Medical image segmentation method, system, electronic device and storage medium

Abstract

The present invention provides a medical image segmentation method, system, electronic device, and storage medium. The medical image segmentation method obtains a first medical image to be segmented based on a first original medical image; obtains an image of a vascular region of interest based on the first medical image to be segmented; obtains a vascular centerline corresponding to the vascular region of interest based on the image of the vascular region of interest; and obtains several second medical images to be segmented of preset sizes by sliding a window along the vascular centerline with a point on the vascular centerline as the center; uses a trained neural network model to segment each second medical image to be segmented to obtain a preliminary segmented medical image; and splices several preliminary segmented medical images to obtain a final target medical image. The present invention can effectively improve detection efficiency and reduce the tedious operations of human-computer interaction. It is also highly versatile and realizes an end-to-end algorithm process, which can better and faster assist doctors in improving diagnostic accuracy.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a medical image segmentation method, system, electronic equipment and storage medium. Background Art

[0002] Vascular diseases, especially cardiovascular and cerebrovascular diseases, have become a major threat to human life. Among cerebrovascular diseases, intracranial aneurysms are third in incidence, after cerebral thrombosis and hypertensive intracerebral hemorrhage, posing a serious threat to human health. A first intracranial aneurysm rupture can result in death in approximately 30% of patients. Even those who survive are highly likely to experience a second rupture, leading to a secondary hemorrhage. The mortality rate for a secondary hemorrhage ranges from 60% to 70%, with the high-risk rupture period often occurring within multiples of seven days. If a third rupture occurs, the mortality rate can reach over 90%. Therefore, diagnosis and treatment of intracranial aneurysms before they bleed is the best way to reduce mortality and disability.

[0003] Digital subtraction angiography (DSA) is the gold standard for diagnosing intracranial aneurysms. It can clearly determine the location, morphology, size, and number of intracranial aneurysms, and ultimately determine the surgical plan. However, the complex structure of cerebral blood vessels results in different characteristics in different locations, making it difficult for doctors to analyze the vascular structure of patients' medical images.

[0004] Current methods for segmenting intracranial aneurysms include: 1. Manual segmentation of aneurysms from medical images by experienced doctors or professionals; 2. Methods based on traditional model matching; 3. Methods based on images or variable models; 4. Segmentation methods based on deep learning.

[0005] The above segmentation methods have the following problems:

[0006] 1. The results of manual intracranial aneurysm segmentation methods are not only highly variable but also require a lot of time and effort.

[0007] 2. Due to the different characteristics of different parts of intracranial blood vessels, the segmentation accuracy based on the model matching method is low.

[0008] 3. Methods based on variable models require user interaction to complete segmentation, have poor robustness, and have low segmentation accuracy.

[0009] 4. The deep learning segmentation method has achieved certain accuracy improvements compared to traditional methods, but the cascade network model used is complex and cumbersome, not efficient enough in prediction, and is limited by hardware configuration. Summary of the Invention

[0010] The purpose of the present invention is to provide a medical image segmentation method, system, electronic device and storage medium, which can not only effectively reduce the tedious operations of human-computer interaction, but also effectively improve the image segmentation accuracy and better assist doctors in improving the accuracy of targeting.

[0011] To achieve the above object, the present invention provides a medical image segmentation method, comprising:

[0012] According to the first original medical image, using a first preset interpolation method, the first original medical image is transformed to a target size to obtain a first medical image to be segmented;

[0013] performing a first segmentation on the first medical image to be segmented using an adaptive threshold algorithm, comparing the volume values ​​of each connected domain in the medical image obtained by the first segmentation, and taking the connected domain with the largest volume value as the trunk region of the region of interest image to obtain a vascular region of interest image;

[0014] A target path method is used to obtain a blood vessel centerline corresponding to the blood vessel region of interest; wherein the target path method includes: obtaining position coordinates of a starting point and at least one ending point based on the image of the blood vessel region of interest, calculating a target path between the starting point and the ending point, and using the target path as the blood vessel centerline corresponding to the blood vessel region of interest;

[0015] Taking the point on the centerline of the blood vessel as the center and sliding the window along the centerline of the blood vessel at a preset pixel distance, to obtain a plurality of second medical images to be segmented of preset sizes;

[0016] A neural network model is used to perform a second segmentation on each of the second medical images to be segmented, and several medical images obtained by the second segmentation are spliced. A second preset interpolation method is used to transform the several spliced ​​medical images to the same size as the first original medical image to obtain a final target medical image.

[0017] Optionally, the method of using the connected domain with the maximum volume value as the trunk region of the region of interest image to obtain the image of the blood vessel region of interest includes:

[0018] According to the maximum volume value and the preset threshold coefficient, a preset volume threshold is obtained; all connected domains are traversed, and all connected domains with volumes greater than the preset volume threshold are retained to obtain the image of the blood vessel region of interest.

[0019] Optionally, before acquiring the blood vessel centerline corresponding to the blood vessel region of interest using the first blood vessel centerline acquisition method, the method further includes:

[0020] For each connected domain of the image of the vascular region of interest, it is determined whether the connected domain is a trunk region. If so, the target path method is used to obtain the vascular centerline corresponding to the vascular region of interest; if not, a 3D skeleton extraction algorithm is used to extract the skeleton of the connected domain, and the result of the skeleton extraction is used as the vascular centerline corresponding to the vascular region of interest.

[0021] Optionally, acquiring a starting point according to the image of the blood vessel region of interest includes:

[0022] S311: Acquire N first continuous slices starting from a starting slice of the image of the vascular region of interest toward an ending slice, where N is greater than or equal to 3, and the starting slice is closer to a starting scanning position of the patient when acquiring the first original medical image than the ending slice;

[0023] S312: Determine whether the ellipticity of the connected domains corresponding to the N first continuous slices is greater than a first preset threshold and whether the deviation of the centroid coordinates of the connected domains corresponding to the first continuous slices on the X-axis and the Y-axis is less than a second preset threshold. If so, obtain the starting point based on the centroid of the connected domain corresponding to the first continuous slices. If not, execute step S313.

[0024] S313: Starting from the adjacent slice second closest to the starting slice in the first continuous slices toward the ending slice, obtain N second continuous slices, and use the second continuous slices as the first continuous slices, and execute step S312.

[0025] Optionally, acquiring a plurality of end points according to the image of the blood vessel region of interest includes:

[0026] S321: performing an opening operation on the image of the blood vessel region of interest, and then subtracting a result of the opening operation from the image of the region of interest to obtain a plurality of end regions of the image of the blood vessel region of interest;

[0027] S322: Performing a connected domain analysis on each of the terminal regions to obtain the centroid of the connected domain corresponding to each of the terminal regions, and using the centroid of the connected domain as a termination point, thereby obtaining a plurality of termination points of the image of the vascular region of interest.

[0028] Optionally, before segmenting each of the second medical images to be segmented using the neural network model, the neural network model is trained by the following steps:

[0029] Step A: Acquire an original sample, wherein the original sample includes a second original medical image and a labeled image corresponding to the second original medical image, wherein the labeled image is a medical image in which a target tissue is annotated;

[0030] Step B: Expanding the original sample to obtain a training sample and a verification sample, wherein the training sample includes an expanded medical training image and a label image corresponding to the expanded medical training image; the verification sample includes an expanded medical verification image and a label image corresponding to the expanded medical verification image;

[0031] Step C: setting initial values ​​of model parameters of the neural network model; and

[0032] The pre-built neural network model is trained according to the training samples, the verification samples and the initial values ​​of the model parameters of the neural network model until a first preset training end condition is met.

