Renal feature extraction method based on magnetic resonance image
By combining plain magnetic resonance imaging (MRI) and diffusion-weighted imaging with a renal parenchyma image segmentation model framework and ADC analysis, the problem of insufficient renal parenchyma feature extraction in MRI image analysis was solved, thus improving the accuracy and efficiency of renal function assessment.
Patent Information
- Application Number
- CN202511198638.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-26
- Publication Date
- 2025-11-18
AI Technical Summary
Existing magnetic resonance image analysis methods lack key features for extracting renal parenchymal image features, which affects the efficiency and accuracy of renal function assessment.
Bilateral renal MRI data of children with hydronephrosis were obtained using T2WI and diffusion-weighted imaging (DWI) sequences. Morphological, omics, and diffusion-weighted features of the hydronephrotic side and the contralateral kidney were extracted by combining a pre-established renal parenchyma image segmentation model framework and the apparent diffusion coefficient (ADC) analysis method.
This method enhances the diffusion-weighted features of the renal parenchyma, highlights the microstructural features of the renal parenchyma, strengthens feature extraction of the region of interest, and improves the accuracy and efficiency of renal function assessment.
Smart Images
Figure CN120976705A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application relates to the technical field of magnetic resonance image processing, and particularly relates to a kidney feature extraction method based on a magnetic resonance image. BACKGROUND
[0002] Hydronephrosis is a kidney damage caused by the obstruction of urine flow from the kidney to the bladder, which is common in pediatric urology diseases. The most common cause is the obstruction of the ureteropelvic junction (UPJO) due to the intrinsic abnormality of the proximal ureter. Hydronephrosis is not a static process, but a dynamic process, and the optimal timing of treatment is still controversial, especially in pediatric patients. Split renal function assessment is an important factor for surgical treatment. Previous reports suggest that children with poor preoperative split renal function have a poor prognosis. Therefore, it is necessary to determine an image biomarker that can non-invasively and accurately assess the split renal function of children with hydronephrosis.
[0003] The assessment of split renal function by radionuclide renal imaging is considered as the reference standard and may contribute to treatment decisions. However, the image quality of radionuclide renal imaging is poor, and the soft tissue resolution is low. In addition, due to the use of radioactive tracers and ionizing radiation, it cannot be used for frequent follow-up. Studies have reported that the renal parenchymal volume assessed by contrast-enhanced CT (CE-CT), non-contrast-enhanced CT (NCE-CT) and contrast-enhanced MRI (CE-MRI) has a moderate to excellent correlation with split renal function. However, CT examination has ionizing radiation, and gadolinium-containing contrast agents may increase the risk of nephrogenic systemic fibrosis. Non-contrast-enhanced MRU (NCE-MRU) is widely used in the assessment of pediatric patients with hydronephrosis due to its high resolution, no radiation exposure, no contrast agent, and short scanning time. Magnetic resonance urography (MRU) has been widely and routinely used for imaging of pediatric patients with hydronephrosis, ensuring a high soft tissue resolution and contrast under the premise of not using contrast agents to achieve renal parenchymal volume measurement. In the paper published by the research team, "Non contrast enhanced magnetic resonance urography for measuring split kidney function in pediatric patients with hydronephrosis: comparison with renal scintigraphy", the high correlation between split renal parenchymal volume and split renal function is demonstrated, avoiding radionuclides and ionizing radiation. Therefore, the assessment of split renal function can be performed by magnetic resonance image analysis.
[0004] However, although the prior art can realize the determination of renal morphological features through magnetic resonance image analysis to evaluate the assessment of renal function of hydronephrosis, the feature extraction method of the kidney parenchyma image with prominent key features is lacking in the magnetic resonance image analysis, which affects the kidney feature extraction effect, and finally affects the efficiency and accuracy performance of the renal function evaluation. SUMMARY
[0005] The purpose of the present application is to provide a kidney feature extraction method based on magnetic resonance images to solve the technical problem that the feature extraction method of the kidney parenchyma image with prominent key features is lacking in the magnetic resonance image analysis in the prior art, which affects the kidney feature extraction effect, and finally affects the efficiency and accuracy performance of the renal function evaluation.
[0006] To solve the above technical problems, the present application specifically provides the following technical solutions: A kidney feature extraction method based on magnetic resonance images, comprising the following steps: Obtaining double kidney magnetic resonance scanning data of a child with hydronephrosis through T2WI sequence and diffusion weighted imaging (DWI) sequence of magnetic resonance plain scan; Performing kidney region image segmentation in the double kidney magnetic resonance scanning data to determine the hydronephrosis side kidney region and the contralateral kidney region; Extracting the hydronephrosis side and contralateral kidney morphological features and omics features by performing feature extraction in the hydronephrosis side kidney region and the contralateral kidney region. Extracting the diffusion weighted features of the hydronephrosis side and contralateral kidney in the hydronephrosis side kidney region and the contralateral kidney region by combining the pre-established kidney parenchyma image segmentation model framework with the apparent diffusion coefficient (ADC) analysis method.
[0007] As a preferred scheme of the present application, the extraction method of the hydronephrosis side and contralateral kidney morphological features comprises: Extracting the morphological features of the hydronephrosis side and contralateral kidney in the hydronephrosis side kidney region and the contralateral kidney region of the double kidney magnetic resonance scanning data obtained by the T2WI sequence, respectively, the morphological features including: kidney thickness parameters, kidney volume parameters, and kidney diameter parameters.
[0008] As a preferred scheme of the present application, the extraction method of the hydronephrosis side and contralateral kidney omics features comprises: Extracting the omics features of the hydronephrosis side and contralateral kidney in the hydronephrosis side kidney region and the contralateral kidney region of the double kidney magnetic resonance scanning data obtained by the T2WI sequence by the gray level co-occurrence matrix method.
