MR image-based vascular tissue segmentation method and system, and storage medium

Abstract

The invention relates to the technical field of medical image processing, in particular to a vascular tissue segmentation method and system based on an MR image and a storage medium, and the method comprises the following steps: enhancing blood vessel information in the MR image by using a Canny operator to obtain an MR enhanced image; respectively constructing a first segmentation model and a second segmentation model based on the MR image and the MR enhanced image by using a UNet network; and utilizing the second segmentation model to guide the first segmentation model to train, and taking an output result of the trained first segmentation model as a segmentation result of the vascular tissue in the MR image. According to the method, the segmentation result on the low-resolution image can be guided to learn from the segmentation result of the high-resolution image, so that the high-precision segmentation result can be directly obtained on the low-resolution image, the data processing process and the model complexity of the segmentation process do not need to be increased, and the accurate segmentation efficiency of the vascular tissue is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical image processing, and particularly relates to a method, a system and a storage medium for segmenting vascular tissues based on MR images. Background Art

[0002] Vascular segmentation aims to automatically segment vascular tissue regions from multi-modal magnetic resonance (MR) images taken by advanced medical imaging devices. By segmenting blood vessels, the morphology and location of blood vessels can be provided, which play a crucial role in disease diagnosis and monitoring.

[0003] In the prior art, deep learning algorithms are usually used to segment vascular tissues in MR images. Such vascular tissue segmentation models established based on deep learning usually enhance the features of vascular tissues in MR images and improve the image resolution before performing segmentation to obtain high-precision segmentation results. Therefore, in the prior art, high-precision segmentation results need to be based on high-resolution images, and it is difficult to directly obtain high-precision segmentation results on low-resolution images. To obtain the same high-precision segmentation results, corresponding resolution improvement processing is required, which correspondingly increases the data processing process and model complexity of the segmentation process, restricting the efficiency of accurate segmentation of vascular tissues. Summary of the Invention

[0004] The purpose of the present invention is to provide a method, a system and a storage medium for segmenting vascular tissues based on MR images to solve the technical problem that it is difficult to directly obtain high-precision resolution results on low-resolution images in the prior art, which restricts the efficiency of accurate segmentation of vascular tissues.

[0005] To solve the above technical problems, the present invention specifically provides the following technical solutions: A method for segmenting vascular tissues based on MR images, comprising the following steps: Obtain an MR image containing vascular tissues; Enhance the vascular information in the MR image by using the Canny operator to obtain an MR enhanced image; Use the UNet network to respectively construct a first segmentation model for segmenting vascular tissues on the MR image and a second segmentation model for segmenting vascular tissues on the MR enhanced image based on the MR image and the MR enhanced image; Use the second segmentation model to guide the training of the first segmentation model, and use the output result of the trained first segmentation model as the segmentation result of vascular tissues in the MR image.

[0006] As a preferred solution of the present invention, the enhancement method of the MR image includes: Edge detection is performed on the MR image using the Canny operator to obtain the edge detection image of the MR image; The edge detection image is superimposed and fused with the MR image to obtain the MR enhanced image.

[0007] As a preferred solution of the present invention, the construction method of the first segmentation model includes: Taking the UNet network as the network structure of the first segmentation model; Taking the MR image as the input of the UNet network and the vascular tissue mask annotation data in the MR image as the output of the UNet network to construct the first segmentation model; The first segmentation model is: ; In the formula, is the vascular tissue mask of the MR image, is the MR image, and UNet is the UNet network.

[0008] As a preferred solution of the present invention, the construction method of the second segmentation model includes: Taking the UNet network as the network structure of the second segmentation model; Taking the MR enhanced image as the input of the UNet network and the vascular tissue mask annotation data in the MR enhanced image as the output of the UNet network to construct the second segmentation model; The second segmentation model is: ; In the formula, is the vascular tissue mask of the MR enhanced image, is the MR enhanced image, and UNet is the UNet network.

[0009] As a preferred solution of the present invention, the method for constructing the loss function for training the first segmentation model includes: Construct the prediction loss of the first segmentation model as: ; In the formula, is the prediction loss, is the vascular tissue mask output by the i-th MR image sample in the training dataset through the first segmentation model, is the true value of the vascular tissue mask of the i-th MR image sample in the training dataset, and m is the total number of MR image samples in the training dataset; Construct the guiding loss of the second segmentation model for the first segmentation model as: ; ; In the formula, is the guidance loss, is the vascular tissue mask output by the first segmentation model for the i-th MR image sample in the training dataset, is the vascular tissue mask output by the second segmentation model for the MR enhanced image corresponding to the i-th MR image sample in the training dataset, is the guidance adjustment coefficient, with high credibility and large guidance weight, is the ground truth of the vascular tissue mask of the MR enhanced image corresponding to the i-th MR image sample in the training dataset, , and are all L2 norm formulas; Combining the prediction loss and the guidance loss to form the loss function for training the first segmentation model as: ; In the formula, is the loss function for training the first segmentation model.