[0033] Optionally, the expanding the original sample to obtain a training sample and a validation sample includes:

[0034] Preprocessing the second original medical image and the label image corresponding to the second original medical image respectively to obtain a preprocessed second original medical image and a preprocessed label image;

[0035] acquiring, according to the preprocessed second original medical image, a blood vessel region image of interest corresponding to the second original medical image and a blood vessel centerline of the blood vessel region image of interest corresponding to the second original medical image;

[0036] Taking the blood vessel centerline of the image of the blood vessel region of interest corresponding to the second original medical image as the center point, sliding a window along the blood vessel centerline and at the preset pixel distance to obtain a plurality of medical images of preset sizes, thereby obtaining an expanded sample image;

[0037] Taking the center point of the plurality of sample images of the preset size as a first center point, obtaining a plurality of second center points corresponding to the first center point on a preprocessed label image corresponding to the preprocessed second original medical image, and obtaining a plurality of medical images of the preset size with the second center point as the center to obtain an expanded label image;

[0038] According to a preset proportional distribution relationship, the expanded sample images are randomly divided into two groups, one of which is used as medical training images and the other is used as medical verification images; the medical training images and the expanded label images corresponding to the medical training images are combined into a group as training data to obtain the training samples; the medical verification images and the expanded label images corresponding to the medical verification images are combined into a group as verification data to obtain the verification samples.

[0039] Optionally, the first preset training end condition is that the number of training times of the neural network model is greater than or equal to a first preset number of iterations;

[0040] In step C, the pre-built neural network model is trained according to the training sample, the verification sample and the initial values ​​of the model parameters of the neural network model until a first preset training end condition is met, including:

[0041] Step C1: using the training sample as input of the neural network model, and obtaining a target tissue prediction result of the medical training image according to the initial values ​​of the model parameters of the neural network model;

[0042] Step C2: calculating a loss function value based on the target tissue prediction result of the medical training image and the expanded label image corresponding to the medical training image; and adjusting the model parameters of the neural network model based on the loss function value and a preliminary training preset error value until the number of training times of the neural network model is greater than or equal to a second preset number of iterations, thereby obtaining a preliminary trained neural network model;

[0043] Step C3: Inputting the verification sample into the preliminarily trained neural network model, obtaining a target tissue prediction result of the medical verification image, calculating a loss function value based on the target tissue prediction result of the medical verification image and the expanded label image corresponding to the medical verification image; and recording a verification result, the verification result including the loss function value and the corresponding model parameters of the preliminarily trained neural network model;

[0044] Step C4: determining whether the number of training times of the neural network model is greater than or equal to the first preset number of iterations; if so, executing step C5; if not, adjusting the model parameters of the neural network model, and updating the initial values ​​of the model parameters of the neural network model to the adjusted model parameters, and returning to executing step C1;

[0045] Step C5: According to the record of the verification result, the neural network model is obtained and the training is completed; wherein, the model parameters of the neural network model are the model parameters of the preliminarily trained neural network model corresponding to the minimum value of the loss function.

[0046] In order to achieve the above object, the present invention further provides a medical image segmentation system, the medical image segmentation system comprising:

[0047] A first medical image to be segmented acquiring unit is configured to transform the first original medical image to a target size using a first preset interpolation method according to the first original medical image to acquire the first medical image to be segmented;

[0048] a vessel region of interest acquisition unit configured to perform a first segmentation on the first medical image to be segmented using an adaptive threshold algorithm, compare the volume values ​​of each connected domain in the medical image obtained by the first segmentation, and use the connected domain with the largest volume value as the trunk region of the region of interest image to acquire the vessel region of interest image;

[0049] a blood vessel centerline acquisition unit configured to acquire the blood vessel centerline corresponding to the blood vessel region of interest using a target path method; wherein the target path method includes: acquiring position coordinates of a starting point and at least one ending point based on the image of the blood vessel region of interest, calculating a target path between the starting point and the ending point, and using the target path as the blood vessel centerline corresponding to the blood vessel region of interest;

[0050] a second medical image to be segmented acquiring unit configured to perform a sliding window operation along the blood vessel centerline at a preset pixel distance with a point on the blood vessel centerline as the center, so as to acquire a plurality of second medical images to be segmented of preset sizes;

[0051] The target medical image acquisition unit is configured to use a neural network model to perform a second segmentation on each of the second medical images to be segmented, and to splice the multiple medical images obtained by the second segmentation, and to use a second preset interpolation method to transform the multiple spliced ​​medical images to the same size as the first original medical image to obtain a final segmented medical image.

[0052] To achieve the above object, the present invention further provides an electronic device, comprising a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the medical image segmentation method described above is implemented.

[0053] To achieve the above-mentioned object, the present invention further provides a readable storage medium, wherein the readable storage medium stores a computer program, and when the computer program is executed by a processor, the medical image segmentation method described above is implemented.

[0054] Compared with the prior art, the medical image segmentation method, system, electronic device, and storage medium provided by the present invention have the following beneficial effects:

[0055] The medical image segmentation method provided by the present invention first transforms a first original medical image to a target size using a first preset interpolation method based on a first original medical image to obtain a first medical image to be segmented; performs a first segmentation on the first medical image to be segmented using an adaptive threshold algorithm, compares the volume values ​​of each connected domain in the medical image obtained by the first segmentation, and uses the connected domain with the largest volume value as the trunk region of the region of interest image to obtain a blood vessel region image of interest; then uses a target path method to obtain the blood vessel centerline corresponding to the blood vessel region of interest; wherein the target path method includes: obtaining a starting point and a target path based on the blood vessel region image of interest; The position coordinates of one less end point are obtained, and the target path between the starting point and the end point is calculated, and the target path is used as the vascular centerline corresponding to the vascular region of interest; then, a window is slid along the vascular centerline according to a preset pixel distance with the point on the vascular centerline as the center to obtain a plurality of second medical images to be segmented of a preset size; finally, a neural network model is used to perform a second segmentation on each of the second medical images to be segmented, and the plurality of medical images obtained by the second segmentation are spliced, and a second preset interpolation method is used to transform the plurality of spliced ​​medical images to the same size as the first original medical image to obtain the final target medical image. Thus, the medical image segmentation method provided by the present invention automatically calculates the vascular centerline based on the acquisition of the vascular region of interest (such as the intracranial artery), and crops the region of interest on the image with the point on the centerline as the center to perform network prediction of the target tissue (such as the intracranial artery and intracranial aneurysm), which can not only effectively improve the detection efficiency, but also greatly improve the overall segmentation accuracy and reduce the tedious operation of human-computer interaction. In addition, the medical image segmentation method provided by the present invention has strong versatility and realizes an end-to-end algorithm process, which can better and faster assist doctors in improving diagnostic accuracy.

[0056] 2. The medical image segmentation method provided by the present invention first obtains an original training sample and an original verification sample, wherein the original training sample includes an original medical training image and a first label image corresponding to the original medical training image, and the first label image is a first medical image with a target tissue marked; the original verification sample includes an original medical verification image and a second label image corresponding to the original medical verification image, and the second label image is a second medical image with a target tissue marked; and the original training sample is expanded to obtain an extended training sample, wherein the extended training sample includes an extended medical training image and a first label image corresponding to the extended medical training image; the original verification sample is expanded to obtain an extended verification sample, wherein the extended verification sample includes an extended medical verification image and a second label image corresponding to the extended medical verification image; then the initial values ​​of the model parameters of the neural network model are set; and the pre-built neural network model is trained according to the extended training sample, the extended verification sample and the initial values ​​of the model parameters of the neural network model until the first preset training end condition is met. Therefore, the medical image segmentation method provided by the present invention can not only solve the problem of insufficient samples by obtaining training samples and verification samples and expanding the training samples and verification samples, but also can train the pre-built neural network model according to the expanded training samples, the expanded verification samples and the initial values ​​of the model parameters of the neural network model, thereby ensuring the generalization and robustness of the neural network model and improving the segmentation precision and accuracy of medical images.

