CN-122024018-A - Automatic segmentation method and device for three-dimensional medical image, electronic equipment and storage medium

CN122024018ACN 122024018 ACN122024018 ACN 122024018ACN-122024018-A

Abstract

The invention discloses an automatic segmentation method, device, electronic equipment and storage medium of a three-dimensional medical image, which comprise the steps of obtaining three-dimensional medical image data input by a target user at the front end of a network, and obtaining a containerized segmentation model and labeling task parameters determined by the target user; the method comprises the steps of generating a segmentation task object based on labeling task parameters and three-dimensional medical image data, adding the segmentation task object to a target task queue, processing the segmentation task object based on a containerized segmentation model to obtain an original three-dimensional segmentation mask, processing the original three-dimensional segmentation mask according to category mapping rules to obtain a target segmentation result, and storing the target segmentation result for a target user to access through a network front end. Based on the technical scheme, the three-dimensional medical image data mechanical energy is segmented at the server to obtain the segmentation result, so that the centralization and the service of the calculation force are realized, the deployment and the maintenance of the model are uniformly carried out, and the manpower resource consumption of the client is obviously reduced.

Inventors

YUAN RONG
Shi Shuyue
YE YONGFANG

Assignees

上海影禾医脉智能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260408

Claims (10)

1. An automatic segmentation method of a three-dimensional medical image is characterized by comprising the following steps: Acquiring three-dimensional medical image data input by a target user at the front end of a network, and acquiring a containerized segmentation model and labeling task parameters determined by the target user, wherein the labeling task parameters comprise category mapping rules and filtering conditions; generating a segmentation task object based on the labeling task parameters and the three-dimensional medical image data, and adding the segmentation task object to a target task queue; processing the segmentation task object based on the containerized segmentation model to obtain an original three-dimensional segmentation mask; And processing the original three-dimensional segmentation mask according to the category mapping rule to obtain a target segmentation result, and storing the target segmentation result so as to enable the target user to access through the network front end.
2. The method of claim 1, wherein adding the split task object to a target task queue comprises: marking the segmented task object as a low-priority task, and adding the marked segmented task object to a priority task queue; and extracting the segmentation task object from the priority task queue under the condition that the high priority task in the priority task queue is empty and the occupancy rate of the computing power resource is lower than a preset threshold value.
3. The method of claim 2, wherein processing the segmentation task object based on the containerized segmentation model results in an original three-dimensional segmentation mask, comprising: Acquiring medical image tag information of three-dimensional medical image data in the segmentation task object, and comparing the medical image tag information with the filtering condition; And under the condition that the medical image label information meets the filtering condition, calling a containerized segmentation model to segment the three-dimensional medical image data, so as to obtain an original three-dimensional segmentation mask output by the containerized segmentation model.
4. The method of claim 3, wherein said comparing said medical image label information to said filtering condition comprises: Extracting metadata information of the three-dimensional medical image from the medical image tag information, wherein the metadata information comprises layer thickness, layer spacing and an inspection part tag; Extracting a corresponding preset threshold range from the filtering condition based on the examination part label, wherein the preset threshold range comprises a layer thickness threshold range and an interlayer spacing threshold range; And comparing the layer thickness and the layer spacing with the preset threshold range to determine whether the medical image label information meets the filtering condition.
5. The method of claim 3, wherein invoking the containerized segmentation model to segment the three-dimensional medical image data comprises: Inputting the three-dimensional medical image data into a shared image encoder in a containerized segmentation model, and extracting a multi-scale feature map, wherein the shared image encoder comprises at least four groups of three-dimensional residual convolution modules; Processing the multi-scale feature map based on an automatic segmentation network to obtain an automatic anatomical structure label, wherein the automatic segmentation network comprises a three-dimensional up-sampling transpose convolution layer and jump connection; Based on an interactive segmentation network, combining the multi-scale feature map and prompt information input by a user, obtaining an interactive segmentation mask, wherein the interactive segmentation network comprises a point embedded mapping layer and a cross attention module; and fusing the interactive segmentation mask and the automatic anatomical structure label to obtain the original three-dimensional segmentation mask.
6. The method of claim 1, wherein processing the original three-dimensional segmentation mask according to the class mapping rule results in a target segmentation result, comprising: determining pixel matrixes corresponding to the original three-dimensional segmentation mask, and identifying a matrix source category of each pixel matrix; And carrying out pixel replacement on each pixel matrix based on the category mapping rule and the matrix source category to obtain the target segmentation result.
7. The method of claim 1, wherein storing the target segmentation result comprises: Compressing the target segmentation result and establishing an association relationship with a data identifier of the three-dimensional medical image data; Storing the compressed target segmentation result into a cloud hierarchical storage, wherein the cloud hierarchical storage comprises a hot storage layer, a warm storage layer and a cold storage layer; Updating the state of the target segmentation result to be accessible, and when the target user initiates a viewing request through the network front end, retrieving the target segmentation result from the cloud hierarchical storage according to the data identification and pushing the target segmentation result to the network front end.
8. An automatic segmentation apparatus for three-dimensional medical images, comprising: The data acquisition module is used for acquiring three-dimensional medical image data input by a target user at the front end of the network, and acquiring a containerized segmentation model and labeling task parameters determined by the target user, wherein the labeling task parameters comprise category mapping rules and filtering conditions; The task generation module is used for generating a segmentation task object based on the labeling task parameters and the three-dimensional medical image data and adding the segmentation task object to a target task queue; The image segmentation module is used for processing the segmentation task object based on the containerized segmentation model to obtain an original three-dimensional segmentation mask; And the data storage module is used for processing the original three-dimensional segmentation mask according to the category mapping rule to obtain a target segmentation result, and storing the target segmentation result so as to enable the target user to access through the network front end.
9. An electronic device, the electronic device comprising: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the method of automatic segmentation of a three-dimensional medical image according to any one of claims 1-7.
10. A computer readable storage medium storing computer instructions for causing a processor to perform the method of automatic segmentation of a three-dimensional medical image according to any one of claims 1-7.

