Training method and device of medical image segmentation model and electronic equipment

Abstract

The invention relates to a medical image segmentation model training method and device and electronic equipment, and the method comprises the steps: carrying out the pre-training of a first image segmentation model through a first data set, and obtaining a pre-trained first image segmentation model, each training sample in the first data set is a medical image marked with position information and morphological feature information of the target area; inputting a second data set composed of unlabeled medical images into the pre-trained first image segmentation model to obtain first label information corresponding to each medical image in the second data set; inputting a third data set composed of the medical images carrying the first label information into the second image segmentation model to obtain second label information corresponding to the medical images in the third data set; and performing optimization training on the pre-trained first image segmentation model by using the first data set and a fourth data set consisting of the medical images carrying the second label information.

Description

Technical Field

[0001] The present disclosure relates to image segmentation technology, and more specifically, to a training method, device, and electronic device for a medical image segmentation model. Background Art

[0002] In recent years, the continuous development of deep learning technology has driven a paradigm shift in computer-assisted diagnosis (CAD) in medical image analysis. A typical deep learning application paradigm is to train high-precision models with plug-and-play capabilities using large-scale annotated datasets. The implementation and development of deep learning methods in CAD have significantly reduced the reliance on professional physicians to interpret images in the traditional diagnostic process.

[0003] However, deep learning methods are data-driven at their core. Training a deep neural network to perform well on a specific task requires large amounts of labeled data to fully fit the deep neural network. This presents unique data challenges in the field of medical image analysis. First, the privacy and sensitivity of medical data make cross-institutional data sharing difficult, and publicly available large-scale labeled datasets are scarce. Second, medical image annotation requires the expertise of radiologists. For example, annotating a lesion in a single CT image can take over 30 minutes, which exponentially increases annotation costs.

[0004] In the past two years, the emergence of universal visual foundational models, such as the Segment Anything Model (SAM), has marked a major breakthrough in foundational models in the field of computer vision (CV), following natural language processing. This model, with its strong generalization capabilities achieved through large-scale data pre-training, provides a new technical paradigm for cross-domain image segmentation tasks. However, in the field of medical image analysis, the scarcity of high-quality annotated data has consistently constrained the clinical application of SAM. To address this industry pain point, researchers have begun exploring SAM's zero-shot learning capabilities for medical image segmentation tasks.

[0005] Therefore, a new technical solution needs to be provided to improve the accuracy of medical image segmentation. Summary of the Invention

[0006] One objective of the present disclosure is to provide a new technical solution for a training method of a medical image segmentation model.

[0007] According to a first aspect of the present disclosure, a method for training a medical image segmentation model is provided, comprising:

[0008] Pre-training a first image segmentation model using a first data set to obtain a pre-trained first image segmentation model, wherein each training sample in the first data set is a medical image annotated with location information and morphological feature information of a target area;

[0009] Inputting a second data set consisting of unlabeled medical images into the pre-trained first image segmentation model to obtain first label information corresponding to each medical image in the second data set, wherein the first label information includes location information and morphological feature information of the target area;

[0010] inputting a third data set consisting of medical images carrying the first label information into a second image segmentation model to obtain second label information corresponding to each medical image in the third data set, wherein the second label information includes optimized location information and morphological feature information of the target area;

[0011] The pre-trained first image segmentation model is optimized and trained using the first data set and a fourth data set consisting of medical images carrying the second label information to obtain the optimized and trained first image segmentation model.

[0012] Optionally, the second data set includes continuous two-dimensional medical image slices determined based on three-dimensional medical images, and the second data set consisting of unlabeled medical images is input into the pre-trained first image segmentation model to obtain first label information corresponding to each medical image in the second data set, including:

[0013] Inputting a second dataset consisting of unlabeled continuous two-dimensional medical image slices into the pre-trained first image segmentation model to obtain first label information and confidence scores corresponding to each two-dimensional medical image slice in the second dataset, wherein the confidence scores are used to indicate the reliability of the corresponding first label information;

[0014] Before inputting a third data set consisting of medical images carrying the first label information into a second image segmentation model to obtain second label information corresponding to each medical image in the third data set, the method further includes:

[0015] According to the confidence level of the first label information corresponding to each two-dimensional medical image slice, two-dimensional medical image slices corresponding to the first label information whose confidence level meets preset requirements are screened to form the third data set.