[0057] 3. Since the medical image segmentation system, electronic device and storage medium provided by the present invention and the medical image segmentation method provided by the present invention belong to the same inventive concept, they have at least the same beneficial effects as the above-mentioned medical image segmentation method and will not be described in detail here. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 is a flowchart of a medical image segmentation method in one embodiment of the present invention;

[0059] Figure 2 is a schematic diagram of an intracranial artery image in a specific example of the present invention;

[0060] Figure 3 for Figure 1 A specific flow chart of obtaining a starting point according to the image of the vascular region of interest provided in one embodiment of step S300;

[0061] Figure 4 for Figure 1 A specific flow chart of obtaining the end point according to the image of the blood vessel of interest provided in one embodiment of step S300;

[0062] Figure 5 A schematic diagram of the overall training process of a neural network model provided by one embodiment of the present invention;

[0063] Figure 6 This is a flow chart of expanding original samples to obtain training samples and verification samples according to one embodiment of the present invention;

[0064] Figure 7 A schematic diagram of the training details of a neural network model provided in one embodiment of the present invention;

[0065] Figure 8 is a schematic diagram of an intracranial aneurysm image obtained by segmentation in a specific example of the present invention;

[0066] Figure 9 A schematic block diagram of a medical image segmentation system provided in one embodiment of the present invention;

[0067] Figure 10 Schematic diagram of the block structure of an electronic device in one embodiment of the present invention;

[0068] The accompanying drawings are numerals as follows:

[0069] 100-intracranial artery, 200-intracranial aneurysm;

[0070] 310 - first medical image to be segmented acquisition unit, 320 - interested blood vessel region acquisition unit, 330 - blood vessel centerline acquisition unit, 340 - second medical image to be segmented acquisition unit, 350 - target medical image acquisition unit, 360 - neural network model training unit, 361 - training sample and verification sample acquisition module, 362 - sample expansion module, 363 - network model training module;

[0071] 410 - processor, 420 - memory, 430 - communication interface, 440 - communication bus. DETAILED DESCRIPTION

[0072] The following is a further detailed description of the medical image segmentation method, system, electronic device and storage medium proposed in the present invention in conjunction with the accompanying drawings and specific embodiments. According to the following description, the advantages and features of the present invention will become clearer. It should be noted that the drawings are in a very simplified form and are not in precise proportions, which are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention. In order to make the purposes, features and advantages of the present invention more obvious and easy to understand, please refer to the drawings. It should be noted that the structures, proportions, sizes, etc. illustrated in the drawings of this specification are only used to match the contents disclosed in the specification for people familiar with this technology to understand and read, and are not used to limit the conditions for the implementation of the present invention. Any modification of the structure, change in the proportional relationship or adjustment of the size should still fall within the scope of the technical content disclosed in the present invention when the effects and purposes that can be achieved are the same or similar to those that can be produced by the present invention.

[0073] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply the existence of any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or device comprising the element.

[0074] The core idea of ​​the present invention is to provide a medical image segmentation method, system, electronic device and storage medium, which can not only effectively reduce the tedious operations of human-computer interaction, but also effectively improve the image segmentation accuracy. It should be noted that although the present invention is explained by taking the segmentation of intracranial blood vessels and intracranial aneurysms from three-dimensional DSA (Digital Subtraction Angiography) volume data images (that is, the input data is a three-dimensional DSA volume data image of the patient's brain) as an example, as those skilled in the art can understand, the present invention can also segment the desired tissues and organs from other medical images, such as the aorta and aortic aneurysm, and the present invention is not limited to this.

[0075] It should be noted that the electronic device of the embodiment of the present invention can be a personal computer, a mobile terminal, etc., and the mobile terminal can be a hardware device with various operating systems such as a mobile phone and a tablet computer. In addition, it should be noted that although the present invention takes the intracranial artery region as the vascular region of interest and illustrates the intracranial aneurysm in the intracranial artery region as an example, this does not constitute a limitation of the present invention. As those skilled in the art will understand, the medical image segmentation method provided by the present invention can also be used to segment hemangiomas (or other vascular diseases) in other parts. In addition, it should be noted that all calculation processes in the present invention are performed in the world coordinate system. Specifically, the coordinates of each pixel point on the first / second original medical image in the image coordinate system can be converted into coordinates in the world coordinate system according to the mapping relationship between the pre-acquired image coordinate system and the world coordinate system, and then the calculation can be performed. The mapping relationship between the image coordinate system and the world coordinate system can be obtained by the parameters of the acquisition device of the acquired original medical image.

[0076] To realize the above idea, one embodiment of the present invention provides a medical image segmentation method. Specifically, please refer to Figure 1 , which schematically shows a flow chart of a medical image segmentation method provided by one embodiment of the present invention, as shown in FIG. Figure 1 As shown, the medical image segmentation method includes the following steps:

[0077] S100: According to a first original medical image, using a first preset interpolation method, transforming the first original medical image to a target size to obtain a first medical image to be segmented;

[0078] S200: performing a first segmentation on the first medical image to be segmented using an adaptive threshold algorithm, and comparing the volume values ​​of each connected domain in the medical image obtained by the first segmentation, and taking the connected domain with the largest volume value as the trunk region of the region of interest image to obtain a blood vessel region of interest image;

[0079] S300: Acquire a blood vessel centerline corresponding to the blood vessel region of interest using a target path method; wherein the target path method includes: acquiring position coordinates of a starting point and at least one ending point based on the image of the blood vessel region of interest, calculating a target path between the starting point and the ending point, and using the target path as the blood vessel centerline corresponding to the blood vessel region of interest;

[0080] S400: Taking the point on the centerline of the blood vessel as the center, sliding a window along the centerline of the blood vessel at a preset pixel distance to obtain a plurality of second medical images to be segmented of preset sizes;

[0081] S500: Using a neural network model to perform a second segmentation on each of the second medical images to be segmented, and splicing the multiple medical images obtained by the second segmentation, and using a second preset interpolation method to transform the multiple spliced ​​medical images to the same size as the first original medical image to obtain a final target medical image.

[0082] Therefore, the medical image segmentation method provided by the present invention automatically calculates the vascular centerline based on the acquisition of the vascular region of interest (such as the intracranial artery). It then crops the region of interest from the image with the point on the centerline as the center to perform network prediction of the target tissue (such as the intracranial artery and intracranial aneurysm). This not only effectively improves detection efficiency, but also greatly improves the overall segmentation accuracy and reduces the tedious human-computer interaction operations. In addition, the medical image segmentation method provided by the present invention is highly versatile and implements an end-to-end algorithm process, which can better and more quickly assist doctors in improving diagnostic accuracy.

[0083] In the present invention, the medical image to be segmented can be a medical image of the patient's brain, or a medical image of other tissues and organs other than the patient's brain, such as the lungs, aorta, etc., and the present invention is not limited to this. The three-dimensional DSA volume data image of the medical image to be segmented can also be other medical images, such as CTA (computed tomography angiography) volume data (three-dimensional data) image. It should be noted that the size of the medical image to be segmented can be set according to the specific situation, and the present invention is not limited to this. For example, the size of the medical image to be segmented can be 512×512×510 pixels. In addition, it should be noted that the medical image to be segmented can be collected by an image acquisition device, such as CT, MRI and other imaging equipment, or can be collected through the Internet, or can be scanned by a scanning device, and the present invention also does not impose any restrictions on this.