Description

Automatic segmentation method and device for three-dimensional medical image, electronic equipment and storage medium Technical Field The present invention relates to the field of image processing technologies, and in particular, to an automatic segmentation method and apparatus for three-dimensional medical images, an electronic device, and a storage medium. Background In order to improve the recognition capability of medical images, sub-images corresponding to each key position are obtained after image segmentation is carried out on the acquired medical images, so that quick and accurate medical image recognition is realized according to the sub-images. However, in the existing medical image segmentation method, 3D segmentation of medical images has the main problems of (1) high hardware cost, complex deployment and maintenance due to the fact that local high-performance computers are relied on, and computational effort is difficult to fully utilize, (2) low automation degree, manual or semi-automatic labeling due to severe dependence on professionals, low efficiency, and becomes a bottleneck of AI model training, (3) model deployment and maintenance dependence on manpower, the professional personnel participation is needed for deploying and updating models at clients, and additional manpower resource consumption and management cost are brought, and (4) poor coordination and expandability, the localization system is difficult to support multi-user and multi-task parallel processing and flexible scheduling, and the model updating and algorithm iteration period is long. Disclosure of Invention The invention provides an automatic segmentation method, device, electronic equipment and storage medium for three-dimensional medical images, wherein three-dimensional medical images and task parameters input by a user are acquired at the front end of a network, segmentation is carried out on the three-dimensional medical images input by the user under the condition of meeting preset conditions, segmentation results are generated, and the centralization and the service of calculation power are realized, so that strong 3D segmentation calculation power distributed according to needs is provided for the user, local hardware differences are shielded, the deployment, the update and the maintenance of a model are uniformly responsible by a service center, the manpower resource consumption of the client is obviously reduced, the latest segmentation model with optimal performance can be provided for the user in time, and the efficient, collaborative and extensible automatic medical image processing is realized. According to an aspect of the present invention, there is provided an automatic segmentation method of a three-dimensional medical image, including: acquiring three-dimensional medical image data input by a target user at the front end of a network, and acquiring a containerized segmentation model and labeling task parameters determined by the target user, wherein the labeling task parameters comprise category mapping rules and filtering conditions; Generating a segmentation task object based on the labeling task parameters and the three-dimensional medical image data, and adding the segmentation task object to a target task queue; processing the segmentation task object based on the containerized segmentation model to obtain an original three-dimensional segmentation mask; and processing the original three-dimensional segmentation mask according to the category mapping rule to obtain a target segmentation result, and storing the target segmentation result for a target user to access through the front end of the network. According to another aspect of the present invention, there is provided an automatic segmentation apparatus for three-dimensional medical images, including: The data acquisition module is used for acquiring three-dimensional medical image data input by a target user at the front end of the network, and acquiring a containerized segmentation model and labeling task parameters determined by the target user, wherein the labeling task parameters comprise category mapping rules and filtering conditions; The task generating module is used for generating a segmentation task object based on the labeling task parameters and the three-dimensional medical image data and adding the segmentation task object to the target task queue; the image segmentation module is used for processing the segmentation task object based on the containerized segmentation model to obtain an original three-dimensional segmentation mask; And the data storage module is used for processing the original three-dimensional segmentation mask according to the category mapping rule to obtain a target segmentation result, and storing the target segmentation result so as to be accessed by a target user through the front end of the network. According to another aspect of the present invention, there is provided an electronic apparatus including: