Complex blood vessel segmentation method based on neural network
A neural network and blood vessel technology, applied in neural learning methods, biological neural network models, neural architectures, etc., can solve problems such as class imbalance and data scarcity, achieve low cost, improve segmentation efficiency, and be suitable for popularization and use
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0036]Seefigure 1 A method of dividing a complex blood vessel based on a neural network, comprising the following steps:
[0037]Step 1: Preprocessing the original medical image image data set, labeled a mask diagram corresponding to each original image to obtain a homemade brain vascular data Coro and ELE, respectively 40, respectively, each pair of pictures including an original map and Corresponding marker mask diagram;
[0038]Step 2: Transform the original image and its corresponding mask map through the PATCH-based data enhancement method input into the split network, perform depth network model training, obtained PATCHES is the size of the random 48 * 48 resolution from the original image Variety;
[0039]Step 3: Output prediction results according to the training parameters obtained after network training, and perform results assessment.
[0040]In this embodiment, only a small amount of data set can complete the segmentation of the vascular image. Based on the depth neural network netw...
Embodiment 2
[0042]The present embodiment is the same as that of the examples, and the particulars are as follows:
[0043]The pretreatment of the step 1 includes the following operations:
[0044]1-1: Adjust the resolution of the original image of 565 × 584 pixels;
[0045]1-2: Binarize the original image to get a binary map;
[0046]1-3: Mark the two-value map to get a binary marker mask, 1 represents a blood vessel structure, 0 represents a picture background;
[0047]1-4: Half of the data set will be used for training, half is used for testing, and then extract 10% from the test concentration for verification.
[0048]The network model training of the step 2 includes the following operations:
[0049]2-1: The split network consists of two consecutive U-Net, each U-NET is 5 floors, and the original image in the data set and the corresponding mask map are enhanced after the PATCH-based data enhancement method. Training, eventually get predicted, complete network structurefigure 2 ;
[0050]2-2: The first U-name is to...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More