[0084] It's important to note that, as those skilled in the art will appreciate, pixel distance defines the physical size of image pixels and ensures the accuracy of actual distance measurements. For example, taking a 2D image (and similarly for 3D images), if the pixel spacing along the X and Y axes is known to be 0.5 mm, then a 10-pixel line in the image will have a length of 5 mm. Similarly, knowing the width and height of the image pixels (e.g., 512×512) allows us to determine the actual size of the image: 512×0.5 mm = 256 mm. Unlike natural images, the true size (image size) of human body parts is crucial in medical imaging. However, because different scanners or acquisition protocols typically produce datasets with varying voxel spacing, which neural network models cannot understand, all medical images are resampled to a consistent pixel size. However, the inventors of the present application have found through extensive investigations and studies that the ratio between the pixel (voxel) distance and the actual physical distance of general three-dimensional DSA volume data fluctuates between 0.3 and 0.5. Therefore, in one embodiment, the target voxel spacing is defined as [0.4, 0.4, 0.4], and the first original medical image is transformed into the target size by a first preset interpolation method to obtain a first image to be segmented. Preferably, in step S100, the first preset difference method is a spline interpolation method. Spline interpolation will include more surrounding pixels in the calculation range, thereby improving the segmentation accuracy while retaining the details of the picture as much as possible. For specific details about spline interpolation, please refer to the prior art, which will not be repeated here. As those skilled in the art will understand, the present invention is not limited to the specific algorithm of the first preset difference method. In other embodiments, other interpolation algorithms may also be used, including but not limited to nearest neighbor interpolation and bilinear interpolation.

[0085] It can be seen that the medical image segmentation method provided by the present invention, in step S100, transforms the first original medical image to the target size by adopting the first preset interpolation method, which can lay the foundation for further segmentation using a neural network model.

[0086] Correspondingly, in step S500, a second preset interpolation method is used to transform the stitched segmented medical image to the same size as the first original medical image, thereby obtaining a final target medical image. With this configuration, the medical image segmentation method provided by the present invention can obtain a target medical image of the same size as the first original medical image, thereby maintaining image consistency.

[0087] To further improve segmentation accuracy, preferably, in one exemplary embodiment, before executing step S200, the medical image segmentation method further includes filtering the first medical image to be segmented to remove noise in the first medical image to be segmented. Thus, filtering the first medical image to be segmented (e.g., Gaussian filtering) effectively removes noise in the first medical image to be segmented, laying a good foundation for obtaining an accurate image of the vascular region of interest.

[0088] Correspondingly, the step S200 is: acquiring an image of a blood vessel region of interest based on the filtered first medical image to be segmented.

[0089] Specifically, as one preferred embodiment, in step S200, taking the connected domain with the maximum volume value as the trunk region of the region of interest image to obtain the vessel region of interest image includes:

[0090] According to the maximum volume value and the preset threshold coefficient, a preset volume threshold is obtained; all connected domains are traversed, and all connected domains with volumes greater than the preset volume threshold are retained to obtain the image of the blood vessel region of interest.

[0091] With such configuration, the medical image segmentation method provided by the present invention is based on the imaging principle of three-dimensional DSA images. The intracranial artery area appears highlighted on the image under the action of contrast agent. By adopting an adaptive threshold algorithm, the algorithm is not only simple and easy to use, but also has high computational efficiency and fast speed.

[0092] More specifically, in one embodiment, all pixel values ​​less than 0 in the first medical image to be segmented are first set to 0 to remove background interference, and the OTSU adaptive threshold algorithm (Otsu method) is used to segment the intracranial artery to obtain the preliminary segmentation result volume data of the intracranial artery (i.e., the medical image obtained by the first segmentation), wherein the intracranial artery area is 1 and other areas are 0. At this time, the preliminary segmentation result volume data is interfered by non-vascular areas and broken small blood vessels (intracranial aneurysms generally do not occur in small blood vessel areas). All connected domains in the preliminary segmentation result volume data are counted and their volumes are calculated respectively, and the connected domain with the largest volume is found. If the corresponding volume is maxV, it is determined that the connected domain is the main trunk area of ​​the intracranial artery; a preset threshold coefficient a is set. It can be understood that the value range of the preset threshold coefficient is between [0-1]. For example, in one embodiment, it is set to 0.2. All connected domains are traversed in a loop, and only connected domains with a volume greater than maxV*a are retained. In this way, non-vascular highlight areas and partially broken small blood vessel areas can be removed, and the image of the blood vessel area of ​​interest (such as the mask image of the intracranial artery volume data) can be obtained. Figure 2, which schematically shows the schematic diagram of the intracranial artery image obtained using the above method. Figure 2 As shown, by segmenting the medical image after filtering, a complete intracranial artery image can be obtained. In the intracranial artery image (such as the intracranial artery body data mask image), the pixel value of the intracranial artery area is 1, and the pixel value of other areas is 0.

[0093] It should be noted that, as those skilled in the art can understand, other image segmentation methods in the prior art, such as threshold segmentation method, neural network segmentation method based on deep learning, etc., can be used to segment the medical image. The present invention does not limit the segmentation method.

[0094] Therefore, by performing a first (i.e., preliminary) segmentation on the first medical image to be segmented (e.g., a three-dimensional DSA volume data image of the patient's brain) and obtaining an image of the vascular region of interest (e.g., a preliminary segmentation result of intracranial arteries and intracranial aneurysms), the tedious human-computer interaction operations can be effectively reduced, the segmentation accuracy can be improved, and the foundation can be laid for the subsequent acquisition of accurate images of the target tissue (e.g., intracranial arteries and intracranial aneurysms).

[0095] It is particularly important to note that, as those skilled in the art will appreciate, the medical image segmentation method provided by the present invention, in step S300, obtains a starting point and an ending point based on the image of the vascular region of interest, and then determines a target path between the starting point and the ending point based on the position coordinates of the starting point and the ending point, and uses the target path as the vascular centerline corresponding to the vascular region of interest. In this way, not only can the vascular centerline be accurately obtained, but the vascular centerline can also be extracted, laying the foundation for improving the efficiency of subsequently obtaining accurate target tissue images.

[0096] Preferably, in one exemplary embodiment, before acquiring the vascular centerline corresponding to the vascular region of interest using the first vascular centerline acquisition method, the method further includes:

[0097] For each connected domain of the image of the vascular region of interest, it is determined whether the connected domain is a trunk region. If so, the target path method is used to obtain the vascular centerline corresponding to the vascular region of interest; if not, a 3D skeleton extraction algorithm is used to extract the skeleton of the connected domain, and the result of the skeleton extraction is used as the vascular centerline corresponding to the vascular region of interest.

[0098] More specifically, it should be noted that the trunk region is the connected domain with the maximum volume value in the image of the blood vessel region of interest. Furthermore, the present invention does not limit the specific method of calculating the target path between the starting point and the end point in step S300, including but not limited to the A* algorithm or the Dijkstra algorithm. For specific details of the skeleton extraction algorithm, A* algorithm or Dijkstra of the connected domain, please refer to the relevant descriptions in the prior art, which will not be elaborated here. Therefore, the medical image segmentation method provided by the present invention adopts different centerline extraction algorithms according to the volume of the connected domain of the blood vessel region of interest, which can take into account the extraction efficiency of the blood vessel centerline without reducing the accuracy of the blood vessel centerline extraction (usually, taking hemangioma as an example, the probability of hemangioma being located in smaller blood vessel tissue is lower), thereby seeking a better balance between accuracy and efficiency.

[0099] Preferably, in one exemplary embodiment, see Figure 3 , which schematically shows a specific flow chart of obtaining the starting point according to the image of the blood vessel region of interest provided by this embodiment, from Figure 3 It can be seen that in step S00, obtaining the starting point according to the image of the blood vessel region of interest includes:

[0100] S311: Acquire N first continuous slices starting from a starting slice of the image of the vascular region of interest toward an ending slice, wherein N≥3, and the starting slice is closer to a starting scanning position of the patient when acquiring the first original medical image than the ending slice.

[0101] S312: Determine whether the ellipticity of the connected domains corresponding to the N first continuous slices is greater than a first preset threshold and whether the deviation of the centroid coordinates of the connected domains corresponding to the first continuous slices on the X-axis and the Y-axis is less than a second preset threshold. If so, obtain the starting point based on the centroid of the connected domain corresponding to the first continuous slices. If not, execute step S313.

[0102] S313: Starting from the adjacent slice second closest to the starting slice in the first continuous slices toward the ending slice, obtain N second continuous slices, and use the second continuous slices as the first continuous slices, and execute step S312.

