计算机断层扫描(computed tomography,CT)可应用于前列腺癌的检查诊断,但是它对软组织结构对比度不高,因此很难从中分割病变;而核磁共振成像(nuclear magnetic resonance imaging,MRI)具有较高的对比度,能为病变提供丰富的影像信息。为...计算机断层扫描(computed tomography,CT)可应用于前列腺癌的检查诊断,但是它对软组织结构对比度不高,因此很难从中分割病变;而核磁共振成像(nuclear magnetic resonance imaging,MRI)具有较高的对比度,能为病变提供丰富的影像信息。为了提升CT图像的前列腺分割精度,本文提出一种新的基于深度学习的多模态U形网络图像分割模型MM-unet,充分运用MRI图像与CT图像间信息互补的特点。具体地,首先运用迁移学习思想分别训练MRI与CT图像的初始分割模型,然后通过设计一种新型的多模态损失函数MM-Loss,建立不同模态分割模型之间的联系,联合训练基于MRI与CT图像的MM-unet。为验证所提模型MM-unet的有效性,我们在某合作医院提供的Prostate数据集上进行了实验,实验结果表明,与U-net方法相比,MM-unet能够获得高出3个百分点Dice的CT图像分割精度。展开更多
针对测地线活动轮廓(geodesic active contour,GAC)模型轮廓演化速度慢的问题,构造一个区域灰度相似性信息项,对GAC模型的能量泛函进行改进,加快轮廓演化速度,将其用于肺部薄扫CT(computed tomography)图像序列中肺实质的自动分割。采...针对测地线活动轮廓(geodesic active contour,GAC)模型轮廓演化速度慢的问题,构造一个区域灰度相似性信息项,对GAC模型的能量泛函进行改进,加快轮廓演化速度,将其用于肺部薄扫CT(computed tomography)图像序列中肺实质的自动分割。采用基于Nystrom逼近的谱聚类算法分割CT图像序列中间位置CT中的肺实质,计算其灰度均值与标准差,构造区域灰度相似性信息项,以分割好的肺实质轮廓作为初始轮廓,分别从上下两个方向采用改进了能量泛函的GAC模型实现其它切片中肺实质的分割。实验结果表明,该方法能够较好实现肺实质的自动分割,与医师分割结果的重合率可达94.83%,时间消耗较少。展开更多
A new method is presented for the segmentation of pulmonary parenchyma. The proposed method is based on the area calculation of different objects in the image. The main purpose of the proposed algorithm is the segment...A new method is presented for the segmentation of pulmonary parenchyma. The proposed method is based on the area calculation of different objects in the image. The main purpose of the proposed algorithm is the segment of the lungs images from the computer tomography(CT) images. The original image is binarized using the bit-plane slicing technique and among the different images the best binarized image is chosen. After binarization, the labeling is done and the area of each label is calculated from which the next level of binarized image is obtained. Then, the boundary tracing algorithm is applied to get another level of binarized image. The proposed method is able to extract lung region from the original images. The experimental results show the significance of the proposed method.展开更多
文摘计算机断层扫描(computed tomography,CT)可应用于前列腺癌的检查诊断,但是它对软组织结构对比度不高,因此很难从中分割病变;而核磁共振成像(nuclear magnetic resonance imaging,MRI)具有较高的对比度,能为病变提供丰富的影像信息。为了提升CT图像的前列腺分割精度,本文提出一种新的基于深度学习的多模态U形网络图像分割模型MM-unet,充分运用MRI图像与CT图像间信息互补的特点。具体地,首先运用迁移学习思想分别训练MRI与CT图像的初始分割模型,然后通过设计一种新型的多模态损失函数MM-Loss,建立不同模态分割模型之间的联系,联合训练基于MRI与CT图像的MM-unet。为验证所提模型MM-unet的有效性,我们在某合作医院提供的Prostate数据集上进行了实验,实验结果表明,与U-net方法相比,MM-unet能够获得高出3个百分点Dice的CT图像分割精度。
文摘针对测地线活动轮廓(geodesic active contour,GAC)模型轮廓演化速度慢的问题,构造一个区域灰度相似性信息项,对GAC模型的能量泛函进行改进,加快轮廓演化速度,将其用于肺部薄扫CT(computed tomography)图像序列中肺实质的自动分割。采用基于Nystrom逼近的谱聚类算法分割CT图像序列中间位置CT中的肺实质,计算其灰度均值与标准差,构造区域灰度相似性信息项,以分割好的肺实质轮廓作为初始轮廓,分别从上下两个方向采用改进了能量泛函的GAC模型实现其它切片中肺实质的分割。实验结果表明,该方法能够较好实现肺实质的自动分割,与医师分割结果的重合率可达94.83%,时间消耗较少。
基金supported (in part) by research funding from Chosun University, Korea, 2013
文摘A new method is presented for the segmentation of pulmonary parenchyma. The proposed method is based on the area calculation of different objects in the image. The main purpose of the proposed algorithm is the segment of the lungs images from the computer tomography(CT) images. The original image is binarized using the bit-plane slicing technique and among the different images the best binarized image is chosen. After binarization, the labeling is done and the area of each label is calculated from which the next level of binarized image is obtained. Then, the boundary tracing algorithm is applied to get another level of binarized image. The proposed method is able to extract lung region from the original images. The experimental results show the significance of the proposed method.
