was further implemented to obtain a more uniform data resolution. Because the UNETR model incorporates a self-attention mechanism, it is more sensitive to the resolution of the input data and the size of the blocks. Therefore, this invention designs the block size and sampling strategy to maximize model performance. After optimization, this invention determines a uniform block size of 128×128×128 voxels, a size that ensures sufficient local feature capture while also meeting the model's need for global contextual information.

[0063] S4 utilizes test-time augmentation techniques to randomly rotate and augment the training samples.

[0064] Specifically, this step involves test time enhancement and model generalization:

[0065] In its specific implementation, this invention utilizes test-time augmentation techniques to randomly rotate and expand the training samples, thereby enhancing the model's generalization ability to different cancers. Since the UNETR model has better global feature capture capabilities, its robustness can be further improved through data augmentation strategies such as color transformation and noise addition.

[0066] S5. The tumor segmentation model is trained using the training set with the segmentation annotation results of the tumor region annotated by the physician as the target, and the trained tumor segmentation model is obtained.

[0067] As an optional implementation of this invention, the tumor segmentation model is trained using a training set with the segmentation and annotation results of the tumor region annotated by the physician as the target. The trained model is obtained by: using the backpropagation algorithm and the stochastic gradient descent method, and performing backpropagation iterative updates of the weights of each layer according to the magnitude of the loss function between the predicted segmentation result and the annotation result of the forward propagation, until the loss function between the predicted segmentation result and the annotation result of the model tends to converge, at which point the training of the model is stopped, and the trained model is obtained.

[0068] Specifically, this invention uses a training set to train the UNETR model with the segmentation and annotation results of the tumor region labeled by physicians as the target. It adopts the backpropagation algorithm and the stochastic gradient descent method. Based on the magnitude of the loss function between the predicted segmentation result and the annotation result of the forward propagation, the weights of each layer are updated iteratively during backpropagation until the loss function between the predicted segmentation result and the annotation result of the model tends to converge. At this point, the training of the model is stopped, and the trained model is obtained.

[0069] S6. Use the validation set to validate and fine-tune the trained tumor segmentation model.

[0070] As an optional implementation of this invention, using a validation set to validate and fine-tune the trained tumor segmentation model includes: selecting the optimal model parameters and hyperparameter configurations by monitoring the Dice and IoU coefficients.

[0071] Specifically, this invention performs model validation and parameter tuning on a validation set. By monitoring the Dice and IoU coefficients, the optimal model parameters and hyperparameter configurations are selected. The loss function of this invention is a joint of the Dice loss and the cross-entropy loss.

[0072] Dice loss is shown below:

[0073]

[0074] Where, p i G represents the predicted value. i This represents the true value, and the summation is performed on all elements.

[0075] The cross-entropy loss is shown below:

[0076] L cE =-∑ i g i log(p i )

[0077] Among them, g i It is the one-hot encoding of the real label, p i It is the predicted probability distribution.

[0078] The total loss function is shown below:

[0079] L loss =L Dice +λL CE

[0080] Here, λ is a balancing coefficient used to adjust the weights between the two losses.

[0081] S7 uses the test set to evaluate the trained tumor segmentation model.

[0082] Specifically, the trained UNETR segmentation model was evaluated on a test set. The Dice coefficient is a key metric for measuring the accuracy of tumor segmentation tasks. It assesses the precision and consistency of segmentation results by calculating the overlap between the predicted segmentation and the actual segmentation annotated by experienced radiologists. The Dice coefficient score ranges from 0 to 1; the closer the score is to 1, the higher the overlap between the predicted segmentation and the standard annotation, and the stronger the similarity. Generally, a Dice coefficient greater than 0.5 indicates a high degree of overlap in the segmentation results, thus reflecting good segmentation performance.

[0083]

[0084] Where A represents the binarized segmentation result predicted by the model, B represents the binarized segmentation result of the gold standard, and |·| represents the number of pixels.

[0085] II. Automatically extract tumor regions from images, including:

[0086] S8 uses the trained tumor segmentation model to segment the image to be segmented, obtains the tumor segmentation results, and visualizes the tumor segmentation results.