[0103] More specifically, as one of the preferred implementations, N is an odd number. In step S312, obtaining the starting point based on the centroid of the connected domain corresponding to each of the first continuous slices includes taking the centroid of the connected domain corresponding to the middle slice of the first continuous slice as the starting point.

[0104] Preferably, in one exemplary embodiment, see Figure 4 , which schematically shows a specific flow chart of obtaining the end point according to the image of the blood vessel region of interest provided by this embodiment, Figure 4 It can be seen that in step S00, the acquisition of several end points based on the image of the blood vessel region of interest includes:

[0105] S321: performing an opening operation on the image of the blood vessel region of interest, and then subtracting a result obtained by the opening operation from the image of the region of interest to obtain a plurality of end regions of the image of the blood vessel region of interest.

[0106] S322: Performing a connected domain analysis on each of the terminal regions to obtain the centroid of the connected domain corresponding to each of the terminal regions, and using the centroid of the connected domain as a termination point, thereby obtaining a plurality of termination points of the image of the vascular region of interest.

[0107] Specifically, the extraction of intracranial artery and intracranial aneurysm images using a three-dimensional DSA volume data image of the patient's brain as the original image is described as an example. First, as a preferred embodiment, the starting point generally needs to be selected at the very beginning of the internal carotid artery. The original three-dimensional DSA image is also scanned upward from the patient's neck. Therefore, for the image of the vascular region of interest (mask volume data), a connected domain analysis is performed step by step starting from the 0th slice (along the Z-axis direction). When the ellipticity of the connected domains between the three adjacent slices (the value of N) is greater than 0.8 (the first preset threshold) and the deviation of the respective centroids on the X and Y axes is less than 3 (the second preset threshold), the centroid of the connected domain corresponding to the middle slice can be considered as the starting point. As can be understood by those skilled in the art, the setting of N to 3, the first preset threshold to 0.8, and the second preset threshold to 3 in this embodiment are merely descriptions of preferred embodiments, and are not limitations of the present invention. In other embodiments (such as for cardiac aorta and aortic aneurysm images), the settings should be reasonable according to actual conditions. Secondly, because intracranial aneurysms generally do not occur in the thin end of the artery, it is more reasonable to select the end point at the thin end of the artery. That is, if an aneurysm exists, it must be between the starting point and the end point of the intracranial artery. Specifically, an opening operation is first performed on the image of the vascular region of interest (such as the mask image of the intracranial artery volume data) (that is, the image of the vascular region of interest is first eroded and then dilated to remove the end area of ​​the blood vessel in the image), and then the result of the opening operation is subtracted from the image of the vascular region of interest (such as the mask image of the intracranial artery volume data) to obtain several end areas of the image of the vascular region of interest (such as the thin end area of ​​the intracranial artery). Connected domain analysis is performed on these areas, and the centroid of each connected domain is used as the end point, ultimately obtaining a series of end points. Finally, based on the starting point and multiple ending points on the image of the vascular region of interest (such as the mask image of the intracranial artery volume data), a shortest path iteration method (such as the A* algorithm, the Dijkstra algorithm, etc.) is used to finally obtain multiple vascular centerlines from the starting point to the ending point, which are reflected as a series of three-dimensional point coordinates [[x1, y1, z1], [x2, y2, z2], [x3, y3, z3], ...].

[0108] It should be noted that, as those skilled in the art will appreciate, while the above description uses the example of automatically obtaining the starting and ending points, the starting and ending points can be selected manually or by a computer according to a pre-set algorithm, and the present invention is not limited thereto. Furthermore, it should be noted that in other embodiments, existing methods for extracting vascular centerlines, such as those based on region growing and vascular centerline models, can also be employed, and the present invention is not limited thereto.

[0109] Preferably, in one exemplary embodiment, before segmenting each of the second medical images to be segmented using the neural network model in step S500, the method further includes training a pre-established neural network model to obtain the neural network model. As will be appreciated by those skilled in the art, the neural network model is pre-trained, and as long as the neural network model training is completed before step S500, the present invention does not limit the order of neural network model training and steps S100 to S400, nor does it limit the specific method of neural network model training. The following is merely a description of a preferred embodiment.

[0110] Specifically, see Figure 5 , which schematically shows a training flow diagram of a neural network model provided by one embodiment of the present invention. Figure 5 It can be seen that the following steps are included:

[0111] Step A: Acquire an original sample, where the original sample includes a second original medical image and a label image corresponding to the second original medical image, where the label image is a medical image in which a target tissue has been marked.

[0112] It is understandable that, as a preference, the label image can be obtained through the gold standard, but this is not a limitation of the present invention. In other embodiments, it can also be obtained manually or by a computer algorithm based on the second original medical image, but it should be understood that the label image should be reliable.

[0113] Step B: Expanding the original sample to obtain a training sample and a verification sample, wherein the training sample includes an expanded medical training image and a label image corresponding to the expanded medical training image; the verification sample includes an expanded medical verification image and a label image corresponding to the expanded medical verification image;

[0114] Step C: setting initial values ​​of model parameters of the neural network model; and

[0115] The pre-built neural network model is trained according to the training samples, the verification samples and the initial values ​​of the model parameters of the neural network model until a first preset training end condition is met.

[0116] As can be understood by those skilled in the art, as mentioned above, before executing step B, it is preferred that the pixels of the second original medical image and the label image corresponding to the second original medical image be resampled to the target resolution scale. The specific sampling method has been described in detail above and will not be repeated here. For example, the label image can be resampled to the target resolution scale using the nearest neighbor interpolation method.

[0117] Preferably, in one preferred embodiment, see Figure 6 , which schematically shows a flow chart of an embodiment of the present invention for expanding the original sample to obtain the training sample and the verification sample, Figure 6 It can be seen that in step B, the original sample is expanded to obtain a training sample and a verification sample, including:

[0118] Step B1: preprocessing the second original medical image and the label image corresponding to the second original medical image respectively to obtain a preprocessed second original medical image and a preprocessed label image.

[0119] Specifically, in order to improve the accuracy of the neural network model, before generating the training sample and the verification sample, the second original medical image may be preprocessed to remove noise in the image and improve the image quality of the training sample.

[0120] Step B2: Based on the preprocessed second original medical image, an image of a blood vessel region of interest corresponding to the second original medical image and a blood vessel centerline of the image of the blood vessel region of interest corresponding to the second original medical image are acquired.

[0121] It should be noted that the basic principle for obtaining the region of interest image and the vascular centerline of the vascular region of interest image corresponding to the second original medical image is similar to the method for obtaining the vascular region of interest image and its vascular centerline of the first original medical image described above. Please refer to the above for details and will not be repeated here. As those skilled in the art will understand, other methods may also be used to obtain the vascular region of interest image and its vascular centerline, and the present invention does not impose any limitation on this.

[0122] Step B3: Taking the blood vessel centerline of the image of the blood vessel region of interest corresponding to the second original medical image as the center point, sliding the window along the blood vessel centerline and according to the preset pixel distance to obtain several medical images of preset sizes to obtain an expanded sample image.

[0123] Specifically, a sliding window can be performed along the center line, with a distance of 64 pixels between two points on the center line as the center, to intercept a region of size [128, 128, 128] from the original image as a sample image. Thus, the medical image segmentation method provided by the present invention can generate multiple expanded sample images when the number of second original medical images is limited. This not only simplifies the algorithm but also significantly improves the training efficiency of the neural network model.

[0124] Step B4: Taking the center point of the expanded sample images of the preset sizes as the first center point, obtaining several second center points corresponding to the first center point on the preprocessed label image corresponding to the preprocessed second original medical image, and obtaining several medical images of the preset sizes with the second center point as the center to obtain the expanded label image.

[0125] Those skilled in the art should be able to understand that the expanded sample image and the expanded label image are one-to-one corresponding: including the position of the expanded sample image on the second original medical image and the position and size of the label image corresponding to the expanded label image on the second original medical image.