[0010] As a preferred solution of the present invention, the training method of the first segmentation model includes: Dividing the training dataset into a training set and a test set; Based on the loss function, training the UNet network on the training set to obtain the first segmentation model; Based on the evaluation metrics, evaluating the performance of the first segmentation model on the test set.

[0011] In a preferred solution of the present invention, the annotation data of the vascular tissue mask in the MR enhanced image and the annotation data of the vascular tissue mask in the MR image are both obtained by manual annotation.

[0012] As a preferred solution of the present invention, the evaluation metrics include Dice metric, IoU metric, and MAE metric.

[0013] As a preferred solution of the present invention, the present invention provides a vascular tissue segmentation system based on MR images, which is applied to a vascular tissue segmentation method based on MR images. The system includes: A data acquisition unit for acquiring MR images containing vascular tissue; A data enhancement unit for enhancing the vascular information in the MR image using the Canny operator to obtain an MR enhanced image; A model construction unit is configured to use a UNet network to construct, based on MR images and MR enhanced images, a first segmentation model for segmenting vascular tissues on MR images and a second segmentation model for segmenting vascular tissues on MR enhanced images respectively; use the second segmentation model to guide the training of the first segmentation model, and use the output result of the trained first segmentation model as the segmentation result of vascular tissues in the MR image. A segmentation output unit is configured to use the trained first segmentation model to obtain the segmentation result of vascular tissues in the MR image.

[0014] As a preferred embodiment of the present invention, there is provided a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method for segmenting vascular tissues based on MR images as described above.

[0015] The present invention has the following beneficial effects compared with the prior art: The present invention constructs a first segmentation model for segmenting vascular tissues on MR images and a second segmentation model for segmenting vascular tissues on MR enhanced images respectively, and uses the second segmentation model to guide the training of the first segmentation model, which can guide the segmentation result on the low-resolution image to learn from the segmentation result on the high-resolution image, so as to directly obtain a high-precision segmentation result on the low-resolution image without increasing the data processing process and model complexity in the segmentation process, thereby improving the efficiency of accurate segmentation of vascular tissues. Description of the Drawings

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary, and for those of ordinary skill in the art, other implementation drawings can be obtained according to the provided drawings without creative efforts.

[0017] Figure 1 It is a flowchart of the method for segmenting vascular tissues based on MR images provided by an embodiment of the present invention. Figure 2 It is a block diagram of the system for segmenting vascular tissues based on MR images provided by an embodiment of the present invention. Detailed Embodiments

[0018] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] As Figure 1 shown, the present invention provides a method for segmenting vascular tissues based on MR images, including the following steps: Obtain an MR image containing vascular tissues; Use the Canny operator to enhance the vascular information in the MR image to obtain an MR enhanced image; Use the UNet network to respectively construct a first segmentation model for segmenting vascular tissues on the MR image and a second segmentation model for segmenting vascular tissues on the MR enhanced image based on the MR image and the MR enhanced image; Use the second segmentation model to guide the training of the first segmentation model, and use the output result of the trained first segmentation model as the segmentation result of vascular tissues in the MR image.

[0020] The present invention first uses the Canny operator to perform edge detection on the MR image, thereby obtaining an edge detection image in the MR image. The linear attribute characteristics of the vascular tissues themselves will be detected in the edge detection, and the overall contour of the vascular tissues will be obtained. The edge detection image containing the overall contour of the vascular tissues is superimposed on the original MR image, thereby realizing pixel enhancement of the vascular tissues in the original MR image and obtaining an MR enhanced image. Compared with the original MR image, the resolution of the vascular tissues is improved.

[0021] The present invention constructs a first segmentation model for segmenting vascular tissues on the MR image to obtain the vascular segmentation result on the MR image, and constructs a second segmentation model for segmenting vascular tissues on the MR enhanced image to obtain the vascular segmentation result on the MR image. Among them, the first segmentation model segments vascular tissues on the MR image with low resolution, and the second segmentation model segments blood vessels on the MR image with high resolution. It can be seen that the first segmentation model is a low-precision segmentation compared with the second segmentation model, and the second segmentation model is a high-precision segmentation. However, the first segmentation model belongs to an efficient segmentation process, and the second segmentation model needs to perform image enhancement first, and the overall belongs to an inefficient segmentation process.