[0087] III. Register plain scan and arterial phase images to the venous phase, including:

[0088] S9: Acquire CT images of children and remove artifacts from them.

[0089] As an optional embodiment of the present invention, acquiring a child's CT image and removing artifacts from the child's CT image includes: using wavelet transform to decompose the child's CT image into different frequency sub-bands; detecting high-frequency noise or abnormal textures in the decomposed sub-bands and performing thresholding on the detected artifacts; reconstructing the thresholded sub-bands back into the image space through inverse wavelet transform to obtain the CT image after artifact removal.

[0090] Specifically, in pediatric CT scans, due to the smaller size and shape of children's organs compared to adults, and the potential movement of children during the scan, artifacts can be generated in the resulting CT images. Traditional artifact removal methods address the reconstruction process from the original projection data to the CT image. This invention only addresses the reconstructed CT image, therefore proposing the following artifact removal method based on pediatric CT images.

[0091] 1) Image decomposition

[0092] The image is decomposed into different frequency sub-bands using wavelet transform:

[0093] {c A ,c D1 ,c D2 ,…,c DB}=W(I)

[0094] Where I is a CT image containing artifacts. It's wavelet transform, c A These are approximation coefficients, c D1 ,c D2 ,…,c DN It is the detail factor.

[0095] 2) Artifact Detection and Thresholding

[0096] In the decomposed subbands, high-frequency noise or anomalous textures are detected, and the detected artifacts are thresholded to reduce their intensity or remove them completely. Simultaneously, to preserve image detail while removing high-frequency artifacts, an adaptive threshold is used for local adjustment. The threshold is automatically adjusted based on the local statistical characteristics (standard deviation) of the image, thus preserving useful details while removing artifacts.

[0097] τ(i,j)=k·σ(i,j)

[0098] c′ D (i,j)=sign(c D (i,j))·max(|c D (i,j)|-τ(i,j),0)

[0099] Where c D (i,j) are the wavelet detail coefficients at position (i,j), σ(i,j) is the local standard deviation at that position, k is a constant, τ(i,j) is the calculated adaptive threshold for that point, and sign(c D (i,j) is c D The symbol for (i,j).

[0100] 3) Image reconstruction

[0101] The sub-bands after adaptive thresholding are reconstructed back into the image space using inverse wavelet transform:

[0102]

[0103] Where I′ is the CT image after artifact removal.

[0104] S10 is used to register the pediatric CT plain scan images, arterial phase images, and venous phase images after artifact removal.

[0105] As an optional embodiment of the present invention, the registration of the pediatric CT plain scan image and arterial phase image and venous phase image after artifact removal includes: using tumor segmentation results and abdominal high-density bone for auxiliary feature point registration, wherein the registration evaluation criterion adopts mutual information (MI). The auxiliary feature point registration using tumor segmentation results and high-density abdominal bone includes: standardizing the pediatric CT plain scan image after artifact removal using Z-scores before registration; segmenting the high-density bone using the Otsu adaptive threshold segmentation method and filling the voids using a closing operation; extracting edge points and corner points from the obtained tumor and high-density bone regions, where edge points are detected using the Canny edge detection algorithm and points are selected at equal intervals along the edges, and corner points are obtained using the Harris corner detection algorithm; finding corresponding feature point pairs in the pediatric CT plain scan image, arterial phase image, and venous phase image after artifact removal using the nearest neighbor matching algorithm; performing registration using affine transformation, estimating the transformation parameters by minimizing the mean square error of the matching point pairs; defining a similarity metric using mutual information and feature point matching error; maximizing the similarity metric using gradient descent; and iterating until the similarity metric no longer increases significantly and the feature point matching error is less than a preset value, at which point the iteration stops, and the registration is complete.

[0106] Specifically, the data used in this invention are plain and enhanced CT images of the abdomen of children. Due to the different acquisition times of the images, there are mismatches between the sequences. In order to conduct subsequent enhancement assessment, it is necessary to first register the plain scan and the arterial and venous phase images. Clinically, radiologists generally believe that the tumor boundary is clearer in the venous phase image. Therefore, the registration process uses the venous phase image as the reference image and registers the plain scan and arterial phase images to the venous phase image.