[0009] As a preferred scheme of the present application, the extraction method of the diffusion weighted features of the hydronephrosis side kidney comprises: The kidney parenchyma region in the hydrops side kidney determined by the hydrops side kidney parenchyma region mask enhanced image output by the kidney parenchyma image segmentation model framework is pixel superimposed and fused with the hydrops side kidney region to obtain a hydrops side kidney region with enhanced kidney parenchyma region , wherein, is the hydrops side kidney region with enhanced kidney parenchyma region, is the hydrops side kidney region, is the kidney parenchyma region corresponding to the hydrops side kidney parenchyma region mask enhanced image, is a pixel superimposition operation formula; The apparent diffusion coefficient ADC analysis is performed on the hydrops side kidney region with enhanced kidney parenchyma region to obtain a diffusion weighted feature of the hydrops side kidney.
[0010] As a preferred scheme of the present application, the extraction method of the diffusion weighted feature of the hydrops side kidney comprises: The kidney parenchyma region in the hydrops side kidney determined by the hydrops side kidney parenchyma region mask enhanced image output by the kidney parenchyma image segmentation model framework is generated into a spatial attention map of the kidney parenchyma region through a spatial attention mechanism; The spatial attention map is multiplied with the hydrops side kidney region element by element to obtain a hydrops side kidney region with enhanced kidney parenchyma region , wherein, is the hydrops side kidney region with enhanced kidney parenchyma region, is the hydrops side kidney region, is a spatial attention mechanism operation formula, is the kidney parenchyma region corresponding to the hydrops side kidney parenchyma region mask enhanced image, is an element by element multiplication operator; The apparent diffusion coefficient ADC analysis is performed on the hydrops side kidney region with enhanced kidney parenchyma region to obtain a diffusion weighted feature of the hydrops side kidney.
[0011] As a preferred scheme of the present application, the extraction method of the diffusion weighted feature of the contralateral kidney comprises: The apparent diffusion coefficient ADC analysis is performed on the contralateral kidney region determined by the contralateral kidney mask image output by the kidney region image segmentation model framework to obtain a diffusion weighted feature of the contralateral kidney.
[0012] As a preferred scheme of the present application, the diffusion weighted feature comprises a mean value, 5th, 25th, 50th, 75th and 90th percentile, non-uniformity, skewness, kurtosis and entropy.
[0013] In a preferred embodiment of the present invention, the apparent diffusion coefficient ADC analysis method generates ADC maps of the hydronephrotic kidney region and the contralateral kidney region using a single exponential model. The single exponential model analysis formula is as follows: ; In the formula, S(b) is the DWI signal value at a specified b value when diffusion exists in the region, S0 is the DWI signal value when diffusion does not exist in the region, and b is the b value, which is used to determine the degree of diffusion motion weighting in the DWI signal.
[0014] As a preferred embodiment of the present invention, the renal parenchyma image segmentation model framework includes a renal parenchyma region enhancement module based on the Unet network, and two renal parenchyma region segmentation modules for hydronephrotic kidneys based on the Mask R-CNN network. The input of the renal parenchyma region enhancement module based on the Unet network is used as the input of the renal parenchyma image segmentation model framework, and the output of the first renal parenchyma region segmentation module for hydronephrotic kidneys is used as the output of the renal parenchyma image segmentation model framework. The renal parenchyma enhancement module is used to segment the renal parenchyma region in the hydronephrotic kidney region using the Unet network to obtain a renal parenchyma mask image of the hydronephrotic kidney, and to perform pixel overlay and fusion of the renal parenchyma mask image of the hydronephrotic kidney and the hydronephrotic kidney region to obtain a hydronephrotic kidney region with enhanced renal parenchyma. The structural expression of the renal parenchyma enhancement module is as follows: ; In the formula, The renal parenchyma enhancement module outputs an enhanced version of the hydronephrotic kidney region. The area of the kidney on the side with hydronephrosis. This is a mask image of the renal parenchyma region of the hydronephrotic kidney obtained by segmentation using the Unet network. Unet is the Unet network itself. The enhancement coefficient of the renal parenchyma region; ; In the formula, GT en This is a true masked image of the renal parenchyma region of the hydronephrosis-related kidney. Masked enhanced image of the renal parenchyma region of the hydronephrosis-related kidney. This is a masked image of the renal parenchyma region of the hydronephrosis-related kidney. For KL divergence calculation, This is a pixel overlay operation formula; The first hydronephrotic kidney parenchyma segmentation module is used to segment the renal parenchyma region in the hydronephrotic kidney region with enhanced renal parenchyma region using the Mask R-CNN network, and obtain the mask-enhanced image of the hydronephrotic kidney parenchyma region. The structural expression of the first hydronephrosis kidney renal parenchyma region segmentation module is: ; In the formula, is a hydronephrosis kidney region mask enhanced image, is a Mask R-CNN network; The second hydronephrosis kidney renal parenchyma region segmentation module is used to segment the renal parenchyma region in the hydronephrosis kidney region by using the Mask R-CNN network, to obtain a hydronephrosis kidney renal parenchyma region mask image; The structural expression of the second hydronephrosis kidney renal parenchyma region segmentation module is: ; In the formula, is a hydronephrosis kidney renal parenchyma region mask image output by the second hydronephrosis kidney renal parenchyma region segmentation module.
[0015] As a preferred scheme of the present application, the kidney diameter parameters include a kidney anteroposterior diameter parameter, a kidney left-right diameter parameter, and a kidney superior-inferior diameter parameter.