[0022] The present invention uses a second segmentation model to train and guide the first segmentation model, such that during the training process, the output of the first segmentation model approaches the output of the second segmentation model, that is, low-precision segmentation approaches high-precision segmentation. Thus, after the training is completed, the first segmentation model can obtain high-precision segmentation results on low-resolution MR images, and ultimately the first segmentation model combines high precision and high efficiency.

[0023] In order to enable the second segmentation model to train and guide the first segmentation model, the present invention additionally adds a guiding loss part to the prediction loss (i.e., L mse , which quantifies the loss between the output of the first segmentation model and the ground truth) of the traditional loss function of the first segmentation model. The guiding loss measures the degree of difference between the output of the first segmentation model and the output of the second segmentation model. Training with this as the loss can ensure that the degree of difference between the output of the first segmentation model and the output of the second segmentation model after the training is completed is minimized, achieving high-precision segmentation of the first segmentation model on low-resolution MR images.

[0024] Furthermore, the present invention also adds a guiding adjustment coefficient to the guiding loss part to control and adjust the training and guiding progress of the second segmentation model on the first segmentation model. The guiding adjustment coefficient is related to the precision performance of the second segmentation model. When the precision performance of the second segmentation model is higher, it indicates that the credibility of the second segmentation model is higher, and the training and guiding effect with this second segmentation model is better, so a large guiding adjustment coefficient is given, making the weight of the guiding loss part in the total loss function larger, and placing the training focus of the first segmentation model on learning from the second segmentation model, so that the first segmentation model obtains high-precision performance. Correspondingly, when the precision performance of the second segmentation model is low, it indicates that the credibility of the second segmentation model is low, and the training and guiding effect with this second segmentation model is poor, so a small guiding adjustment coefficient is given, making the weight of the guiding loss part in the total loss function smaller, and placing the training focus of the first segmentation model on maintaining the current minimum prediction loss, so that the first segmentation model maintains the current best precision performance, that is, the highest precision performance on MR images, and avoids being wrongly guided in training, resulting in the reduction or loss of the existing precision (basic disk).

[0025] Therefore, the present invention adds a guiding adjustment coefficient to the guiding loss part to achieve adaptive regulation of the guiding process and ensure effective guidance.

[0026] The present invention first performs edge detection on the MR image using the Canny operator, thereby obtaining the edge detection image of the MR image. The linear attribute characteristics of the vascular tissue itself will be detected during edge detection, obtaining the overall contour of the vascular tissue. The edge detection image containing the overall contour of the vascular tissue is superimposed on the original MR image, thereby realizing pixel enhancement of the vascular tissue in the original MR image and obtaining the MR enhanced image. Compared with the original MR image, the resolution of the vascular tissue is improved, specifically as follows: The method for enhancing the MR image includes: Performing edge detection on the MR image using the Canny operator to obtain the edge detection image of the MR image; Superimposing and fusing the edge detection image with the MR image to obtain the MR enhanced image.

[0027] The present invention constructs a first segmentation model for segmenting vascular tissue on the MR image to obtain the vascular segmentation result on the MR image, specifically as follows: The construction method of the first segmentation model includes: Taking the UNet network as the network structure of the first segmentation model; Taking the MR image as the input of the UNet network and the mask annotation data of the vascular tissue in the MR image as the output of the UNet network to construct the first segmentation model; The first segmentation model is: ; In the formula, is the mask of the vascular tissue of the MR image, is the MR image, and UNet is the UNet network.

[0028] The present invention constructs a second segmentation model for segmenting vascular tissue on the MR enhanced image to obtain the vascular segmentation result on the MR image, specifically as follows: The construction method of the second segmentation model includes: Taking the UNet network as the network structure of the second segmentation model; Taking the MR enhanced image as the input of the UNet network and the mask annotation data of the vascular tissue in the MR enhanced image as the output of the UNet network to construct the second segmentation model; The second segmentation model is: ; In the formula, is the mask of the vascular tissue of the MR enhanced image, is the MR enhanced image, and UNet is the UNet network.

