基于U-net的直肠癌智能分割

Intelligent segmentation of rectal cancer based on U-net

近年来,随着医学影像与大数据的发展,通过人工智能来处理图像,从而给医生带来辅助性的诊断参考成为广大学者的研究热点。文章利用U-net网络对直肠癌CT影像进行智能分割,加入了图像增强、批归一化等技巧缓解过拟合现象,并通过多次实验确定最佳初始学习速率和卷积核数目,在验证集上的Dice系数达到0.9329。实验表明,U-Net对小数据集的医学图像分割有很好效果,对正负样本极度偏斜的数据集,使用Dice系数可以准确地衡量分割的相似程度。

In recent years,with the development of medical imaging and big data,the use of artificial intelligence to process images to bring auxiliary diagnostic references to doctors has become a research hotspot for scholars.In this paper,the U-net network is used to intelligently segment rectal cancer CT images.Techniques such as image enhancement and batch normalization are added to alleviate the over-fitting phenomenon.Through multiple experiments,the optimal initial learning rate and the number of convolution kernels are determined,and the Dice coefficient on the validation set reaches 0.9329.Experiments show that U-Net has a good effect on medical image segmentation of small data sets.In the extremely skewed data set of positive and negative samples,the Dice coefficient can accurately measure the similarity of the segmentation.