[0016] Compared with the prior art, the present application has the following beneficial effects: The present application obtains the magnetic resonance plain scan and diffusion weighted imaging of the hydronephrosis child, uses the artificial intelligence algorithm to analyze and extract the kidney image features, and enhances the image of the renal parenchyma region in the diffusion weighted feature extraction process of the hydronephrosis kidney. After the enhanced hydronephrosis kidney image is analyzed by ADC, the diffusion weighted features of the renal parenchyma region can be enhanced, the key features in the diffusion weighted features can be more easily extracted on the basis of extracting the diffusion weighted features of the entire hydronephrosis kidney region, the diffusion weighted feature details of the renal parenchyma region can be highlighted, the microstructure features of the renal parenchyma region are highlighted, and the feature extraction of the region of interest is strengthened. BRIEF DESCRIPTION OF DRAWINGS
[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only exemplary, and those skilled in the art can obtain other implementation drawings according to the provided drawings without any creative labor.
[0018] Figure 1 The kidney feature extraction method flowchart based on magnetic resonance imaging provided by the embodiment of the present application; Figure 2 The kidney region image segmentation model framework schematic diagram provided by the embodiment of the present application; Figure 3 a kidney region image segmentation result provided by the embodiment of the present application; Figure 4 a kidney parenchyma image segmentation model framework schematic diagram provided by the embodiment of the present application; Figure 5 kidney morphological features and omics features provided by the embodiment of the present application; Figure 6 diffusion weighted features provided by the embodiment of the present application. DETAILED DESCRIPTION
[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative labor fall within the protection scope of the present application.
[0020] As shown in Figure 1 The present application provides a kidney feature extraction method based on magnetic resonance images, comprising the following steps: Obtain double kidney magnetic resonance scanning data of a hydronephrosis child through T2WI sequence and diffusion weighted imaging (DWI) sequence of magnetic resonance plain scanning; The image scanning parameters are as follows: T2WI scanning parameters: 3.0 T MR scanner, 8-channel heart coil, TR 5300 ms, TE 66.55 ms, layer thickness 3 mm, FA 110°, DWI parameters: TR / TE (ms), 4000 / shortest TE; matrix 128x96, bandwidth, 250 kHz; layer thickness 3 mm, b value (s / mm2) 0, 800.
[0021] Perform kidney region image segmentation in the double kidney magnetic resonance scanning data to determine the hydronephrosis side kidney region and the contralateral kidney region; Obtain the hydronephrosis side and contralateral kidney morphological features and omics features by performing feature extraction in the hydronephrosis side kidney region and the contralateral kidney region; Extract the diffusion weighted features of the hydronephrosis side and contralateral kidney in the hydronephrosis side kidney region and the contralateral kidney region through the pre-established kidney parenchyma image segmentation model framework combined with apparent diffusion coefficient (ADC) analysis method.
[0022] In order to obtain the morphological characteristics of the water accumulation side and the contralateral kidney, first, image segmentation of the water accumulation side and the contralateral kidney (or normal side) is performed on the dual kidney magnetic resonance scan data or the magnetic resonance scan image, that is, the water accumulation side and the contralateral kidney (or normal side) region is segmented on the magnetic resonance scan image, and then the morphological characteristics of the kidney are extracted according to the segmented water accumulation side kidney region and the contralateral kidney region.
[0023] As shown in Figure 2 , the dual kidney magnetic resonance scan data is subjected to kidney region image segmentation through a pre-established kidney region image segmentation model framework to determine the water accumulation side kidney region and the contralateral kidney region, as shown in Figure 3 . The kidney region image segmentation model framework includes a water accumulation side kidney enhancement module based on a Unet network and two water accumulation side kidney and contralateral kidney segmentation modules based on Mask R-CNN networks, the input of the water accumulation side kidney enhancement module based on the Unet network is taken as the input of the kidney region image segmentation model framework, and the output of the first water accumulation side kidney and contralateral kidney segmentation module is taken as the output of the kidney region image segmentation model framework. The water accumulation side kidney enhancement module is used to segment the water accumulation side kidney region from the dual kidney magnetic resonance scan data by using the Unet network to obtain a water accumulation side kidney mask image, and to fuse the water accumulation side kidney mask image and the dual kidney magnetic resonance scan data to obtain water accumulation side kidney enhanced dual kidney magnetic resonance scan data. The Unet network is a convolutional neural network for image segmentation, which is in the shape of "U" and mainly consists of an encoder (Contracting Path) and a decoder (Expanding Path). The encoder part gradually extracts features through convolutional layers and pooling layers, while the decoder part gradually restores the resolution of the image through upsampling and convolution operations. The uniqueness of U-Net lies in its skip connections, which directly pass the feature maps of the encoder part to the decoder part, helping to restore more accurate details. This design enables UNet to better preserve detailed information when processing image segmentation tasks, thereby improving the accuracy and precision of segmentation.
[0024] Mask R-CNN (Mask Region-based Convolutional Neural Network) is a target detection and instance segmentation model extended from Faster R-CNN. It not only can detect the position of objects in the image and identify their categories, but also can generate pixel-level segmentation masks for each detected object, thereby achieving more fine object segmentation.