[0126] Step B5: According to a preset proportional distribution relationship, the expanded sample images are randomly divided into two groups, one of which is used as medical training images and the other is used as medical verification images; the medical training images and the expanded label images corresponding to the medical training images are combined into a group as training data to obtain the training samples; the medical verification images and the expanded label images corresponding to the medical verification images are combined into a group as verification data to obtain the verification samples.

[0127] Preferably, 80% of the expanded sample images and the expanded label images corresponding to the expanded sample images can be used as training images, and the remaining 20% ​​can be used as verification images. However, this is not a limitation. In one embodiment, for example, if there are 150 expanded sample images and the label images corresponding to the 150 expanded sample images, 120 of them can be used as training samples, and the remaining 30 can be used as test samples.

[0128] Preferably, in one exemplary embodiment, the neural network model is a 3D V-Net network model. V-Net is a fully convolutional neural network for volume data segmentation. As will be understood by those skilled in the art, the model parameters of a neural network model include two categories: feature parameters and hyperparameters. Feature parameters are used to learn image features, such as the texture, geometry, and location features of an aneurysm. Feature parameters include weight parameters and bias parameters. Hyperparameters are parameters manually set during training. Only by setting appropriate hyperparameters can feature parameters be learned from samples. Hyperparameters may include the learning rate, the number of hidden layers, the convolution kernel size, the number of training iterations, and the batch size per iteration. It is understood that the specific structure of the V-Net (such as the size and dimensions of the convolution kernel) can be appropriately set based on actual conditions. For example, the present invention can set the number of hidden layers to 16, 32, 64, 128, or 256, the convolution kernel size to 3×3, the number of training iterations to 500, and the batch size to 2. If the GPU memory limit is not reached, the infill step size can be gradually increased. Preferably, the loss function is the sum of the dice value and the binary cross entropy. The cross entropy can be used to improve the unstable training fluctuations that occur when using only the dice value as the loss function. The dice value is more effective when the intracranial aneurysm occupies a small area of ​​the entire image.

[0129] In step C, the first preset training end condition is that the number of training times of the neural network model is greater than or equal to the first preset number of iterations (for example, 500 times). Figure 7 , which schematically gives a detailed flow chart of the training process of the provided neural network model, starting from Figure 7 It can be seen that step C of training the pre-built neural network model according to the training sample, the verification sample and the initial values ​​of the model parameters of the neural network model until the first preset training end condition is met includes:

[0130] Step C1: using the training sample as the input of the neural network model, and obtaining the target tissue prediction result of the medical training image according to the initial values ​​of the model parameters of the neural network model.

[0131] Step C2: Calculate the value of the loss function based on the target tissue prediction result of the medical training image and the expanded label image corresponding to the medical training image; and adjust the model parameters of the neural network model based on the value of the loss function and the preliminary training preset error value until the number of training times of the neural network model is greater than or equal to the second preset number of iterations (for example, 10 times), thereby obtaining a preliminary trained neural network model.

[0132] Step C3: Input the verification sample into the preliminarily trained neural network model to obtain the target tissue prediction result of the medical verification image, calculate the value of the loss function based on the target tissue prediction result of the medical verification image and the expanded label image corresponding to the medical verification image; and record the verification result, which includes the value of the loss function and the corresponding model parameters of the preliminarily trained neural network model.

[0133] Step C4: Determine whether the number of training times of the neural network model is greater than or equal to the first preset number of iterations. If so, execute step C5; if not, adjust the model parameters of the neural network model, and update the initial values ​​of the model parameters of the neural network model to the adjusted model parameters, and return to execute step C1.

[0134] Step C5: According to the record of the verification result, the trained neural network model is obtained and the training is completed; wherein, the model parameters of the neural network model are the model parameters of the preliminarily trained neural network model corresponding to the minimum value of the loss function.

[0135] Therefore, the method of obtaining several medical images of preset sizes for sample expansion by sliding the window along the center line and according to the preset pixel distance not only has high expansion efficiency, but also has high training efficiency and strong generalization ability and good robustness of the obtained neural network model by alternating training and verification of training samples and verification samples during network model training.

[0136] Next, the medical image segmentation method step S400 is described. In one embodiment, the vascular centerline is obtained through steps S100-S300. Thus, a sliding window can be performed along the vascular centerline, centered at a point on the vascular centerline, with a pixel distance of 64 between the two center points. A cropped region of size [128, 128, 128] (i.e., the second medical image to be segmented) is intercepted from the original image, thereby obtaining a plurality of second medical images to be segmented of a predetermined size. Finally, step S500 is executed, whereby the plurality of second medical images to be segmented of the predetermined size are input into the neural network model, thereby obtaining a plurality of preliminary segmented medical images corresponding to the second medical image to be segmented.

[0137] It should be noted that, in this embodiment, since there is overlap in the cropped areas (the pixel distance between the two center points is 64, and the size of the cropped area is [128, 128, 128]), based on the result predicted by the neural network model under normal circumstances (i.e., the second segmentation), the accuracy of the central area is greater than the cognition of the edge area. Therefore, as a preference, a weight template of size [128, 128, 128] (the same as the cropped area) is defined, and the internal weight parameter is a Gaussian distribution, that is, the center weight is large and the edge weight is small. Before splicing, the preliminary segmented medical image corresponding to the predicted [128, 128, 128] second medical image to be segmented is first multiplied by the template weight, and then spliced ​​(i.e., the reverse operation is performed according to step S400), and then the spliced ​​result is adjusted to the same size as the first original medical image through nearest interpolation to obtain the final target medical image (patient's intracranial aneurysm image). Specifically, please refer to Figure 2 and Figure 8 ,in Figure 8 Schematic diagram of an intracranial aneurysm image obtained by segmentation in a specific example of the present invention. Figure 2 and Figure 8 It can be seen that by adopting the medical image segmentation method provided by the present invention to segment the medical image to be segmented (such as a three-dimensional DSA image), the desired target tissue (such as the intracranial artery 100 and the intracranial aneurysm 200) can be accurately segmented from the medical image to be segmented, and the segmentation result obtained by the medical segmentation method provided by the present invention can not only accurately segment the tissue and organ images, but also reduce the human-computer interaction operation process, which can significantly improve the doctor's diagnostic efficiency and accuracy.

[0138] Another embodiment of the present invention provides a medical image segmentation system. Specifically, see Figure 9 , which schematically shows the block structure diagram of the medical image segmentation system provided by this embodiment, Figure 9 It can be seen that the medical image segmentation system provided in this embodiment includes a first medical image to be segmented acquisition unit 310, a blood vessel region of interest acquisition unit 320, a blood vessel centerline acquisition unit 330, a second medical image to be segmented acquisition unit 340 and a target medical image acquisition unit 350.

[0139] Specifically, the first medical image to be segmented acquiring unit 310 is configured to transform the first original medical image to a target size using a first preset interpolation method based on the first original medical image, thereby acquiring a first medical image to be segmented. The vascular region of interest acquiring unit 320 is configured to perform a first segmentation on the first medical image to be segmented using an adaptive threshold algorithm, compare the volume values ​​of each connected domain in the medical image obtained by the first segmentation, and select the connected domain with the largest volume value as the trunk region of the vascular region of interest image to acquire the vascular region of interest image. The vascular centerline acquiring unit 330 is configured to acquire the vascular centerline corresponding to the vascular region of interest using a target path method. The target path method includes: acquiring the position coordinates of a starting point and at least one ending point based on the vascular region of interest image, calculating a target path between the starting point and the ending point, and using the target path as the vascular centerline corresponding to the vascular region of interest. The second medical image to be segmented acquiring unit 340 is configured to perform a sliding window operation along the vascular centerline at a preset pixel distance, centered on a point on the vascular centerline, to acquire a plurality of second medical images to be segmented of preset sizes. The target medical image acquisition unit 350 is configured to use a neural network model to perform a second segmentation on each of the second medical images to be segmented, and to splice the multiple medical images obtained by the second segmentation, and to use a second preset interpolation method to transform the multiple spliced ​​medical images to the same size as the first original medical image to obtain the final segmented medical image. Thus, the medical image segmentation system provided by the present invention automatically calculates the centerline of the blood vessel on the basis of obtaining the blood vessel region of interest (such as the intracranial artery), and crops the region of interest on the image with the point on the centerline as the center to perform network prediction of the target tissue (such as the intracranial artery and intracranial aneurysm), which can not only effectively improve the detection efficiency, but also greatly improve the overall segmentation accuracy and reduce the tedious operations of human-computer interaction. In addition, the medical image segmentation method provided by the present invention has strong versatility and realizes an end-to-end algorithm process, which can better and faster assist doctors in improving diagnostic accuracy.