[0029] The present invention uses a second segmentation model to train and guide the first segmentation model, so that during the training process, the output of the first segmentation model approaches the output of the second segmentation model, that is, the low-precision segmentation approaches the high-precision segmentation. Thus, after the training is completed, the first segmentation model can obtain a high-precision segmentation result on a low-resolution MR image, and finally the first segmentation model has both high precision and high efficiency. In order to achieve the training and guidance of the first segmentation model by the second segmentation model, the present invention adds an additional guidance loss part to the prediction loss (i.e., L mse , quantifying the loss between the output of the first segmentation model and the ground truth) of the traditional loss function of the first segmentation model, as follows: The method for constructing the loss function for training the first segmentation model includes: Construct the prediction loss of the first segmentation model as: ; In the formula, is the prediction loss, is the vascular tissue mask output by the first segmentation model for the i-th MR image sample in the training dataset, is the ground truth of the vascular tissue mask of the i-th MR image sample in the training dataset, and m is the total number of MR image samples in the training dataset; Construct the guidance loss of the second segmentation model for the first segmentation model as: ; ; In the formula, is the guidance loss, is the vascular tissue mask output by the first segmentation model for the i-th MR image sample in the training dataset, is the vascular tissue mask output by the second segmentation model for the MR enhanced image corresponding to the i-th MR image sample in the training dataset, is the guidance adjustment coefficient, is the ground truth of the vascular tissue mask of the MR enhanced image corresponding to the i-th MR image sample in the training dataset, , and are all L2 norm formulas; Combine the prediction loss and the guidance loss to form the loss function for training the first segmentation model as: ; In the formula, is the loss function for training the first segmentation model.

[0030] The present invention also adds a guidance adjustment coefficient to the guidance loss part, which is used to control and adjust the training guidance progress of the second segmentation model for the first segmentation model. The guidance adjustment coefficient is related to the accuracy performance of the second segmentation model. The higher the accuracy performance of the second segmentation model, the higher the credibility of the second segmentation model, and the better the training guidance effect with this second segmentation model. Then a large guidance adjustment coefficient is given, making the weight of the guidance loss part in the total loss function larger, and placing the training focus of the first segmentation model on learning from the second segmentation model, so that the first segmentation model can obtain high accuracy performance.

[0031] Correspondingly, when the accuracy performance of the second segmentation model is low, it indicates that the credibility of the second segmentation model is low, and the training guidance effect with this second segmentation model is not good. Then a small guidance adjustment coefficient is given, making the weight of the guidance loss part in the total loss function smaller, and placing the training focus of the first segmentation model on maintaining the current minimum prediction loss, so that the first segmentation model can maintain its current best accuracy performance, that is, the highest accuracy performance on MR images, and avoid being wrongly guided in training, resulting in the reduction or loss of the existing accuracy (basic disk).

[0032] Therefore, the present invention adds a guidance adjustment coefficient to the guidance loss part to realize the adaptive regulation of the guidance process and ensure effective guidance.

[0033] The training method of the first segmentation model includes: Dividing the training data set into a training set and a test set; Based on the loss function, training the UNet network on the training set to obtain the first segmentation model; Based on the evaluation metrics, evaluating the performance of the first segmentation model on the test set.

[0034] The vascular tissue mask annotation data in the MR enhanced image and the vascular tissue mask annotation data in the MR image are both obtained by manual annotation.

[0035] The evaluation metrics include Dice metric, IoU metric, and MAE metric.

[0036] As Figure 2 shown, the present invention provides a vascular tissue segmentation system based on MR images, which is applied to a vascular tissue segmentation method based on MR images. The system includes: A data acquisition unit for acquiring MR images containing vascular tissue; A data enhancement unit for enhancing the vascular information in the MR image using the Canny operator to obtain an MR enhanced image; A model construction unit, which is used to construct, based on an MR image and an MR enhanced image by using a UNet network, a first segmentation model for segmenting vascular tissue on the MR image and a second segmentation model for segmenting vascular tissue on the MR enhanced image respectively; use the second segmentation model to guide the training of the first segmentation model, and use the output result of the trained first segmentation model as the segmentation result of vascular tissue in the MR image; A segmentation output unit, which is used to obtain the segmentation result of vascular tissue in the MR image by using the trained first segmentation model.

[0037] The present invention provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the vascular tissue segmentation method based on an MR image is implemented.

[0038] The present invention constructs a first segmentation model for segmenting vascular tissue on an MR image and a second segmentation model for segmenting vascular tissue on an MR enhanced image respectively, and uses the second segmentation model to guide the training of the first segmentation model, which can guide the segmentation result on a low-resolution image to learn from the segmentation result on a high-resolution image, so as to directly obtain a high-precision segmentation result on the low-resolution image, and there is no need to increase the data processing process and model complexity in the segmentation process, thus improving the efficiency of accurate segmentation of vascular tissue.

[0039] The above embodiments are only exemplary embodiments of the present application and are not used to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art can make various modifications or equivalent replacements within the essence and protection scope of the present application, and such modifications or equivalent replacements should also be regarded as falling within the protection scope of the present application.