[0025] The structural expression of the hydronephrosis side kidney enhancement module is: ; In the formula, is the double kidney magnetic resonance scan data of the hydronephrosis side kidney enhancement, is the double kidney magnetic resonance scan data, is the hydronephrosis side kidney mask image segmented by the Unet network, and Unet is the Unet network, is the enhancement coefficient of the hydronephrosis side kidney, is a pixel superposition operation formula; ; In the formula, GT is the true mask image of the hydronephrosis side kidney region, is the hydronephrosis side kidney mask enhancement image, is the hydronephrosis side kidney mask image output by the second hydronephrosis side kidney and contralateral kidney segmentation module, is the KL divergence operation, and correspond to the compare module in Figure 3 ; The first hydronephrosis side kidney and contralateral kidney segmentation module is used to segment the hydronephrosis side kidney region and the contralateral kidney region in the double kidney magnetic resonance scan data of the hydronephrosis side kidney enhancement by using the Mask R-CNN network, to obtain the hydronephrosis side kidney mask enhancement image and the contralateral kidney mask image; The structural expression of the first hydronephrosis side kidney and contralateral kidney segmentation module is: ; In the formula, is the contralateral kidney mask image, is the hydronephrosis side kidney mask enhancement image, and Mask R-CNN is the Mask R-CNN network; The second hydronephrosis side kidney and contralateral kidney segmentation module is used to segment the hydronephrosis side kidney region and the contralateral kidney region in the double kidney magnetic resonance scan data by using the Mask R-CNN network, to obtain the hydronephrosis side kidney mask image and the contralateral kidney mask image; The structural expression of the second hydronephrosis side kidney and contralateral kidney segmentation module is: ; In the formula, is the hydronephrosis side kidney mask image output by the second hydronephrosis side kidney and contralateral kidney segmentation module.
[0026] The application constructs a kidney region image segmentation model framework for implementing segmentation of the hydronephrosis side kidney region and the contralateral kidney region, which is composed of a hydronephrosis side kidney enhancement module based on a Unet network and two hydronephrosis side kidney and contralateral kidney segmentation modules based on Mask R-CNN networks.
[0027] Since the application expects to evaluate the split kidney function of children with hydronephrosis, the hydronephrosis side kidney is more concerned in the evaluation process, that is, the hydronephrosis side kidney morphological characteristics play a dominant role, that is, the accuracy requirement for extracting the kidney morphological characteristics is higher, and the unilateral enhancement of the hydronephrosis side kidney region is adopted to meet the accuracy requirement of the hydronephrosis side kidney morphological characteristics, while the corresponding normal side kidney morphological characteristics play an auxiliary role, so that the normal side kidney can be reduced in requirements and does not need to be enhanced, which can meet the requirement of kidney morphological characteristic extraction while reducing the redundant calculation amount.
[0028] Then, the double kidney magnetic resonance scanning data with the enhanced details of the hydronephrosis side kidney is reprocessed through the Mask R-CNN network structure (i.e., the first hydronephrosis side kidney and contralateral kidney segmentation module) for higher-precision segmentation of the hydronephrosis side kidney region, and more accurate segmentation results of the hydronephrosis side kidney region are obtained through the enhanced details.
[0029] In the kidney region image segmentation model framework, a Mask R-CNN network structure (i.e., the second hydronephrosis side kidney and contralateral kidney segmentation module) is also provided for segmentation of the hydronephrosis side kidney region on the original double kidney magnetic resonance scanning data, which serves as a quantitative component for the hydronephrosis side kidney enhancement effect, that is, it monitors the segmentation performance of the hydronephrosis side kidney region obtained on the original double kidney magnetic resonance scanning data, and is used for comparison with the segmentation performance of the hydronephrosis side kidney region obtained on the double kidney magnetic resonance scanning data with the enhanced details of the hydronephrosis side kidney. When the former performance is lower than the latter, the feedback is given to the enhancement coefficient of the hydronephrosis side kidney The start of the hydronephrosis side kidney enhancement module is triggered, that is, the segmentation of the first hydronephrosis side kidney and contralateral kidney segmentation module is based on the double kidney magnetic resonance scanning data with the enhanced hydronephrosis side kidney In addition, the hydronephrosis side kidney preliminary segmentation performance in the hydronephrosis side kidney enhancement module is integrated , The lower, the better the preliminary segmentation performance of the hydronephrosis side kidney is, the higher the reliability of the preliminary segmentation result of the hydronephrosis side kidney superimposed on the original double-kidney magnetic resonance scanning data should be, the higher the superimposition degree corresponds to the setting, and the better the detail enhancement effect will be, so that the setting When The lower, The higher, the high-reliability superimposition can be ensured, and the segmentation of the final first hydronephrosis side kidney and contralateral kidney segmentation module is based on the double-kidney magnetic resonance scanning data The redundant enhancement operation is added to reduce the efficiency in exchange for the accuracy improvement.
[0030] When the performance of the former is higher than or equal to the latter, it is indicated that the enhancement of the hydronephrosis side kidney does not bring positive effects to improve the segmentation performance, and the feedback is also fed back to the enhancement coefficient of the hydronephrosis side kidney The upper, The closing of the hydronephrosis side kidney enhancement module is triggered, and the segmentation of the final first hydronephrosis side kidney and contralateral kidney segmentation module is based on the double-kidney magnetic resonance scanning data The upper, that is, based on the original double-kidney magnetic resonance scanning data, at this time, the region enhancement operation is a redundant operation, and does not need to be performed, and the efficiency is improved in exchange for the accuracy.
[0031] In summary, the present application can realize the self-adaptation of the hydronephrosis side kidney detail enhancement, the hydronephrosis side kidney detail enhancement is performed in the case of improving the segmentation performance, and the hydronephrosis side kidney detail enhancement is not needed in the case of not improving the segmentation performance, the redundant calculation is reduced, the adaptive enhancement can achieve the balance between the segmentation accuracy and the segmentation efficiency, and therefore the high-efficiency and accurate kidney region image segmentation can be achieved.
[0032] The method for extracting the morphological features of the hydronephrosis side kidney and the contralateral kidney comprises: Morphological features of the hydronephrosis side kidney and the contralateral kidney are extracted in the hydronephrosis side kidney region and the contralateral kidney region of the double-kidney magnetic resonance scanning data obtained by the T2WI sequence, and the morphological features comprise kidney thickness parameters, kidney volume parameters and kidney diameter parameters.
