Unlock instant, AI-driven research and patent intelligence for your innovation.

Lightweight brain tumor segmentation method based on depth separable convolution

A brain tumor, lightweight technology, applied in the field of lightweight brain tumor segmentation, can solve the problems of high hardware resource requirements, affecting segmentation accuracy, large memory usage, etc., achieve high precision, reduce computing consumption, and balance training intensity effect

Pending Publication Date: 2021-04-20
TIANJIN UNIV
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The 2D network segmentation speed is fast and the computing resources are low, but usually multiple output images are mechanically stitched into a 3D image as the segmentation result, and the context information from adjacent slices is not effectively used, and problems such as jaggedness and faults often occur. Segmentation accuracy
The 3D convolutional network has sufficient spatial feature extraction and good segmentation effect. However, due to its huge memory usage and high requirements on hardware resources, it is usually necessary to make compromises in the network structure by sacrificing accuracy or training speed. given memory budget

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Lightweight brain tumor segmentation method based on depth separable convolution
  • Lightweight brain tumor segmentation method based on depth separable convolution
  • Lightweight brain tumor segmentation method based on depth separable convolution

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0024] The first step, image preprocessing

[0025] The experimental data of the present invention comes from the BraTS 2018 data set, in which the training set contains 210 high-grade glioma patient samples, 75 low-grade glioma patient samples, and the verification set contains 66 unlabeled patient samples. In the training set, each sample contains 4 MRI modalities and the real-value label map manually marked by multiple professional physicians. Through a series of data enhancement methods such as random interception of brain tumor images with a size of 240×240×155 to 128× 128×128, randomly flip in the axial, coronal, sagittal and other directions, randomly rotate in the range of [-10°, 10°], etc., to avoid overfitting problems caused by insufficient training set data. The processed brain tumor MRI4 modality is input in the form of 4 channels and trained through the encoder-decoder network.

[0026] The second step is to construct an improved U-Net and design a weighted mixe...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention provides a lightweight brain tumor segmentation method based on depth separable convolution. The method comprises the following steps: brain tumor MRI preprocessing; according to the method, a deep convolutional network being constructed and trained, 3D U-Net being modified firstly, convolution in the network being replaced with depth separable convolution, an improved weighting loss function being used for increasing the segmentation accuracy and accelerating the model convergence speed, and the method comprises the following steps: constructing the improved 3D U-Net; designing a weighted mixed loss function and training the network; and carrying out post-processing on the segmentation result.

Description

technical field [0001] The invention belongs to the field of image segmentation based on the combination of computer vision and medical image processing, and relates to a lightweight brain tumor segmentation method. Background technique [0002] Brain tumor is an abnormal cell group that grows in the brain, and it is a kind of tumor that seriously endangers the life of patients. According to the global cancer statistics report of A Cancer Journal for Clinicians (CA), as of 2019, there were about 238,000 new cases of brain tumors, accounting for about 1.3% of all new cases, and about 17.7 deaths. million, accounting for about 2.9% of all cancer deaths. Brain tumors are mainly divided into primary brain tumors that form in the brain or in nerves originating from the brain and secondary brain tumors that metastasize to the brain from other parts of the body. The most common primary brain tumors in adults are primary central nervous system lymphoma and glioma, in which glioma ...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06T7/11G06T7/00G06N3/04G06N3/08
Inventor 赵奕名李锵关欣
Owner TIANJIN UNIV