[0140] Preferably, in one exemplary embodiment, please continue to refer to Figure 9 The medical image segmentation system also includes a neural network model training unit 360, which includes a training sample and verification sample acquisition module 361, a sample expansion unit 362 and a network model training unit 363.

[0141] Specifically, the training sample and validation sample acquisition module 361 is configured to acquire original training samples and original validation samples, wherein the original training samples include an original medical training image and a first labeled image corresponding to the original medical training image, wherein the first labeled image is a first medical image with the target tissue labeled; the original validation samples include an original medical validation image and a second labeled image corresponding to the original medical validation image, wherein the second labeled image is a second medical image with the target tissue labeled. The sample expansion module 362 is configured to expand the original training samples to obtain expanded training samples, wherein the expanded training samples include the expanded medical training image and the first labeled image corresponding to the expanded medical training image; and to expand the original validation samples to obtain expanded validation samples, wherein the expanded validation samples include the expanded medical validation image and the second labeled image corresponding to the expanded medical validation image. The network model training module 363 is configured to set initial values ​​for model parameters of the neural network model; and to train a pre-built neural network model based on the expanded training samples, the expanded validation samples, and the initial values ​​of the model parameters of the neural network model until a first preset training termination condition is met.

[0142] Therefore, the medical image segmentation system provided by the present invention, its neural network model training unit 360, can not only obtain training samples and verification samples, but also can expand the training samples and verification samples, not only can alleviate the shortage of samples, but also can train the pre-built neural network model according to the expanded training samples, the expanded verification samples and the initial values ​​of the model parameters of the neural network model, thereby ensuring the generalization and robustness of the network model, and improving the segmentation precision and accuracy of medical images.

[0143] Therefore, since the medical image segmentation system provided by the embodiment of the present invention and the medical image segmentation method provided by the above-mentioned embodiments belong to the same inventive concept, they have at least the same beneficial effects. Please refer to the above-mentioned beneficial effects of the medical image segmentation method for details, and no further details will be given here.

[0144] Based on the same inventive concept, an embodiment of the present invention further provides an electronic device, please refer to Figure 10 , which schematically shows a block diagram of an electronic device provided by one embodiment of the present invention. Figure 10 As shown, the electronic device includes a processor 410 and a memory 420 , wherein the memory 420 stores a computer program, and when the computer program is executed by the processor 410 , the medical image segmentation method described above is implemented.

[0145] like Figure 10As shown, the electronic device further includes a communication interface 430 and a communication bus 440, wherein the processor 410, the communication interface 430, and the memory 420 communicate with each other via the communication bus 440. The communication bus 440 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 440 may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not mean that there is only one bus or one type of bus. The communication interface 430 is used for communication between the above-mentioned electronic device and other devices.

[0146] The processor 410 referred to in the present invention may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor. The processor 410 is the control center of the electronic device, connecting various parts of the entire electronic device using various interfaces and lines.

[0147] The memory 420 may be used to store the computer program. The processor 410 implements various functions of the electronic device by running or executing the computer program stored in the memory 420 and calling the data stored in the memory 420.

[0148] The memory 420 may include nonvolatile and / or volatile memory. Nonvolatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM).

[0149] Yet another embodiment of the present invention provides a readable storage medium, wherein the readable storage medium stores a computer program, and when the computer program is executed by a processor, the medical image segmentation method described above can be implemented.

[0150] The readable storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, a system, device or component of electricity, magnetism, light, electromagnetic, infrared or semiconductor, or any combination thereof. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer hard disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this article, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0151] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0152] Computer program code for performing the operations of the present invention may be written in one or more programming languages, or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0153] In summary, compared with the prior art, the medical image segmentation method, system, electronic device, and storage medium provided by the present invention have the following advantages:

[0154] The medical image segmentation method provided by the present invention first obtains a first medical image to be segmented based on a first original medical image; then obtains a vascular region image of interest based on the first medical image to be segmented; then obtains a vascular centerline corresponding to the vascular region of interest based on the vascular region image of interest; and then slides a window along the vascular centerline at a preset pixel distance with a point on the vascular centerline as the center to obtain a plurality of second medical images to be segmented of a preset size; then segments each of the second medical images to be segmented using a neural network model to obtain a preliminary segmented medical image corresponding to the second medical image to be segmented; and then splices the plurality of preliminary segmented medical images to obtain a final target medical image. Thus, the medical image segmentation method provided by the present invention automatically calculates the vascular centerline based on the acquisition of the vascular region of interest (such as an intracranial artery), and crops the vascular centerline on the image with the point on the centerline as the center to perform network prediction of the target tissue (such as an intracranial artery and an intracranial aneurysm), which can not only effectively improve detection efficiency, but also greatly improve the overall segmentation accuracy and reduce the tedious human-computer interaction operations. In addition, the medical image segmentation method provided by the present invention is highly versatile and realizes an end-to-end algorithm process, which can better and more quickly assist doctors in improving diagnostic accuracy.

[0155] Since the medical image segmentation system, electronic device and storage medium provided by the present invention and the medical image segmentation method provided by the present invention belong to the same inventive concept, they have at least the same beneficial effects as the above-mentioned medical image segmentation method and are not described in detail here.

[0156] It should be noted that the devices and methods disclosed in the embodiments of this document may also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of the devices, methods, and computer program products according to the various embodiments of this document. In this regard, each box in the flowchart or block diagram may represent a module, program, or portion of code, wherein the module, program segment, or portion of code contains one or more executable instructions for implementing a specified logical function, and the module, program segment, or portion of code contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the boxes may also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes may actually be executed substantially in parallel, or they may sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system for performing the specified function or action, or may be implemented by a combination of dedicated hardware and computer instructions.

[0157] In addition, the functional modules in the various embodiments of this document may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0158] The above description is merely a description of preferred embodiments of the present invention and does not limit the scope of the present invention. Any changes or modifications made by persons skilled in the art based on the above disclosure are within the scope of protection of the present invention. Obviously, various modifications and variations may be made by persons skilled in the art without departing from the spirit and scope of the present invention. Thus, if such modifications and variations fall within the scope of the present invention and its equivalents, the present invention is intended to include such modifications and variations.

Claims

1. A medical image segmentation method, characterized in that: include: According to the first original medical image, using a first preset interpolation method, the first original medical image is transformed to a target size to obtain a first medical image to be segmented; performing a first segmentation on the first medical image to be segmented using an adaptive threshold algorithm, comparing the volume values ​​of each connected domain in the medical image obtained by the first segmentation, and taking the connected domain with the largest volume value as the trunk region of the region of interest image to obtain a vascular region of interest image; A target path method is used to obtain a blood vessel centerline corresponding to the blood vessel region of interest; wherein the target path method includes: obtaining position coordinates of a starting point and at least one ending point based on the image of the blood vessel region of interest, calculating a target path between the starting point and the ending point, and using the target path as the blood vessel centerline corresponding to the blood vessel region of interest; Taking the point on the centerline of the blood vessel as the center and sliding the window along the centerline of the blood vessel at a preset pixel distance, to obtain a plurality of second medical images to be segmented of preset sizes; A neural network model is used to perform a second segmentation on each of the second medical images to be segmented, and several medical images obtained by the second segmentation are spliced. A second preset interpolation method is used to transform the several spliced ​​medical images to the same size as the first original medical image to obtain a final target medical image.