[0033] The method for extracting the histological features of the hydronephrosis side kidney and the contralateral kidney comprises: The histological features of the hydronephrosis side kidney and the contralateral kidney are extracted in the hydronephrosis side kidney region and the contralateral kidney region of the double-kidney magnetic resonance scanning data obtained by the T2WI sequence by a gray level co-occurrence matrix method.
[0034] The method for extracting the diffusion-weighted features of the hydronephrosis side kidney comprises: The kidney parenchyma region in the hydrops side kidney determined by the hydrops side kidney kidney parenchyma region mask enhanced image output by the kidney parenchyma image segmentation model framework is pixel superimposed and fused with the hydrops side kidney region to obtain a hydrops side kidney region with enhanced kidney parenchyma region , wherein, is the hydrops side kidney region with enhanced kidney parenchyma region, is the hydrops side kidney region, is the kidney parenchyma region corresponding to the hydrops side kidney kidney parenchyma region mask enhanced image, is a pixel superimposition operation formula; The apparent diffusion coefficient ADC analysis is performed on the hydrops side kidney region with enhanced kidney parenchyma region to obtain the diffusion weighted feature of the hydrops side kidney.
[0035] The extraction method of the diffusion weighted feature of the hydrops side kidney includes: The kidney parenchyma region in the hydrops side kidney determined by the hydrops side kidney kidney parenchyma region mask enhanced image output by the kidney parenchyma image segmentation model framework is generated into a spatial attention map of the kidney parenchyma region through a spatial attention mechanism; The spatial attention map is multiplied with the hydrops side kidney region element by element to obtain a hydrops side kidney region with enhanced kidney parenchyma region , wherein, is the hydrops side kidney region with enhanced kidney parenchyma region, is the hydrops side kidney region, is a spatial attention mechanism operation formula, is the kidney parenchyma region corresponding to the hydrops side kidney kidney parenchyma region mask enhanced image, is an element-by-element multiplication operator; The operation process of the spatial attention mechanism performs average pooling (calculating the average value of each spatial position along the channel dimension) and maximum pooling (taking the maximum value of each spatial position along the channel dimension) on the kidney parenchyma region in the hydrops side kidney determined by the hydrops side kidney kidney parenchyma region mask enhanced image output by the kidney parenchyma image segmentation model framework, concatenates the features, performs convolution processing: applies a convolution kernel (such as 7x7) to the concatenated feature map, and finally maps the convolution output to the [0, 1] interval through the Sigmoid function to obtain the spatial attention map.
[0036] The apparent diffusion coefficient ADC analysis is performed on the hydrops side kidney region with enhanced kidney parenchyma region to obtain the diffusion weighted feature of the hydrops side kidney.
[0037] After the renal parenchymal region is accurately segmented from the hydronephrosis side kidney region, the renal parenchymal region is superimposed and fused into the hydronephrosis side kidney region, image enhancement of the renal parenchymal region in the hydronephrosis side kidney region is realized, and after ADC analysis, the diffusion weighted features corresponding to the renal parenchymal region can be enhanced, so that the key features in the diffusion weighted features can be more easily extracted on the basis of the diffusion weighted features of the entire hydronephrosis side kidney region.
[0038] The method of superimposing and fusing the renal parenchymal region into the hydronephrosis side kidney region after the renal parenchymal region is accurately segmented from the hydronephrosis side kidney region has two kinds, the first kind is to superimpose and fuse the renal parenchymal region in the hydronephrosis side kidney determined by the hydronephrosis image segmentation model framework output renal parenchymal region mask enhanced image into the hydronephrosis kidney region, to obtain the renal parenchymal region enhanced hydronephrosis side kidney region , directly enhance the details of the key region of the image at the pixel level, so as to highlight the key features in the subsequent ADC extraction of the diffusion weighted features.
[0039] The second kind is to generate a spatial attention map of the renal parenchymal region through a spatial attention mechanism, multiply the spatial attention map with the hydronephrosis side kidney region element by element, and obtain the renal parenchymal region enhanced hydronephrosis side kidney region , the spatial attention mechanism is used to excavate the key region (i.e. the renal parenchymal region) in the hydronephrosis side kidney region, and high weight is given to the key region, and the high weight is used to enhance the key region of the image, so as to highlight the key features in the subsequent ADC extraction of the diffusion weighted features.
[0040] In actual use, the enhancement method can be selected as needed.
[0041] The method for extracting the diffusion weighted features of the contralateral kidney includes: Performing apparent diffusion coefficient ADC analysis on the contralateral kidney region determined by the kidney region image segmentation model framework output contralateral kidney mask image to obtain the diffusion weighted features of the contralateral kidney.
[0042] The diffusion weighted features include mean, 5th, 25th, 50th, 75th and 90th percentiles, heterogeneity, skewness, kurtosis and entropy, as shown in Figure 6 .
[0043] In the apparent diffusion coefficient ADC analysis method, the ADC maps of the hydronephrosis side kidney region and the contralateral kidney region are generated by single exponential model analysis, as shown in Figure 5 , and the single exponential model analysis formula is: ; In the formula, S (b) is the DWI signal value at a specified b value when diffusion exists in the region, S0 is the DWI signal value when diffusion does not exist in the region, and b is the b value used to determine the degree of diffusion motion weighting in the DWI signal.