[0107] Considering that abdominal solid tumors in children often present with space-occupying lesions, organ-assisted registration is difficult. Therefore, this invention employs a combination of tumor segmentation results and high-density abdominal bone for auxiliary feature point registration. Furthermore, considering the differences in grayscale distribution between plain, arterial, and venous phase images due to contrast agents, the registration evaluation criterion is based on mutual information (MI). MI measures the similarity between two images without requiring identical grayscale distributions, making it more suitable for this invention. The specific registration method is as follows:

[0108] 1) Image standardization

[0109] Image standardization before registration using Z-scores:

[0110]

[0111] Where I′ is the standardized image, and min(I) and max(I) are the minimum and maximum values ​​in the image, respectively.

[0112] 2) Feature point extraction

[0113] Tumor regions are obtained using an automatic segmentation model. High-density bone is automatically segmented from the image using the Otsu adaptive thresholding method.

[0114] Bone=(I>τ B )

[0115] Where τ B An adaptive threshold for the image.

[0116] The bone region obtained by threshold segmentation usually produces discrete small points due to low-density areas within the bone. The voids are automatically filled by closing operations.

[0117] Bone c =C(Bone)

[0118] C represents the closing operation.

[0119] Edge points and corner points (bone ridge points) were extracted from the obtained tumor and high-density bone regions respectively. Edge detection was performed using the Canny edge detection algorithm, and edge points were obtained by selecting points at equal intervals at the edges. Corner points were obtained using the Harris corner detection algorithm.

[0120] E Tumor =Canny(Tumor)

[0121] E Bone =Canny(Bone)

[0122] C Tumor =Harris (Tumor)

[0123] C Bone =Harris (Bone)

[0124] Where E Tumor E Bone C Tumor c Bone These are the obtained edge points and corner points of the tumor and bone, respectively.

[0125] 3) Feature point matching

[0126] The nearest neighbor matching algorithm was used to find corresponding feature point pairs in the plain scan, arterial phase image, and venous phase image:

[0127]

[0128] Where p is a feature point in the plain scan or arterial phase image, q is the corresponding feature point in the venous phase image, d(p,q) is the Euclidean distance between points p and q, and Q is the feature point set of the venous phase image.

[0129] 4) Transformation model estimation

[0130] Since the overall shapes of the phases are similar across different periods, affine transformation is used for registration. The transformation parameters are estimated by minimizing the mean square error of the matching point pairs.

[0131] T(x; θ) = Ax + t

[0132] Where T is the transformation function, x is the original coordinate, θ is the transformation parameter, A is a 2×2 matrix, and t is the translation vector.

[0133] 5) Similarity measurement

[0134] A similarity metric is defined using mutual information and feature point matching error:

[0135] S(θ)=αMI(T(I1;θ),I2)-βD(M,T(P1;θ))

[0136]

[0137] Where MI is mutual information, D is feature point matching error, and α and β are the weighting coefficients of the two, respectively.

[0138] 6) Optimize registration

[0139] Maximize the similarity measure using gradient descent:

[0140]

[0141] in λ is the gradient of the similarity measure with respect to the transformation parameter θ, and λ is the learning rate.

[0142] 7) Iteration

[0143] Repeat steps 3)-6) until the similarity metric no longer increases significantly and the feature point matching error is less than the preset value, at which point the iteration stops and the registration is complete.

[0144] IV. Calculation of normalized tumor enhancement images based on subtraction methods, including:

[0145] S11, the tumor region in the tumor segmentation result is directly mapped onto the plain scan and arterial phase images after registration transformation. The subtraction images of the arterial phase, venous phase and plain scan are calculated respectively to obtain the absolute enhancement value images of the tumor's artery and vein. The absolute enhancement value image is divided by the plain scan image to obtain the normalized enhancement image.

[0146] Specifically, the tumor region obtained through the tumor segmentation model described above can be directly mapped onto the plain and arterial phase images after registration and transformation. Subtraction images of the arterial, venous, and plain phases are calculated separately to obtain the absolute enhancement value images of the tumor's arteries and veins. To reduce the influence of individual differences among different patients, the obtained absolute enhancement value images are divided by the plain phase images to obtain normalized enhancement images.

[0147] I sub =I post -I pre

[0148]

[0149] Among them I post For arterial or venous phase images, I pre For plain scan images, I sub and I′ sub These are the calculated absolute enhancement value image and the normalized enhancement image, respectively.

[0150] V. Quantifying tumor region enhancement information, including:

[0151] S12, using absolute enhancement value images and normalized enhancement images to evaluate the degree of enhancement.

[0152] As an optional implementation of this invention, the enhancement degree assessment using absolute enhancement value images and normalized enhancement images includes: calculating the enhancement distribution information of the patient's tumor region based on the normalized enhancement image; and calculating the proportion of the portion of the tumor with enhancement greater than a set threshold to the total tumor volume.

[0153] Specifically, the present invention first calculates the enhancement distribution information of the patient's tumor region based on the calculated normalized enhancement image:

[0154]

[0155] Where H sub The histogram is calculated when (x) is reached, p(x) i ) is I′ sub (x i ) in x i The probability density at point I′ sub (x i ) is in x i The pixel value at that location.

[0156] The percentage of tumor volume with enhancement exceeding a set threshold is calculated, with the threshold incremented from 0.2 to 5.

[0157]

[0158] Where V sub V represents the tumor volume. enhenced For the enhancement level to be greater than the set threshold T threshold The volume is the volume, and VolumeRatio is the final calculated volume ratio that meets the requirements.

[0159] Therefore, the method for evaluating the activity of common abdominal solid tumors in children provided by this invention applies deep learning convolutional neural network technology to feature learning of enhanced abdominal CT medical image data of abdominal solid tumors in children and optimizes the model; it analyzes the changes in the degree of enhancement during plain scan and post-enhanced venous phase, thereby improving the model's ability to evaluate activity.

[0160] Furthermore, this invention utilizes neural networks to extract two-dimensional and high-order features from input abdominal enhanced CT image frames, thereby improving the model's discriminative ability. Simultaneously, through corresponding computer-aided diagnostic systems, it enables clinicians to perform high-level diagnoses, allowing primary care physicians and young doctors to acquire diagnostic capabilities comparable to those of senior experts, thus effectively improving the imaging assessment level of abdominal solid tumors in children in my country.

[0161] The present invention also provides an active imaging assessment device for common abdominal solid tumors in children, comprising: a processor and a memory;

[0162] Memory, used to store computer programs;

[0163] The processor is used to execute the above-described methods for assessing the activity of common abdominal solid tumors in children by calling a computer program.

[0164] Therefore, the active image assessment device for common abdominal solid tumors in children provided by the present invention applies deep learning convolutional neural network technology to feature learning of enhanced abdominal CT medical image data of abdominal solid tumors in children and optimizes the model; it analyzes the changes in the degree of enhancement during plain scan and post-enhanced venous phase, thereby improving the model's ability to assess activity.

[0165] The present invention also provides a computer-readable storage medium having a computer program stored thereon, characterized in that the program, when executed by a processor, implements the above-described method for evaluating the activity of common abdominal solid tumors in children using imaging.

[0166] Therefore, the computer-readable storage medium provided in this invention enables the application of deep learning convolutional neural network technology to feature learning of enhanced abdominal CT medical image data of pediatric abdominal solid tumors, and to optimize the model; it also analyzes the changes in enhancement degree during plain scan and post-enhanced venous phase, thereby improving the model's ability to assess activity.