2. The medical image segmentation method according to claim 1, characterized in that: The method of using the connected region with the maximum volume value as the trunk region of the region of interest image to obtain the image of the blood vessel region of interest includes: According to the maximum volume value and the preset threshold coefficient, a preset volume threshold is obtained; all connected domains are traversed, and all connected domains with volumes greater than the preset volume threshold are retained to obtain the image of the blood vessel region of interest.

3. The medical image segmentation method according to claim 1, wherein: Before acquiring the blood vessel centerline corresponding to the blood vessel region of interest using the first blood vessel centerline acquisition method, the method further includes: For each connected domain of the image of the vascular region of interest, it is determined whether the connected domain is a trunk region. If so, the target path method is used to obtain the vascular centerline corresponding to the vascular region of interest; if not, a 3D skeleton extraction algorithm is used to extract the skeleton of the connected domain, and the result of the skeleton extraction is used as the vascular centerline corresponding to the vascular region of interest.

4. The medical image segmentation method according to claim 1, wherein: The step of obtaining a starting point according to the image of the blood vessel region of interest includes: S311: Acquire N first continuous slices starting from a starting slice of the image of the vascular region of interest toward an ending slice, where N is greater than or equal to 3, and the starting slice is closer to a starting scanning position of the patient when acquiring the first original medical image than the ending slice; S312: Determine whether the ellipticity of the connected domains corresponding to the N first continuous slices is greater than a first preset threshold, and whether the deviation of the centroid coordinates of the connected domains corresponding to the first continuous slices on the X-axis and the Y-axis is less than a second preset threshold. If so, obtain the starting point based on the centroid of the connected domain corresponding to the first continuous slices; if not, execute step S313; S313: Starting from the adjacent slice second closest to the starting slice in the first continuous slices toward the ending slice, obtain N second continuous slices, and use the second continuous slices as the first continuous slices, and execute step S312.

5. The medical image segmentation method according to claim 1, wherein: The step of obtaining a plurality of end points based on the image of the blood vessel region of interest includes: S321: performing an opening operation on the image of the blood vessel region of interest, and then subtracting a result of the opening operation from the image of the region of interest to obtain a plurality of end regions of the image of the blood vessel region of interest; S322: Performing a connected domain analysis on each of the terminal regions to obtain the centroid of the connected domain corresponding to each of the terminal regions, and using the centroid of the connected domain as a termination point, thereby obtaining a plurality of termination points of the image of the vascular region of interest.

6. The medical image segmentation method according to claim 1, wherein: Before segmenting each of the second medical images to be segmented using the neural network model, the method further includes training the neural network model through the following steps: Step A: Acquire an original sample, wherein the original sample includes a second original medical image and a labeled image corresponding to the second original medical image, wherein the labeled image is a medical image in which a target tissue is annotated; Step B: Expanding the original sample to obtain a training sample and a verification sample, wherein the training sample includes an expanded medical training image and a label image corresponding to the expanded medical training image; the verification sample includes an expanded medical verification image and a label image corresponding to the expanded medical verification image; Step C: setting the initial values ​​of the model parameters of the neural network model; as well as The pre-built neural network model is trained according to the training samples, the verification samples and the initial values ​​of the model parameters of the neural network model until a first preset training end condition is met.

7. The medical image segmentation method according to claim 6, characterized in that: The step of expanding the original sample to obtain a training sample and a verification sample includes: Preprocessing the second original medical image and the label image corresponding to the second original medical image respectively to obtain a preprocessed second original medical image and a preprocessed label image; acquiring, according to the preprocessed second original medical image, a blood vessel region image of interest corresponding to the second original medical image and a blood vessel centerline of the blood vessel region image of interest corresponding to the second original medical image; Taking the blood vessel centerline of the image of the blood vessel region of interest corresponding to the second original medical image as the center point, sliding a window along the blood vessel centerline and at the preset pixel distance to obtain a plurality of medical images of preset sizes, thereby obtaining an expanded sample image; Taking the center points of the plurality of expanded sample images of the preset sizes as a first center point, obtaining a plurality of second center points corresponding to the first center point on a preprocessed label image corresponding to the preprocessed second original medical image, and obtaining a plurality of medical images of the preset sizes with the second center point as the center to obtain an expanded label image; According to a preset proportional distribution relationship, the expanded sample images are randomly divided into two groups, one of which is used as medical training images and the other is used as medical verification images; the medical training images and the expanded label images corresponding to the medical training images are combined into a group as training data to obtain the training samples; the medical verification images and the expanded label images corresponding to the medical verification images are combined into a group as verification data to obtain the verification samples.

8. The medical image segmentation method according to claim 6, characterized in that: The first preset training end condition is that the number of training times of the neural network model is greater than or equal to the first preset number of iterations; In step C, the pre-built neural network model is trained according to the training sample, the verification sample and the initial values ​​of the model parameters of the neural network model until a first preset training end condition is met, including: Step C1: using the training sample as input of the neural network model, and obtaining a target tissue prediction result of the medical training image according to the initial values ​​of the model parameters of the neural network model; Step C2: calculating a loss function value based on the target tissue prediction result of the medical training image and the expanded label image corresponding to the medical training image; and adjusting the model parameters of the neural network model based on the loss function value and a preliminary training preset error value until the number of training times of the neural network model is greater than or equal to a second preset number of iterations, thereby obtaining a preliminary trained neural network model; Step C3: Inputting the verification sample into the preliminarily trained neural network model, obtaining a target tissue prediction result of the medical verification image, calculating a loss function value based on the target tissue prediction result of the medical verification image and the expanded label image corresponding to the medical verification image; and recording a verification result, the verification result including the loss function value and the corresponding model parameters of the preliminarily trained neural network model; Step C4: determining whether the number of training times of the neural network model is greater than or equal to the first preset number of iterations; if so, executing step C5; if not, adjusting the model parameters of the neural network model, and updating the initial values ​​of the model parameters of the neural network model to the adjusted model parameters, and returning to executing step C1; Step C5: According to the record of the verification result, the neural network model is obtained and the training is completed; wherein, the model parameters of the neural network model are the model parameters of the preliminarily trained neural network model corresponding to the minimum value of the loss function.

9. A medical image segmentation system, characterized in that: include: A first medical image to be segmented acquiring unit is configured to transform the first original medical image to a target size using a first preset interpolation method according to the first original medical image to acquire the first medical image to be segmented; a vessel region of interest acquisition unit configured to perform a first segmentation on the first medical image to be segmented using an adaptive threshold algorithm, compare the volume values ​​of each connected domain in the medical image obtained by the first segmentation, and use the connected domain with the largest volume value as the trunk region of the region of interest image to acquire the vessel region of interest image; a blood vessel centerline acquisition unit configured to acquire the blood vessel centerline corresponding to the blood vessel region of interest using a target path method; wherein the target path method includes: acquiring position coordinates of a starting point and at least one ending point based on the image of the blood vessel region of interest, calculating a target path between the starting point and the ending point, and using the target path as the blood vessel centerline corresponding to the blood vessel region of interest; a second medical image to be segmented acquiring unit configured to perform a sliding window operation along the blood vessel centerline at a preset pixel distance with a point on the blood vessel centerline as the center, so as to acquire a plurality of second medical images to be segmented of preset sizes; The target medical image acquisition unit is configured to use a neural network model to perform a second segmentation on each of the second medical images to be segmented, and to splice the multiple medical images obtained by the second segmentation, and to use a second preset interpolation method to transform the multiple spliced ​​medical images to the same size as the first original medical image to obtain a final segmented medical image.

10. An electronic device, characterized in that: The medical image segmentation method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the medical image segmentation method according to any one of claims 1 to 8 is implemented.

11. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the medical image segmentation method according to any one of claims 1 to 8 is implemented.

Images