A fully connected crf cascaded fcn and k-means brain tumor segmentation algorithm

Abstract

The invention relates to a brain tumor segmentation algorithm based on the combination of deep learning and traditional segmentation algorithms, in particular to a fully connected CRF cascaded FCN and K-mean brain tumor segmentation algorithm. DenseCRF matches all pixels in the original image with each pixel in the segmentation result of the FCN algorithm, finds pixels with the same attributes, supplements and smooths the input, improves the detail information of the segmentation result, and improves the segmentation accuracy. At the same time, in different segmentation algorithms, the segmentation standards will also be different. By integrating the deep learning algorithm FCN with different segmentation standards and the traditional segmentation algorithm K-means clustering, the segmentation results obtained by the algorithms based on different segmentation standards are mutually exclusive. Supplement to make the segmentation result closer to the real segmented image. Thus, the brain tumor MRI image can be segmented more accurately, and a more accurate tumor image can be provided for the high-precision repeatable measurement and evaluation of the brain tumor MRI image.

Description

technical field [0001] The invention relates to a brain tumor segmentation algorithm based on the combination of deep learning and traditional segmentation algorithms, in particular to a cascaded fully convolutional neural network (Fully Connected Conditional Random Field, DenseCRF) post-processing based on The brain tumor segmentation algorithm fused with Fully Convolution Neural Network (FCN) and K-means clustering algorithm model can be used to segment brain tumor MRI images more accurately, and provide more accurate and repeatable measurement and evaluation of brain tumor MRI images. Accurate tumor images. Background technique [0002] In order to use neuroimaging to evaluate the manifestations of brain tumors and the effect of treatment before and after treatment, it is inevitable to carry out high-precision and repeatable measurement and evaluation of the lesion area. Therefore, accurate segmentation of medical images is a necessary step for measurement and evaluation....

AI Technical Summary

Problems solved by technology

[0003] With the rapid development of computer hardware, in the medical field, computer-aided medical diagnosis methods have become an important research field in medical imaging, diagnostic radiation, and computer science. Among them, deep learning, which is better than traditional algorithms, has entered the medical field. , has achieved many excellent results, but the tumor data set has more complex irregular shape features than most natural image data sets, and the network characteristic of FCN is that it can obtain detailed bottom-level information, and the collection of upper-level information does not Not ideal, so the processing of edge details is rough and lacks optimization

Images