[0016] Optionally, the step of screening, based on the confidence level of the first label information corresponding to each two-dimensional medical image slice, two-dimensional medical image slices corresponding to the first label information whose confidence level meets a preset requirement to form the third data set includes:

[0017] Dividing the continuous medical image slices into a plurality of medical image groups, wherein the medical image slices in each medical image group are continuous;

[0018] Based on the confidence of the first label information corresponding to each two-dimensional medical image slice in each medical image group, the two-dimensional medical image slice corresponding to the first label information with the highest confidence is screened to form the third data set.

[0019] Optionally, the optimization training of the pre-trained first image segmentation model is iterative optimization training, wherein the method further includes:

[0020] After completing one iterative optimization training on the pre-trained first image segmentation model, optimizing the first label information corresponding to each medical image in the second data set using the first image segmentation model that has completed one iterative optimization training to obtain optimized first label information;

[0021] Inputting the data set consisting of medical images carrying the optimized first label information into the second image segmentation model again to obtain the optimized second label information corresponding to each medical image in the corresponding data set;

[0022] The pre-trained first image segmentation model is further iteratively optimized and trained using the first data set and a data set consisting of medical images carrying the optimized second label information.

[0023] Optionally, before inputting the third data set consisting of medical images carrying the first label information into the second image segmentation model to obtain the second label information corresponding to each medical image in the third data set, the method further includes:

[0024] The second image segmentation model is pre-trained using a priori medical knowledge base to obtain a pre-trained second image segmentation model, wherein the priori medical knowledge base includes medical images annotated with location information and morphological feature information of the target area.

[0025] Optionally, the method further includes: inputting the second data set into the optimized and trained first image segmentation model to obtain label information corresponding to each medical image in the second data set;

[0026] determining a Dice coefficient corresponding to each medical image in the second dataset based on label information manually annotated for each medical image in the second dataset and label information corresponding to each medical image in the second dataset obtained based on the optimized trained first image segmentation model;

[0027] determining the segmentation accuracy of the optimized trained first image segmentation model according to the Dice coefficient corresponding to each medical image in the second data set;

[0028] When the segmentation accuracy of the first image segmentation model after the optimized training does not meet the preset segmentation accuracy requirement, the first image segmentation model after the pre-training is optimized and trained using the first data set and a fourth data set consisting of medical images carrying the second label information.

[0029] Optionally, the first image segmentation model is a U-Net model, and the second image segmentation model is a SAM2 model.

[0030] According to a second aspect of the present disclosure, a training device for a medical image segmentation model is provided, comprising:

[0031] a pre-training module, configured to pre-train a first image segmentation model using a first data set to obtain a pre-trained first image segmentation model, wherein each training sample in the first data set is a medical image annotated with location information and morphological feature information of a target region;

[0032] a first label information determination module, configured to input a second data set consisting of unlabeled medical images into the pre-trained first image segmentation model to obtain first label information corresponding to each medical image in the second data set, wherein the first label information includes location information and morphological feature information of the target area;

[0033] a second label information determination module, configured to input a third data set consisting of medical images carrying the first label information into a second image segmentation model to obtain second label information corresponding to each medical image in the third data set, wherein the second label information includes optimized location information and morphological feature information of the target area;

[0034] An optimization training module is used to optimize the pre-trained first image segmentation model using the first data set and a fourth data set consisting of medical images carrying the second label information to obtain the optimized trained first image segmentation model.

[0035] According to a third aspect of the present disclosure, an electronic device is provided, comprising a memory and a processor, wherein the memory stores a computer program, and the computer program is used to control the processor to operate to execute the training method of the medical image segmentation model according to any one of the first aspects of the present disclosure.