Claims

1. A vascular tissue segmentation method based on MR images, characterized in that: The following steps are involved: Acquiring MR images containing vascular tissue; The Canny operator is used to enhance the vascular information in the MR image to obtain an MR enhanced image; Using a UNet network, based on the MR image and the MR enhanced image, respectively construct a first segmentation model for segmenting vascular tissue on the MR image and a second segmentation model for segmenting vascular tissue on the MR enhanced image; The second segmentation model is used to guide the first segmentation model for training, and the output result of the first segmentation model after training is used as the vascular tissue segmentation result in the MR image.

2. The vascular tissue segmentation method based on MR images according to claim 1, characterized in that: The MR image enhancement method comprises: The Canny operator is used to perform edge detection on the MR image to obtain an edge detection image of the MR image; The edge detection image is superimposed and fused with the MR image to obtain the MR enhanced image.

3. The vascular tissue segmentation method based on MR images according to claim 2, characterized in that: The method for constructing the first segmentation model includes: Use the UNet network as the network structure of the first segmentation model; Using the MR image as the input of the UNet network and the vascular tissue mask annotation data in the MR image as the output of the UNet network to construct the first segmentation model; The first segmentation model is: ； In the formula, is the vascular tissue mask of the MR image, is the MR image, and UNet is the UNet network.

4. The vascular tissue segmentation method based on MR images according to claim 3, characterized in that: The method for constructing the second segmentation model includes: Use the UNet network as the network structure of the second segmentation model; Using the MR enhanced image as the input of the UNet network, and using the vascular tissue mask annotation data in the MR enhanced image as the output of the UNet network, to construct the second segmentation model; The second segmentation model is: ； In the formula, is the vascular tissue mask of the MR enhanced image, is the MR enhanced image, and UNet is the UNet network.

5. The vascular tissue segmentation method based on MR images according to claim 4, characterized in that: The loss function method for constructing the first segmentation model training includes: The prediction loss for building the first segmentation model is: ； In the formula, To predict the loss, is the vascular tissue mask output by the first segmentation model for the i-th MR image sample in the training data set, is the true value of the vascular tissue mask of the i-th MR image sample in the training data set, and m is the total number of MR image samples in the training data set; The guided loss of the second segmentation model for the first segmentation model is: ； ； In the formula, To guide the loss, is the vascular tissue mask output by the first segmentation model for the i-th MR image sample in the training data set, is the vascular tissue mask output by the second segmentation model for the MR enhanced image corresponding to the i-th MR image sample in the training data set, To guide the adjustment coefficient, it has high credibility and great guiding power. is the true value of the vascular tissue mask of the MR enhanced image corresponding to the i-th MR image sample in the training data set, , and All are L2 norm forms; The loss function for training the first segmentation model is formed by combining the prediction loss and the guidance loss: ； In the formula, is the loss function used for training the first segmentation model.

6. The vascular tissue segmentation method based on MR images according to claim 5, characterized in that: The training method of the first segmentation model includes: Divide the training data set into training set and test set; Based on the loss function, the UNet network is trained on the training set to obtain the first segmentation model; Based on the evaluation indicators, the performance of the first segmentation model is evaluated on the test set.

7. The vascular tissue segmentation method based on MR images according to claim 6, characterized in that: The vascular tissue mask annotation data in the MR enhanced image and the vascular tissue mask annotation data in the MR image are both obtained by manual annotation.

8. The vascular tissue segmentation method based on MR images according to claim 7, characterized in that: The evaluation indicators include Dice indicator, IoU indicator, and MAE indicator.

9. A vascular tissue segmentation system based on MR images, characterized in that: A method for segmenting vascular tissue based on MR images as described in any one of claims 1 to 8, the system comprising: A data acquisition unit, used for acquiring an MR image containing vascular tissue; A data enhancement unit, used for enhancing the blood vessel information in the MR image using the Canny operator to obtain an MR enhanced image; A model building unit is used to use a UNet network to build a first segmentation model for segmenting vascular tissue on the MR image and a second segmentation model for segmenting vascular tissue on the MR enhanced image based on the MR image and the MR enhanced image respectively; use the second segmentation model to guide the first segmentation model to be trained, and use the output result of the first segmentation model after the training as the vascular tissue segmentation result in the MR image; The segmentation output unit is used to obtain the segmentation result of blood vessel tissue in the MR image by using the first segmentation model after training.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when a processor executes the computer-executable instructions, the method according to any one of claims 1 to 8 is implemented.

Images