CN-122023445-A - Brain choroid plexus vesicle segmentation method

Abstract

The invention discloses a brain choroid plexus vesicle segmentation method, which comprises the steps of enhancing the signal contrast of the whole choroid plexus by using a high-resolution T1 enhanced image, and then adopting two steps of the whole choroid plexus segmentation and vesicle segmentation to realize stable segmentation of vesicles, wherein in the vesicle segmentation process, since the vesicles are filled with cerebrospinal fluid, the invention sets a vesicle segmentation reference threshold by using the difference between the entity (high signal) of the choroid plexus and the cerebrospinal fluid (low signal) in the T1 enhanced image, and realizes the separation of vesicle voxels by low-pass filtering. The method can divide the vesicle structure of the cerebral choroid plexus and calculate the volume of the vesicle structure, and the cerebral choroid plexus vesicle volume can be used as an image marker for auxiliary diagnosis of nerve diseases such as Alzheimer disease and the like.

Inventors

• HUANG PEIYU
• ZHEN ZHIMING
• ZHANG RUITING
• LIN MIAO

Assignees

• 浙江大学医学院附属第二医院

Dates

Publication Date

20260512

Application Date

20260204

Claims (4)

1. 1. A method for dividing a brain choroid plexus vesicle, comprising the steps of: 1) Performing format conversion and offset field correction on the high-resolution T1 enhanced craniocerebral magnetic resonance image, eliminating intensity non-uniformity, and obtaining an optimized reference image; 2) Dividing the optimized reference image into a choroid cluster whole, and outputting a choroid cluster binarization image, namely a choroid cluster mask 1; 3) The choroid plexus mask 1 is expanded outwards by 1 voxel to obtain a choroid plexus mask 2, the optimized reference image is processed by using a whole brain structure segmentation algorithm to obtain a lateral ventricle mask, the choroid plexus mask 2 is removed from the lateral ventricle mask to obtain a pure cerebrospinal fluid mask, an average signal and a standard deviation thereof are extracted from the pure cerebrospinal fluid mask, and the average signal +/-2 x standard deviation is used as a vesicle segmentation reference threshold; 4) Multiplying the optimized reference image with a choroid plexus mask 1 to obtain a choroid plexus image, performing low-pass filtering on the choroid plexus image by using the vesicle segmentation reference threshold to obtain a cerebrospinal fluid signal voxel of a choroid plexus region, and storing the cerebrospinal fluid signal voxel of the choroid plexus region as a binarized image to obtain a choroid plexus vesicle candidate region mask image; 5) Detecting and separating different connected areas in the mask image of the choroid cluster candidate area according to the space connectivity among different voxel points, secondly, calculating the number of voxels of each connected area, removing the connected areas with the number of voxels smaller than 4, finally, merging and storing all the remained connected areas in the mask image of the choroid cluster candidate area into a binarized image to generate a final choroid cluster image, and multiplying the number of voxels with the signal of 1 in the choroid cluster image by the single voxel volume of the choroid cluster image to obtain the choroid cluster volume.
2. 2. The method for dividing the cerebral choroid plexus vesicles according to claim 1 is characterized in that in the step 2), the optimized reference image is divided into the whole choroid plexus, specifically, the method is realized based on a cerebral choroid plexus whole division model, the training method of the cerebral choroid plexus whole division model comprises the steps of carrying out choroid tissue sketching on the optimized reference image by a professional radiologist, sketching the whole choroid plexus, including a choroid plexus entity and vesicles as gold standard mask images, taking 2/3 as training sets and 1/3 as test sets from all gold standard mask images, constructing a U-Net neural network, training the choroid plexus whole division model by using training set data, and testing the accuracy of the whole choroid plexus division by using the test set data, so that the whole choroid plexus division model can be obtained.
3. 3. The method for dividing the cerebral choroid plexus vesicles according to claim 1, wherein in the step 5), different connected regions in the choroid plexus vesicle candidate region mask image are detected and separated according to the spatial connectivity between different voxel points, and then the number of voxels of each connected region is calculated, and the connected region with the number of voxels smaller than 4 is removed, and the specific method comprises the steps of: Defining a space connectivity rule as plane adjacency, determining the connection relation between voxels in the mask image of the choroid cluster candidate region according to the plane adjacency, traversing each voxel in the mask image of the choroid cluster candidate region according to the connection relation between the voxels, determining the connectivity between the voxels and surrounding voxels, distributing a unique object label for all interconnected foreground voxels, outputting the image as a label array, wherein the voxels in each interconnected region have the same label value, the background is kept to be 0, traversing the label array according to the total number of labels, creating a new blank binary image for each non-zero label, positioning all voxels equal to the label value in the label array as the foreground, obtaining the mask of each interconnected region, and finally, calculating the overall prime number of each interconnected region, and removing the interconnected region with the prime number smaller than 4.
4. 4. A brain choroid plexus vesicle volume as an imaging marker for the assisted diagnosis of neurological diseases, characterized in that said brain choroid plexus vesicle volume is obtained based on a brain choroid plexus vesicle segmentation method according to any one of claims 1-3.

Description

Brain choroid plexus vesicle segmentation method Technical Field The invention relates to the field of medical imaging and image analysis, in particular to a brain choroid plexus vesicle segmentation method. Background The human cerebral choroid plexus is present in the lateral, third and fourth ventricles, and is composed of a monolayer of epithelial cells surrounding blood vessels and matrix components, with a highly folded surface, which is covered with microvilli. The choroid plexus has several important functions (1) the choroid plexus is the main secretion source of cerebrospinal fluid, the daily production amount is about 500ml, the cerebrospinal fluid circulation is important for removing metabolic waste and maintaining the brain microenvironment steady state, (2) the choroid plexus epithelial cells form a blood-cerebrospinal fluid barrier (BCSFB) through tight connection, the central nervous system and the peripheral circulation are isolated, and the invasion of exogenous pathogens into the central nervous system can be prevented, (3) a plurality of immune cells including Kolmer cells, macrophages and the like exist on the surface and in the matrix of the choroid plexus. These cells are able to sense pathogens and inflammatory signals in the cerebrospinal fluid and regulate inflammation in the brain by releasing inflammatory factors, attracting inflammatory cell migration. Taken together, the above-described functions of the choroid plexus have a crucial role in maintaining brain health. The damage to the choroid plexus may cause the above important dysfunction, and cause imbalance of brain homeostasis, which is closely related to the occurrence and development of various neurological diseases such as Alzheimer's disease and multiple sclerosis. In clinical images, assessment of the choroid plexus is focused primarily on the overall volume of the choroid plexus. While recent studies based on high-resolution, multi-modal images have shown that there are vesicle structures within the choroid plexus, whose internal signals resemble cerebrospinal fluid. The number of choroid plexus vesicles increased with age, and the number of vesicles was significantly increased in patients with alzheimer's disease compared to the control group. Therefore, choroid plexus vesicles can serve as important imaging markers of choroid plexus degeneration and brain aging. At present, some rough segmentation methods such as Freesurfer segmentation and Gaussian Mixture Models segmentation methods are used for the whole choroid plexus, but no segmentation and quantification method for the choroid plexus vesicles in high-resolution structural images is found at home and abroad. Because the choroid plexus vesicle is small in volume, the wall is thin, signals are weak, the number and morphology of individuals are extremely different, and the difficulty of segmentation calculation is high. At present, no method capable of automatically dividing choroid plexus vesicles and calculating vesicle volumes exists at home and abroad. Disclosure of Invention In order to solve the technical problems, the application provides a method for dividing cerebral choroid plexus vesicles. Aiming at the characteristics of thin choroid plexus vesicle wall, weak signal and easy coverage by cerebrospinal fluid signal, the application firstly uses high-resolution T1 to enhance the signal contrast of the whole image reinforced choroid plexus. Secondly, because the number, the morphology and the distribution of the vesicles have large differences, a simple neural network model cannot acquire enough information, the application adopts two steps of choroid plexus integral segmentation and vesicle segmentation to realize stable segmentation of the vesicles. Third, since the vesicle is filled with cerebrospinal fluid, the application uses the difference between the choroid plexus entity (high signal) and cerebrospinal fluid (low signal) in the T1 enhanced image to set a vesicle segmentation reference threshold, and realizes the separation of vesicle voxels through low-pass filtering. The method can be used for dividing the cerebral choroid plexus vesicles and calculating the volume of the cerebral choroid plexus vesicles. The invention is realized by adopting the following technical scheme: A method for dividing a brain choroid plexus vesicle, comprising the steps of: 1. Image preprocessing And performing format conversion and offset field correction on the high-resolution T1 enhanced craniocerebral magnetic resonance image, eliminating intensity non-uniformity, obtaining an optimized reference image, and providing a unified template for subsequent analysis. The high-resolution T1 enhanced craniocerebral magnetic resonance image is an isotropic T1 enhanced image which is acquired by a field intensity magnetic resonance instrument of 3T or more and has resolution of 0.5 mm. 2. Integral division of choroid plexus And (3) performing choroid plexus integral segment