[0036] According to a fourth aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the training method of the medical image segmentation model described in any one of the first aspects is implemented.

[0037] The present disclosure provides a training method for a medical image segmentation model, which proposes a model framework that combines a proprietary medical image segmentation model (i.e., a first image segmentation model) with a visual base model for medical image segmentation (i.e., a second image segmentation model). The visual base model for medical image segmentation is incorporated into a semi-supervised learning framework, and is coordinated with a proprietary medical image segmentation model pre-trained based on a small amount of labeled data to obtain a large number of training samples to obtain an expanded training data set, and optimize the training of the proprietary medical image segmentation model, thereby achieving training of the proprietary medical image segmentation model using only a small number of labeled training samples.

[0038] Features and advantages of the embodiments of the present specification will become apparent from the following detailed description of exemplary embodiments of the present specification with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and, together with the description, serve to explain the principles of the embodiments of the specification.

[0040] Figure 1 A processing flow chart of a method for training a medical image segmentation model according to an embodiment of the present disclosure is shown.

[0041] Figure 2 A schematic diagram of iterative training of a pre-trained first medical image segmentation model according to an embodiment of the present disclosure is shown.

[0042] Figure 3 A principle block diagram of a training device for a medical image segmentation model according to an embodiment of the present disclosure is shown.

[0043] Figure 4 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0044] Various exemplary embodiments of the present specification will now be described in detail with reference to the accompanying drawings.

[0045] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the embodiments of this specification, its application, or uses.

[0046] It should be noted that like reference numerals and letters refer to like items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0047] To solve the above technical problems, the embodiments of the present disclosure provide a training method for a medical image segmentation model, and propose a model framework that combines a proprietary medical image segmentation model (i.e., a first image segmentation model) with a visual base model for medical image segmentation (i.e., a second image segmentation model). The general visual base model is incorporated into the framework of semi-supervised learning, and is coordinated with the proprietary medical image segmentation model obtained by pre-training based on a small amount of labeled data to obtain a large number of training samples to obtain an expanded training data set, and optimize the training of the proprietary medical image segmentation model, thereby achieving the training of the proprietary medical image segmentation model using only a small number of labeled training samples.

[0048] Figure 1 FIG. 1 is a flow chart showing a method for training a medical image segmentation model according to an embodiment of the present disclosure. Figure 1 As shown, the method includes steps S110 to S140.

[0049] Step S110 , pre-training the first image segmentation model using the first data set to obtain the pre-trained first image segmentation model, wherein each training sample in the first data set is a medical image annotated with location information and morphological feature information of the target area.

[0050] The first image segmentation model can be a U-net model. The annotation information for each training sample in the first dataset is manually annotated. The target region is an organ tissue or bone tissue. The location information of the target region is the location information of a tissue in a medical image and can be represented using the tissue boundary information. The morphological feature information of the target region is the identification information of a tissue and can be distinguished using different grayscale values.

[0051] The first data set includes a small number of training samples. Therefore, the first image segmentation model is pre-trained using the first data set, so that the pre-trained first image segmentation model can initially learn the relevant features of the medical tissue structure, but does not have the ability to achieve high segmentation accuracy.

[0052] In step S120, a second data set consisting of unlabeled medical images is input into the pre-trained first image segmentation model to obtain first label information corresponding to each medical image in the second data set, wherein the first label information includes location information and morphological feature information of the target area.

[0053] The pre-trained first image segmentation model is used to segment the unlabeled medical images in the second dataset to obtain the location information and morphological feature information of the target area in each medical image.

[0054] Since the pre-trained first image segmentation model does not have a high-accuracy segmentation capability, the quality of the first label information generated for the unlabeled medical image is poor and needs further optimization.

[0055] In step S130, a third data set consisting of medical images carrying the first label information is input into the second image segmentation model to obtain second label information corresponding to each medical image in the third data set, where the second label information includes the optimized location information and morphological feature information of the target area.

[0056] The second image segmentation model may be a SAM2 model.

[0057] The first label information carried by each medical image in the third dataset serves as prompt information when the second image segmentation model segments each medical image in the third dataset, thereby optimizing the segmentation results generated by the second image segmentation model. This means optimizing the first label information carried by each medical image to obtain more accurate second label information. The optimization of the first label information carried by each medical image specifically includes optimizing the position information of the target region and correcting the morphological feature information. The optimization of the position information of the target region refers to optimizing the boundary information of the target region in the medical image. Correcting the morphological feature information refers to correcting the tissue identification information corresponding to the target region.

[0058] Step S140 , optimizing and training the pre-trained first image segmentation model using the first data set and a fourth data set consisting of medical images carrying second label information, to obtain the optimized and trained first image segmentation model.

[0059] An expanded training dataset is formed by the first dataset and the fourth dataset consisting of medical images carrying the second label information, which can provide sufficient training samples for the optimized training of the pre-trained first image segmentation model, thereby helping to improve the segmentation performance of the optimized trained first image segmentation model.

[0060] In some embodiments, the second data set includes continuous two-dimensional medical image slices determined based on the three-dimensional medical image. The two-dimensional medical image slices are two-dimensional cross-sectional images obtained along any one of the axial, sagittal, and coronal planes based on the three-dimensional medical image.

[0061] In this embodiment, step S120 specifically includes: inputting a second data set consisting of unlabeled continuous two-dimensional medical image slices into the pre-trained first image segmentation model to obtain first label information and confidence corresponding to each two-dimensional medical image slice in the second data set, wherein the confidence is used to indicate the degree of credibility of the corresponding first label information.

[0062] In this embodiment, before inputting the third data set composed of medical images carrying the first label information into the second image segmentation model to obtain the second label information corresponding to each medical image in the third data set, the method also includes: based on the confidence of the first label information corresponding to each two-dimensional medical image slice, screening the two-dimensional medical image slices corresponding to the confidence of the first label information that meets the preset requirements to form the third data set.

[0063] Since the confidence level of the label information carried by each medical image slice in the third data set is relatively high, when the second image segmentation model segments each medical image slice in the third data set, the label information carried by each medical image slice serves as an accurate prompt, thereby making the second label information corresponding to each medical image slice more accurate, thereby providing more accurate training data for the optimized training of the pre-trained first image segmentation model.

[0064] In some embodiments, based on the confidence of the first label information corresponding to each two-dimensional medical image slice, two-dimensional medical image slices corresponding to the confidence of the first label information meeting the preset requirements are screened to form a third data set, specifically including: dividing the continuous medical image slices into multiple medical image groups, wherein the medical image slices in each medical image group are continuous; based on the confidence of the first label information corresponding to each two-dimensional medical image slice in each medical image group, two-dimensional medical image slices corresponding to the highest confidence of the first label information are screened to form the third data set.

[0065] The continuity of each medical image slice is manifested as spatial continuity. The number of frames of each medical image slice in each medical image group can be set according to demand, for example, three frames.

[0066] In some embodiments, the optimization training of the pre-trained first image segmentation model is iterative optimization training. The method further includes: after completing one iterative optimization training of the pre-trained first image segmentation model, using the first image segmentation model that has completed one iterative optimization training to optimize the first label information corresponding to each medical image in the second data set to obtain optimized first label information; inputting the data set composed of medical images carrying the optimized first label information into the second image segmentation model again to obtain optimized second label information corresponding to each medical image in the corresponding data set; and using the first data set and the data set composed of medical images carrying the optimized second label information to perform another iterative optimization training on the pre-trained first image segmentation model.

[0067] The iterative optimization training of the first image segmentation model after pre-training can be found in Figure 2 .

[0068] according to Figure 2 As shown, a first image segmentation model is pre-trained using a first dataset to obtain a pre-trained first image segmentation model. A second dataset consisting of unlabeled medical images is input into the pre-trained first image segmentation model to obtain first label information corresponding to each medical image in the second dataset. A third dataset consisting of medical images carrying the first label information is input into the second image segmentation model to obtain second label information corresponding to each medical image in the third dataset. The pre-trained first image segmentation model is then iteratively optimized using the first dataset and a fourth dataset consisting of medical images carrying the second label information.

[0069] Combine Figure 2 After completing one iterative optimization training of the pre-trained first image segmentation model, the first label information corresponding to each medical image in the second data set is optimized using the first image segmentation model that has completed one iterative optimization training to obtain the optimized first label information; the data set composed of medical images carrying the optimized first label information is input into the second image segmentation model again to obtain the optimized second label information corresponding to each medical image in the corresponding data set; using the first data set and the data set composed of medical images carrying the optimized second label information, the pre-trained first image segmentation model is iteratively optimized again until the segmentation accuracy of the optimized trained first image segmentation model reaches the preset segmentation accuracy requirement.

[0070] In this embodiment, during the iterative optimization training of the pre-trained first image segmentation model, the entire iterative process forms a closed-loop feedback system of "label information optimization-pre-trained first image segmentation model optimization". Through multiple iterative optimization trainings, the segmentation performance of the pre-trained first image segmentation model is gradually improved until the segmentation accuracy of the pre-trained first image segmentation model converges to a stable state.

[0071] In some embodiments, before step S130, the method further includes: pre-training the second image segmentation model using a priori medical knowledge base to obtain a pre-trained second image segmentation model, wherein the priori medical knowledge base includes medical images annotated with location information and morphological feature information of the target area. Inputting a third data set consisting of medical images carrying the first label information into the pre-trained second image segmentation model to obtain second label information corresponding to each medical image in the third data set. By pre-training the second image segmentation model using the priori medical knowledge base, the pre-trained second image segmentation model can initially learn relevant features of medical tissue structure and is suitable for medical image segmentation.

[0072] Taking knee joint images from the prior medical knowledge base as an example, cartilage is always attached to the femur and tibia. Based on the knee joint images, four target regions are segmented: the femur, tibia, cartilage attached to the femur, and cartilage attached to the tibia. The second image segmentation model is pre-trained using the prior medical knowledge base, enabling it to initially learn the relevant structural features of the knee joint.

[0073] In some embodiments, the method further includes: inputting the second data set into the optimized and trained first image segmentation model to obtain label information corresponding to each medical image in the second data set; determining the Dice coefficient corresponding to each medical image in the second data set based on the label information manually annotated for each medical image in the second data set and the label information corresponding to each medical image in the second data set obtained based on the optimized and trained first image segmentation model; determining the segmentation accuracy of the optimized and trained first image segmentation model based on the Dice coefficient corresponding to each medical image in the second data set; if the segmentation accuracy of the optimized and trained first image segmentation model does not meet the preset segmentation accuracy requirement, continuing to use the first data set and a fourth data set composed of medical images carrying the second label information to optimize the training of the pre-trained first image segmentation model.

[0074] The Dice coefficient is one of the most commonly used evaluation metrics in the field of medical image segmentation. It measures the degree of similarity between the manually annotated label information of a medical image in a second dataset and the label information of the corresponding medical image in the second dataset obtained using an optimized and trained first image segmentation model. A larger Dice coefficient indicates a more accurate segmentation result from the optimized and trained first image segmentation model.

[0075] In this embodiment, based on the Dice coefficient corresponding to each medical image in the second dataset, the degree of growth of the Dice coefficient corresponding to each medical image is determined as the segmentation accuracy of the first image segmentation model after optimization training. When the degree of growth of the Dice coefficient corresponding to at least a preset number of medical images is greater than a preset threshold, the pre-trained first image segmentation model is further optimized using the first dataset and a fourth dataset consisting of medical images carrying second label information. The degree of growth of the Dice coefficient corresponding to each medical image is the difference between the Dice coefficient obtained based on the current iterative training and the Dice coefficient obtained based on the previous iterative training. The greater the degree of growth of the Dice coefficient corresponding to each medical image, the more room there is for further optimization of the first image segmentation model after optimization training, and further optimization training is required.

[0076] The present disclosure also provides a training device for a medical image segmentation model for implementing any of the above method embodiments. Figure 3 FIG. 4 shows a structural block diagram of a training device for a medical image segmentation model according to some embodiments. Figure 3 As shown, the training device 300 for the medical image segmentation model may include a pre-training module 310 , a first label information determination module 320 , a second label information determination module 330 , and an optimization training module 340 .

[0077] The pre-training module 310 is used to pre-train the first image segmentation model using the first data set to obtain the pre-trained first image segmentation model, wherein each training sample in the first data set is a medical image annotated with location information and morphological feature information of the target area.

[0078] The first label information determination module 320 is used to input a second data set consisting of unlabeled medical images into the pre-trained first image segmentation model to obtain the first label information corresponding to each medical image in the second data set, wherein the first label information includes the location information and morphological feature information of the target area.

[0079] The second label information determination module 330 is used to input a third data set consisting of medical images carrying the first label information into the second image segmentation model to obtain the second label information corresponding to each medical image in the third data set. The second label information includes the optimized location information and morphological feature information of the target area.

[0080] The optimization training module 340 is used to optimize the pre-trained first image segmentation model using the first data set and a fourth data set consisting of medical images carrying second label information to obtain the optimized trained first image segmentation model.

[0081] In some embodiments, the second dataset includes continuous two-dimensional medical image slices determined based on three-dimensional medical images. The first label information determination module 320 is configured to input the second dataset, consisting of unlabeled continuous two-dimensional medical image slices, into the pre-trained first image segmentation model to obtain first label information and confidence levels corresponding to each two-dimensional medical image slice in the second dataset, where the confidence levels are used to indicate the degree of credibility of the corresponding first label information.

[0082] The device further includes a screening module configured to screen, based on the confidence of the first label information corresponding to each two-dimensional medical image slice, two-dimensional medical image slices corresponding to the first label information whose confidence meets a preset requirement, to form a third data set.

[0083] In some embodiments, the screening module is used to divide continuous medical image slices into multiple medical image groups, wherein the medical image slices in each medical image group are continuous; based on the confidence of the first label information corresponding to each two-dimensional medical image slice in each medical image group, the two-dimensional medical image slice corresponding to the first label information with the highest confidence is screened to form a third data set.

[0084] In some embodiments, the optimization training of the pre-trained first image segmentation model is iterative optimization training. The optimization training module 340 is used to optimize the first label information corresponding to each medical image in the second data set using the first image segmentation model that has completed the iterative optimization training after completing one iterative optimization training on the pre-trained first image segmentation model to obtain optimized first label information; input the data set composed of medical images carrying the optimized first label information into the second image segmentation model again to obtain optimized second label information corresponding to each medical image in the corresponding data set; and use the first data set and the data set composed of medical images carrying the optimized second label information to perform another iterative optimization training on the pre-trained first image segmentation model.

[0085] In some embodiments, the apparatus further includes a second image segmentation model pre-training module. The second image segmentation model pre-training module is configured to pre-train the second image segmentation model using a priori medical knowledge base to obtain a pre-trained second image segmentation model, wherein the priori medical knowledge base includes medical images annotated with location information and morphological feature information of target regions.

[0086] In some embodiments, the device further includes a verification module. The verification module is configured to input the second data set into the optimized and trained first image segmentation model to obtain label information corresponding to each medical image in the second data set; determine the Dice coefficient corresponding to each medical image in the second data set based on the label information manually annotated for each medical image in the second data set and the label information corresponding to each medical image in the second data set obtained based on the optimized and trained first image segmentation model; determine the segmentation accuracy of the optimized and trained first image segmentation model based on the Dice coefficient corresponding to each medical image in the second data set; and, if the segmentation accuracy of the optimized and trained first image segmentation model does not meet the preset segmentation accuracy requirement, continue to optimize and train the pre-trained first image segmentation model using the first data set and a fourth data set consisting of medical images carrying the second label information.

[0087] The present disclosure also provides an electronic device for implementing any of the above method embodiments. Figure 4 FIG2 shows a block diagram of a structure of an electronic device 4 according to some embodiments. The electronic device 4 may be a PC, a workstation, a notebook computer, a server, etc., which is not limited here.

[0088] like Figure 4 As shown, the electronic device 4 includes a processor 410 and a memory 420 for storing instructions executable by the processor 410. The processor 410 is configured to implement the training method of the medical image segmentation model according to any embodiment of the present disclosure when executing the instructions stored in the memory 420.

[0089] The processor 410 is used to execute computer instructions, which can be written using an instruction set of an architecture such as x86, Arm, RISC, MIPS, or SSE. The memory 420 includes, for example, ROM (read-only memory), RAM (random access memory), and non-volatile memory such as a hard disk, which are not limited here.

[0090] The present disclosure also provides a non-volatile computer-readable storage medium having computer program instructions stored thereon. When the computer program instructions are executed by a processor, the training method of the medical image segmentation model provided in any of the above embodiments is implemented.

[0091] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. For the device embodiments, the relevant parts can be referred to the partial description of the method embodiments.

[0092] The foregoing description of this specification describes specific embodiments. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in an order different from that described in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order shown or the sequential order to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0093] The embodiments of this specification may be systems, methods, and / or computer program products. The computer program product may include a computer-readable storage medium carrying computer instructions for causing a processor to implement various aspects of the embodiments of this specification.

[0094] A computer-readable storage medium can be a tangible device that can hold and store computer instructions for use by a computer instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, such as a punch card or a raised structure in a groove on which computer instructions are stored, and any suitable combination thereof. As used herein, a computer-readable storage medium is not to be construed as a transient signal per se, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (e.g., a light pulse through a fiber optic cable), or an electrical signal transmitted through an electrical wire.

[0095] The computer instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network layer, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network layer can include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network layer adapter card or network layer interface in each computing / processing device receives computer instructions from the network layer and forwards the computer instructions for storage in a computer-readable storage medium in each computing / processing device.

[0096] The flowcharts and block diagrams in the accompanying drawings show the possible implementation architectures, functions and operations of the systems, methods and computer program products according to multiple embodiments of this specification. In this regard, each box in the flowchart or block diagram can represent a module, program segment or part of a computer instruction, and the module, program segment or part of a computer instruction contains one or more executable computer instructions for implementing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can 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, can be implemented using a dedicated hardware-based system that performs the specified function or action, or can be implemented using a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that implementation by hardware, implementation by software, and implementation by a combination of software and hardware are all equivalent.

[0097] The embodiments of the present specification have been described above. The above description is illustrative and not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope of the described embodiments. The terminology used herein is selected to best explain the principles of the embodiments, their practical applications, or improvements to existing technologies, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A training method for a medical image segmentation model, characterized in that: include: Pre-training a first image segmentation model using a first data set to obtain a pre-trained first image segmentation model, wherein each training sample in the first data set is a medical image annotated with location information and morphological feature information of a target area; Inputting a second data set consisting of unlabeled medical images into the pre-trained first image segmentation model to obtain first label information corresponding to each medical image in the second data set, wherein the first label information includes location information and morphological feature information of the target area; inputting a third data set consisting of medical images carrying the first label information into a second image segmentation model to obtain second label information corresponding to each medical image in the third data set, wherein the second label information includes optimized location information and morphological feature information of the target area; The pre-trained first image segmentation model is optimized and trained using the first data set and a fourth data set consisting of medical images carrying the second label information to obtain the optimized and trained first image segmentation model.

2. The method according to claim 1, characterized in that The second data set includes continuous two-dimensional medical image slices determined based on three-dimensional medical images. Inputting the second data set consisting of unlabeled medical images into the pre-trained first image segmentation model to obtain first label information corresponding to each medical image in the second data set includes: Inputting a second dataset consisting of unlabeled continuous two-dimensional medical image slices into the pre-trained first image segmentation model to obtain first label information and confidence scores corresponding to each two-dimensional medical image slice in the second dataset, wherein the confidence scores are used to indicate the reliability of the corresponding first label information; Before inputting a third data set consisting of medical images carrying the first label information into a second image segmentation model to obtain second label information corresponding to each medical image in the third data set, the method further includes: According to the confidence level of the first label information corresponding to each two-dimensional medical image slice, two-dimensional medical image slices corresponding to the first label information whose confidence level meets preset requirements are screened to form the third data set.

3. The method according to claim 2, characterized in that The method of screening, based on the confidence level of the first label information corresponding to each two-dimensional medical image slice, two-dimensional medical image slices corresponding to the first label information whose confidence level meets a preset requirement to form the third data set includes: Dividing the continuous medical image slices into a plurality of medical image groups, wherein the medical image slices in each medical image group are continuous; Based on the confidence of the first label information corresponding to each two-dimensional medical image slice in each medical image group, the two-dimensional medical image slice corresponding to the first label information with the highest confidence is screened to form the third data set.

4. The method according to claim 1, wherein The optimization training of the pre-trained first image segmentation model is an iterative optimization training, wherein the method further includes: After completing one iterative optimization training on the pre-trained first image segmentation model, optimizing the first label information corresponding to each medical image in the second data set using the first image segmentation model that has completed one iterative optimization training to obtain optimized first label information; Inputting the data set consisting of medical images carrying the optimized first label information into the second image segmentation model again to obtain the optimized second label information corresponding to each medical image in the corresponding data set; The pre-trained first image segmentation model is further iteratively optimized and trained using the first data set and a data set consisting of medical images carrying the optimized second label information.

5. The method according to claim 1, characterized in that Before inputting the third data set composed of medical images carrying the first label information into the second image segmentation model to obtain the second label information corresponding to each medical image in the third data set, the method further includes: The second image segmentation model is pre-trained using a priori medical knowledge base to obtain a pre-trained second image segmentation model, wherein the priori medical knowledge base includes medical images annotated with location information and morphological feature information of the target area.

6. The method according to claim 1, characterized in that The method further comprises: Inputting the second data set into the optimized and trained first image segmentation model to obtain label information corresponding to each medical image in the second data set; determining a Dice coefficient corresponding to each medical image in the second dataset based on label information manually annotated for each medical image in the second dataset and label information corresponding to each medical image in the second dataset obtained based on the optimized trained first image segmentation model; determining the segmentation accuracy of the optimized trained first image segmentation model according to the Dice coefficient corresponding to each medical image in the second data set; When the segmentation accuracy of the first image segmentation model after the optimized training does not meet the preset segmentation accuracy requirement, the first image segmentation model after the pre-training is optimized and trained using the first data set and a fourth data set consisting of medical images carrying the second label information.

7. The method according to any one of claims 1 to 6, characterized in that: The first image segmentation model is a U-Net model, and the second image segmentation model is a SAM2 model.

8. A training device for a medical image segmentation model, characterized in that: include: a pre-training module, configured to pre-train a first image segmentation model using a first data set to obtain a pre-trained first image segmentation model, wherein each training sample in the first data set is a medical image annotated with location information and morphological feature information of a target region; a first label information determination module, configured to input a second data set consisting of unlabeled medical images into the pre-trained first image segmentation model to obtain first label information corresponding to each medical image in the second data set, wherein the first label information includes location information and morphological feature information of the target area; a second label information determination module, configured to input a third data set consisting of medical images carrying the first label information into a second image segmentation model to obtain second label information corresponding to each medical image in the third data set, wherein the second label information includes optimized location information and morphological feature information of the target area; An optimization training module is used to optimize the pre-trained first image segmentation model using the first data set and a fourth data set consisting of medical images carrying the second label information to obtain the optimized trained first image segmentation model.

9. An electronic device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the computer program is used to control the processor to operate so as to execute the training method of the medical image segmentation model according to any one of claims 1 to 7.

10. A non-volatile computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the training method of the medical image segmentation model according to any one of claims 1 to 7 is implemented.

Images