[0044] As shown in Figure 4 The renal parenchyma image segmentation model framework includes a renal parenchyma region enhancement module based on a Unet network and two Mask R-CNN network-based hydronephrosis side kidney renal parenchyma region segmentation modules, the input of the renal parenchyma region enhancement module based on the Unet network is taken as the input of the renal parenchyma image segmentation model framework, and the output of the first hydronephrosis side kidney renal parenchyma region segmentation module is taken as the output of the renal parenchyma image segmentation model framework. The renal parenchyma region enhancement module is used to segment the renal parenchyma region in the hydronephrosis side kidney region by using the Unet network to obtain a hydronephrosis side kidney renal parenchyma mask image, and to perform pixel superposition fusion on the hydronephrosis side kidney renal parenchyma mask image and the hydronephrosis side kidney region to obtain a hydronephrosis side kidney region with enhanced renal parenchyma region. The structural expression of the renal parenchyma region enhancement module is: ; In the formula, is the hydronephrosis side kidney region with enhanced renal parenchyma region output by the renal parenchyma region enhancement module, is the hydronephrosis side kidney region, is the hydronephrosis side kidney renal parenchyma region mask image segmented by the Unet network, and Unet is the Unet network, is the enhancement coefficient of the renal parenchyma region; ; In the formula, GT en is the true mask image of the hydronephrosis side kidney renal parenchyma region, is the hydronephrosis side kidney renal parenchyma region mask enhancement image, is the hydronephrosis side kidney renal parenchyma region mask image, is the KL divergence operation, is a pixel superposition operation formula; The first hydronephrosis side kidney renal parenchyma region segmentation module is used to segment the renal parenchyma region in the hydronephrosis side kidney region with enhanced renal parenchyma region by using the Mask R-CNN network to obtain a hydronephrosis side kidney renal parenchyma region mask enhancement image. The structural expression of the first hydronephrosis side kidney renal parenchyma region segmentation module is: ; In the formula, is the hydronephrosis side kidney renal parenchyma region mask enhancement image, a Mask R-CNN network; The second hydronephrosis-side kidney renal parenchyma region segmentation module is configured to segment the renal parenchyma region in the hydronephrosis-side kidney region by using the Mask R-CNN network to obtain a hydronephrosis-side kidney renal parenchyma region mask image. The structure expression of the second hydronephrosis-side kidney renal parenchyma region segmentation module is as follows: In the formula, is the hydronephrosis-side kidney renal parenchyma region mask image output by the second hydronephrosis-side kidney renal parenchyma region segmentation module.
[0045] The renal parenchyma is the main functional tissue of the kidney, which is located below the renal capsule and is composed of the renal cortex and the renal medulla. The renal cortex is located on the outer layer and contains glomeruli and renal tubules. The renal medulla is located on the inner layer and is composed of renal pyramids and collecting ducts. The diffusion-weighted feature is mainly in the renal parenchyma, and therefore, the ADC analysis mainly focuses on the renal parenchyma region.
[0046] In the extraction of the diffusion-weighted feature of the hydronephrosis-side kidney, the hydronephrosis-side kidney is taken as the region of interest, and the hydronephrosis-side kidney contains the renal parenchyma region and the hydronephrosis region. Since the diffusion-weighted feature is mainly in the renal parenchyma, the renal parenchyma region in the hydronephrosis-side kidney is more interesting in the region of interest. That is, the ADC analysis of the hydronephrosis-side kidney focuses on the entire hydronephrosis-side kidney region and more focuses on the renal parenchyma region in the hydronephrosis-side kidney region. In this way, similar to the above-mentioned hydronephrosis-side kidney enhancement, the renal parenchyma region in the hydronephrosis-side kidney is also enhanced, and the hydronephrosis-side kidney after the enhancement of the renal parenchyma region is subjected to ADC analysis to obtain the diffusion-weighted feature of the entire hydronephrosis-side kidney. The diffusion-weighted feature details of the renal parenchyma region are highlighted in the overall feature, the microstructure feature of the renal parenchyma region is highlighted, and the feature extraction of the region of interest is strengthened.
[0047] Therefore, this invention first performs precise segmentation of the renal parenchyma region in the hydronephrosis-side kidney, and then superimposes it onto the hydronephrosis-side kidney region to achieve precise enhancement of the renal parenchyma region in the hydronephrosis-side kidney. This invention transfers the kidney region image segmentation model framework to the renal parenchyma image segmentation model framework (using the hydronephrosis-side kidney region from the bilateral renal MRI scan data obtained from the DWI sequence as the input of the kidney region image segmentation model framework, and the renal parenchyma region in the hydronephrosis-side kidney as the output of the renal parenchyma image segmentation model framework, and performs transfer learning on the kidney region image segmentation model framework to achieve renal parenchyma image segmentation model framework in the hydronephrosis-side kidney region), which is used to achieve renal parenchyma region segmentation in the hydronephrosis-side kidney region. It consists of a renal parenchyma region enhancement module based on the Unet network, and two Mask-based modules. The R-CNN network consists of a segmentation module for the renal parenchyma region of the hydronephrotic kidney. The renal parenchyma enhancement module first uses the Unet network to initially segment the renal parenchyma region of the hydronephrotic kidney region (determined by the kidney region image segmentation model framework) on the original hydronephrotic kidney region. The segmentation mask result of the renal parenchyma region is then superimposed and fused onto the original hydronephrotic kidney region to form an enhanced hydronephrotic kidney region. This can enhance a specific region (the renal parenchyma region) in the original hydronephrotic kidney region while keeping other regions unchanged, thereby highlighting the details of the renal parenchyma region.
[0048] Then, the hydronephrosis-side kidney region, with enhanced details, is re-segmented using the Mask R-CNN network structure (i.e., the first hydronephrosis-side kidney parenchyma region segmentation module) for more accurate renal parenchyma region segmentation, obtaining more precise renal parenchyma region segmentation results through enhanced details.