[0167] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for assessing the activity of common abdominal solid tumors in children using imaging techniques, characterized in that, include: A tumor segmentation model is determined, wherein the tumor segmentation model is the UNETR model; The input image is resampled according to a preset standard interval; The resampled data is block-sampled according to the preset block size; The training samples are randomly rotated and augmented using time-of-test (TOT) augmentation techniques. The tumor segmentation model is trained using the training set with the segmentation annotation results of the tumor region annotated by the physician as the target, and the trained tumor segmentation model is obtained. The trained tumor segmentation model was validated and its parameters were tuned using a validation set. The trained tumor segmentation model was evaluated using a test set. The trained tumor segmentation model is used to segment the image to be segmented, and the tumor segmentation result is obtained and visualized. Acquire CT images of children and remove artifacts from the CT images of children; Artifact-removed plain CT images of children were registered with arterial and venous phase images. The tumor region in the tumor segmentation result is directly mapped onto the plain scan and arterial phase images after registration and transformation. The subtraction images of the arterial phase, venous phase and plain scan are calculated respectively to obtain the absolute enhancement value images of the tumor's artery and vein. The absolute enhancement value images are divided by the plain scan images to obtain the normalized enhancement images. The degree of enhancement is evaluated using the absolute enhancement value image and the normalized enhancement image; in: The registration of artifact-removed pediatric CT plain scan images and arterial and venous phase images includes: The tumor segmentation results and abdominal high-density bone were used to perform auxiliary feature point registration, wherein the registration evaluation criterion was based on mutual information (MI). The auxiliary feature point registration using the tumor segmentation results and abdominal high-density bone includes: Standardization of pediatric CT plain scan images after artifact removal was performed using Z-scores before registration. High-density bone was segmented using the Otsu adaptive threshold segmentation method, and voids were filled using a closing operation. Edge points and corner points were extracted from the obtained tumor and high-density bone regions, respectively. The edge points were obtained by edge detection using the Canny edge detection algorithm and points were selected at equal intervals at the edges. The corner points were obtained by the Harris corner detection algorithm. The nearest neighbor matching algorithm was used to find the corresponding feature point pairs in the artifact-removed pediatric CT plain scan image, the arterial phase image, and the venous phase image; Registration is performed using affine transformation, and the transformation parameters are estimated by minimizing the mean square error of the matching point pairs. A similarity metric is defined using mutual information and feature point matching error; Maximize the similarity measure using gradient descent; The iteration continues until the similarity metric no longer increases significantly and the feature point matching error is less than a preset value, at which point the iteration stops and the registration is complete.

2. The method according to claim 1, characterized in that, The process of training the tumor segmentation model using a training set with the segmentation annotation results of the tumor region annotated by the physician as the target, resulting in a trained model including: Using the backpropagation algorithm and stochastic gradient descent method, the weights of each layer are iteratively updated based on the magnitude of the loss function between the predicted segmentation result and the labeled result of the forward propagation. The training of the model is stopped when the loss function between the predicted segmentation result and the labeled result of the model tends to converge, and the trained model is obtained.

3. The method according to claim 1, characterized in that, The process of validating and tuning the trained tumor segmentation model using a validation set includes: By monitoring the Dice and IoU coefficients, the optimal model parameters and hyperparameter configurations can be selected.

4. The method according to claim 1, characterized in that, The preset standard spacing is 1.0 mm × 1.0 mm × 1.0 mm; the preset block size is 128 voxels × 128 voxels × 128 voxels.

5. The method according to claim 1, characterized in that, The process of acquiring pediatric CT images and removing artifacts from the pediatric CT images includes: The pediatric CT images were decomposed into different frequency sub-bands using wavelet transform; In the decomposed subbands, textures exhibiting high-frequency noise or abnormalities are detected, and the detected artifacts are thresholded. The sub-bands after thresholding are reconstructed back into the image space using inverse wavelet transform to obtain the CT image after artifact removal.

6. The method according to claim 1, characterized in that, The enhancement degree evaluation using the absolute enhancement value image and the normalized enhancement image includes: The enhancement distribution information of the patient's tumor region is calculated based on the normalized enhanced image; Calculate the proportion of the tumor volume whose enhancement value is greater than a set threshold.

7. A device for evaluating the activity of common abdominal solid tumors in children using imaging techniques, characterized in that, include: Processor, memory; The memory is used to store computer programs; The processor is configured to execute, by invoking the computer program, the method for assessing the activity of common abdominal solid tumors in children as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the program implements the method for assessing the activity of common abdominal solid tumors in children as described in any one of claims 1 to 6.

Images