基于U-Net网络与分水岭相结合的脑肿瘤分割

Brain Tumor Segmentation Based on U-Net Network and Watershed

有效的MRI脑肿瘤图像分割能够为医生对患者的诊断和治疗提供可靠的依据。虽然卷积神经网络在医学图像分割的领域取得了显著的进展,但脑部结构过于复杂,误分割率较高,MRI脑肿瘤图像分割仍存在许多不足。U-Net网络的全对称结构能够使其只需少量训练即可提取足够的特征,但由于U-Net网络每次卷积图像都会小一圈,导致上采样和下采样所还原的像素尺寸不一样,无法对肿瘤边缘进行准确地分割。为解决上述问题,提出了一种基于U-Net网络与分水岭相结合的脑肿瘤分割算法。利用U-Net网络模型中的跳跃连接结合压缩路径和扩展路径的特征图,得到对感兴趣区域的初分割,然后通过添加基于重建的开闭操作的分水岭算法优化初分割图边界,得到最终的分割结果。实验结果表明,该方法在准确率Acc、特异性Sp、灵敏度Sn和精度PPV上分别可达到0.896、0.988 8、0.905 9和0.980 1,能够有效地分割出病变区域,具有明显的研究价值。

Effective MRI brain tumor image segmentation can provide a reliable basis for doctors to diagnose and treat patients.Although convolutional neural network has made remarkable progress in the field of medical image segmentation,the brain structure is too complex and the false segmentation rate is high.There are still many deficiencies in MRI brain tumor image segmentation.The fully symmetrical structure of u-net network can extract enough features with only a small amount of training.However,because the convolution image size of U-Net network will be one circle smaller each time,the pixel size restored by up sampling and down sampling is different,and the tumor edge cannot be segmented accurately.In order to solve the above problems,a brain tumor segmentation algorithm based on the combination of U-Net network and watershed is proposed.Using the jump connection in the U-Net network model,the characteristic graphs of the underlying network and the high-level network are fused to obtain the initial segmentation of the region of interest,and then the boundary of the initial segmentation graph is optimized by adding the watershed algorithm based on the reconstruction opening and closing operation to obtain the final segmentation result.The experimental results show that the accuracy Acc,specificity Sp,sensitivity Sn and accuracy PPV of this method can reach 0.896,0.988 8,0.905 9and 0.980 1 respectively.It can effectively segment the lesion area and has obvious research value.