[0049] The renal parenchyma image segmentation model framework also includes a Mask R-CNN network structure (i.e., a segmentation module for the renal parenchyma region of the second hydronephrotic kidney). This module is used to segment the renal parenchyma region on the original hydronephrotic kidney region. It essentially acts as a quantification component for enhancing the renal parenchyma region, monitoring the segmentation performance of the renal parenchyma region obtained on the original hydronephrotic kidney region and comparing it with the segmentation performance of the renal parenchyma region obtained on the hydronephrotic kidney region with enhanced renal parenchyma details. When the former's performance is lower than the latter's, it is fed back to the enhancement coefficient of the renal parenchyma region. superior, The activation of the renal parenchymal region enhancement module is triggered, meaning the segmentation of the first hydronephrosis-side renal parenchymal region segmentation module is based on the hydronephrosis-side renal parenchymal region with enhanced renal parenchymal region. Furthermore, it integrates the preliminary segmentation performance of the renal parenchyma region from the renal parenchyma region enhancement module. , The lower, the better the preliminary segmentation performance of the renal parenchymal region is, the higher the reliability of the preliminary segmentation result of the renal parenchymal region superimposed on the original hydronephrosis side kidney region should be, and the higher the superimposition degree corresponding setting is, the better the detail enhancement effect will be, so that the setting When The lower, The higher, the high reliability of the superimposition can be ensured, and the segmentation of the final first hydronephrosis side kidney renal parenchymal region is based on the enhanced hydronephrosis side kidney region , and the redundant enhancement operation is increased to reduce the efficiency in exchange for the accuracy improvement.
[0050] When the former performance is higher than or equal to the latter, it indicates that the enhancement of the renal parenchymal region does not bring positive effect to improve the segmentation performance, and the enhancement coefficient of the renal parenchymal region is fed back to trigger the closing of the renal parenchymal region enhancement module, and the segmentation of the final first hydronephrosis side kidney renal parenchymal region is based on the hydronephrosis side kidney region , that is, based on the original hydronephrosis side kidney region, at this time, the region enhancement operation is redundant operation, and there is no need to perform it, and the efficiency is improved in exchange for the accuracy.
[0051] In summary, the present application can realize the self-adaptation of the renal parenchymal region detail enhancement, and the renal parenchymal region detail enhancement is performed in the case of improving the segmentation performance, and the renal parenchymal region detail enhancement is not needed in the case of not improving the segmentation performance, so that the redundant calculation is reduced, the adaptive enhancement can achieve the balance between the segmentation accuracy and the segmentation efficiency, and therefore the high-efficiency and accurate renal parenchymal region segmentation can be achieved.
[0052] The renal diameter parameters include the front-to-back diameter parameter, the left-to-right diameter parameter and the superior-to-inferior diameter parameter of the kidney.
[0053] The present application obtains the image data of magnetic resonance plain scanning and diffusion weighted imaging of a child with hydronephrosis, analyzes and extracts the renal image features by using an artificial intelligence algorithm, enhances the image of the renal parenchymal region in the diffusion weighted feature extraction process of the hydronephrosis side kidney, and after the enhanced hydronephrosis side kidney image is analyzed by using ADC, the diffusion weighted features of the renal parenchymal region can be enhanced, the key features in the diffusion weighted features are more easily extracted on the basis of extracting the diffusion weighted features of the entire hydronephrosis side kidney region, the diffusion weighted feature details of the renal parenchymal region are highlighted, the microstructure features of the renal parenchymal region are highlighted, and the feature extraction of the region of interest is strengthened.
[0054] The above examples are only exemplary embodiments of the present application, and are not intended to limit the present application, and the protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements to the present application within the spirit and protection scope of the present application, and such modifications or equivalent replacements are also considered to fall within the protection scope of the present application.
Claims
1. A method for extracting kidney features based on magnetic resonance imaging, characterized in that, Includes the following steps: Bilateral renal MRI data of children with hydronephrosis were obtained using T2WI sequences and diffusion-weighted imaging (DWI) sequences. In the bilateral renal magnetic resonance scan data, kidney region image segmentation was performed to determine the region of the hydronephrotic kidney and the contralateral kidney region; By extracting features from the hydronephrotic kidney region and the contralateral kidney region, the morphological and omics features of the hydronephrotic kidney and the contralateral kidney are obtained. By combining a pre-established renal parenchyma image segmentation model framework with the apparent diffusion coefficient (ADC) analysis method, diffusion-weighted features of the hydronephrotic and contralateral kidneys are extracted in the hydronephrotic and contralateral kidney regions.
2. The method for extracting kidney features based on magnetic resonance imaging according to claim 1, characterized in that: The methods for extracting the morphological features of the hydronephrotic side and the contralateral kidney include: Morphological features of the hydronephrotic kidney and the contralateral kidney were extracted from bilateral renal magnetic resonance imaging data obtained from T2WI sequences. These morphological features included kidney thickness parameters, kidney volume parameters, and kidney diameter parameters.
3. The method for extracting kidney features based on magnetic resonance imaging according to claim 1, characterized in that: The methods for extracting the renal omics features of the hydronephrotic side and the contralateral kidney include: Omics features of the hydronephrotic kidney and the contralateral kidney were extracted from bilateral renal magnetic resonance imaging data obtained from T2WI sequences using the gray-level co-occurrence matrix method.
4. The method for extracting kidney features based on magnetic resonance imaging according to claim 1, characterized in that: The method for extracting the diffusion-weighted features of the hydronephrotic kidney includes: The renal parenchymal region of the hydronephrosis side kidney, determined by the mask-enhanced image of the hydronephrosis side kidney output by the renal parenchymal image segmentation model framework, is pixel-wise superimposed and fused with the hydronephrosis side kidney region to obtain the enhanced renal parenchymal region of the hydronephrosis side kidney. In the formula, The hydronephrotic region shows enhancement of the renal parenchyma. The area of the kidney on the side with hydronephrosis. This is the renal parenchyma region corresponding to the masked enhanced image of the renal parenchyma region of the hydronephrosis side. This is a pixel overlay operation formula; Apparent diffusion coefficient (ADC) analysis was performed on the hydronephrotic region with enhanced renal parenchyma to obtain the diffusion-weighted characteristics of the hydronephrotic kidney.
5. The method for extracting kidney features based on magnetic resonance imaging according to claim 1, characterized in that: The method for extracting the diffusion-weighted features of the hydronephrotic kidney includes: The renal parenchymal region in the hydronephrosis side kidney, determined by the mask-enhanced image of the renal parenchymal region of the hydronephrosis side kidney output by the renal parenchymal image segmentation model framework, is used to generate a spatial attention map of the renal parenchymal region through a spatial attention mechanism. The spatial attention map is then multiplied element-wise with the hydronephrosis region to obtain the hydronephrosis region with enhanced renal parenchyma. In the formula, The hydronephrotic region shows enhancement of the renal parenchyma. The area of the kidney on the side with hydronephrosis. This is the spatial attention mechanism operation formula. This is the renal parenchyma region corresponding to the masked enhanced image of the renal parenchyma region of the hydronephrosis side. This is the element-wise multiplication operator; Apparent diffusion coefficient (ADC) analysis was performed on the hydronephrotic region with enhanced renal parenchyma to obtain the diffusion-weighted characteristics of the hydronephrotic kidney.
6. The method for extracting kidney features based on magnetic resonance imaging according to claim 1, characterized in that: The method for extracting the diffusion-weighted features of the contralateral kidney includes: Apparent diffusion coefficient ADC analysis was performed on the contralateral kidney region determined by the contralateral kidney mask image output by the kidney region image segmentation model framework to obtain the diffusion-weighted features of the contralateral kidney.
7. The method for extracting kidney features based on magnetic resonance imaging according to claim 1, characterized in that: The diffusion-weighted features include mean, 5th, 25th, 50th, 75th and 90th percentiles, heterogeneity, skewness, kurtosis and entropy.
8. The method for extracting kidney features based on magnetic resonance imaging according to claim 1, characterized in that: The apparent diffusion coefficient ADC analysis method generates ADC maps of the hydronephrotic kidney region and the contralateral kidney region using a single exponential model. The single exponential model analysis formula is as follows: ； In the formula, S(b) is the DWI signal value at a specified b value when diffusion exists in the region, S0 is the DWI signal value when diffusion does not exist in the region, and b is the b value, which is used to determine the degree of diffusion motion weighting in the DWI signal.
9. The method for extracting kidney features based on magnetic resonance imaging according to claim 1, characterized in that: The renal parenchyma image segmentation model framework includes a renal parenchyma region enhancement module based on the Unet network, and two renal parenchyma region segmentation modules for hydronephrotic kidneys based on the Mask R-CNN network. The input of the renal parenchyma region enhancement module based on the Unet network is used as the input of the renal parenchyma image segmentation model framework, and the output of the first hydronephrotic kidney renal parenchyma region segmentation module is used as the output of the renal parenchyma image segmentation model framework. The renal parenchyma enhancement module is used to segment the renal parenchyma region in the hydronephrotic kidney region using the Unet network to obtain a renal parenchyma mask image of the hydronephrotic kidney, and to perform pixel overlay and fusion of the renal parenchyma mask image of the hydronephrotic kidney and the hydronephrotic kidney region to obtain a hydronephrotic kidney region with enhanced renal parenchyma. The structural expression of the renal parenchyma enhancement module is as follows: ； In the formula, The renal parenchyma enhancement module outputs an enhanced version of the hydronephrotic kidney region. The area of the kidney on the side with hydronephrosis. This is a mask image of the renal parenchyma region of the hydronephrotic kidney obtained by segmentation using the Unet network. Unet is the Unet network itself. The enhancement coefficient of the renal parenchyma region; ； In the formula, GT en This is a true masked image of the renal parenchyma region of the hydronephrosis-related kidney. Masked enhanced image of the renal parenchyma region of the hydronephrosis-related kidney. This is a masked image of the renal parenchyma region of the hydronephrosis-related kidney. For KL divergence calculation, This is a pixel overlay operation formula; The first hydronephrotic kidney parenchyma segmentation module is used to segment the renal parenchyma region in the hydronephrotic kidney region with enhanced renal parenchyma region using the Mask R-CNN network, and obtain the mask-enhanced image of the hydronephrotic kidney parenchyma region. The structural expression for the segmentation module of the renal parenchyma region of the first hydronephrotic kidney is: ； In the formula, This is a masked enhanced image of the renal region of the hydronephrosis-related kidney. For Mask R-CNN network; The second hydronephrotic kidney parenchyma segmentation module is used to segment the renal parenchyma region in the hydronephrotic kidney region using the Mask R-CNN network to obtain a mask image of the renal parenchyma region of the hydronephrotic kidney. The structural expression for the segmentation module of the renal parenchyma region of the second hydronephrotic kidney is as follows: ； In the formula, The image is a mask image of the renal parenchyma region of the hydronephrotic kidney, output by the segmentation module of the renal parenchyma region of the second hydronephrotic kidney.
10. A method for extracting kidney features based on magnetic resonance imaging according to claim 2, characterized in that: The parameters of the kidney diameter include the anteroposterior diameter, lateral diameter, and superior-inferior diameter.