本文提出了一种新的无监督域自适应框架,称为协同图像和特征自适应(SIFA),以有效地解决域移位的问题。在最近的深度学习研究中,领域适应已经成为一个重要的热点,目的是在将神经网络应用于新的测试领域时恢复性能退化。我们提出的SIFA是一个优雅的学习图表,它从图像和特征的角度展示了协同融合的适应性。特别是,同时转换跨域的图像外观并增强提取的特征对于分割任务的域不变性。两个视角共享特征编码器层,以便在端到端学习过程中掌握它们的相互利益。在不使用目标域的任何注释的情况下,我们的统一模型的学习是由对抗性损失引导的,并且从各个方面采用了多个鉴别器。我们已经通过对心脏结构的模态医学图像分割的挑战性应用广泛地验证了我们的方法。实验结果表明,我们的SIFA模型将性能从17.2%恢复到73.0%,并且通过显着的优势超越了最先进的方法。
translated by 谷歌翻译
深度神经网络(DNN)在各种医学图像分析任务中取得了巨大成功。然而,这些成就必不可少地依赖于精确注释的数据集。如果使用带噪声标记的图像,训练过程将立即遇到困难,导致非理想的分类器。考虑到注释质量需要很多专业知识,这个问题在医学领域更为重要。在本文中,我们提出了一个有效的迭代学习框架,用于噪声标记的医学图像分类,以对抗缺乏高质量的注释医学数据。具体地,提出了一种在线不确定性样本挖掘方法,以消除噪声标记图像的干扰。接下来,我们设计一个样本加权策略,以保留正确标记的硬样本的有用性。我们提出的方法在皮肤病变分类任务上得到验证,取得了很好的效果。
translated by 谷歌翻译
在这项工作中,我们报告了结合IEEE国际生物医学成像研讨会(ISBI)2016和国际医学影像计算机辅助干预会议(MICCAI)2017年组织的肝肿瘤分割基准(LITS)的设置和结果。将24种有效的最先进的肝脏和肝脏肿瘤分段算法应用于一组131个计算机断层扫描(CT)体积,具有不同类型的肿瘤对比度水平(高强度/低强度),组织异常(转移瘤)大小和不同程度的病变。已提交的算法已在70个未公开的卷上进行了测试。该数据集是与七家医院和研究机构合作创建的,由三位独立的放射科医师手动审查。我们发现没有一种算法对肝脏和肿瘤表现最佳。最佳肝脏分割算法的Dice评分为0.96(MICCAI),而对于肿瘤分割,最佳算法评估为0.67(ISBI)和0.70(MICCAI)。 LITS图像数据和手动注释继续通过在线评估系统公开提供,作为持续的基准测试资源。
translated by 谷歌翻译
深度卷积网络已经证明了各种医学图像计算任务的最新表现。利用来自不同模态的图像进行相同的分析任务具有临床优势。然而,深度模型对具有不同分布的测试数据的泛化能力仍然是一个主要挑战。在本文中,我们提出了PnPAdaNet(即插即用对抗域适应网络),用于在医学图像的不同模态之间分割网络,例如,MRI和CT。我们建议通过以无监督的方式对齐源域和目标域的特征空间来解决重要的域移位。具体地,域适配模块灵活地替换源网络的早期编码器层,并且域之间共享更高层。具有对抗性学习,我们建立了两个鉴别器,其输入分别是多级特征和预测的分割掩模。我们已经在非配对MRI和CT中对心脏结构分割的域适应方法进行了验证。综合消融研究的实验结果证明了我们提出的PnP-AdaNet的优良功效。此外,我们为心脏数据集引入了一个新的基于无监督跨模态域适应任务的基准。我们将公开提供我们的代码和数据库,旨在促进未来对医学成像这一具有挑战性且重要的研究课题的研究。
translated by 谷歌翻译
The analysis of glandular morphology within colon histopathology images is an important step in determining the grade of colon cancer. Despite the importance of this task, manual segmentation is laborious, time-consuming and can suffer from subjectivity among pathologists. The rise of computational pathology has led to the development of automated methods for gland segmentation that aim to overcome the challenges of manual segmentation. However, this task is non-trivial due to the large variability in glandular appearance and the difficulty in differentiating between certain glandular and non-glandular histological structures. Furthermore, a measure of uncertainty is essential for diagnostic decision making. To address these challenges, we propose a fully convolutional neural network that counters the loss of information caused by max-pooling by reintroducing the original image at multiple points within the network. We also use atrous spatial pyramid pooling with varying dilation rates for preserving the resolution and multi-level aggregation. To incorporate uncertainty, we introduce random transformations during test time for an enhanced segmentation result that simultaneously generates an uncertainty map, highlighting areas of ambiguity. We show that this map can be used to define a metric for disregarding predictions with high uncertainty. The proposed network achieves state-of-the-art performance on the GlaS challenge dataset and on a second independent colorectal adenocarcinoma dataset. In addition, we perform gland instance segmentation on whole-slide images from two further datasets to highlight the generalisability of our method. As an extension, we introduce MILD-Net + for simultaneous gland and lumen segmentation, to increase the diagnostic power of the network.
translated by 谷歌翻译
卷积网络(ConvNets)在各种具有挑战性的视觉任务中取得了巨大成功。但是,当遇到域转换时,ConvNets的性能会降低。在生物医学图像分析领域中,域适应性更具有挑战性,而生物医学图像分析中的跨模态数据具有很大不同的分布。鉴于注释医学数据特别昂贵,受监督的转移学习方法并不是最佳的。在本文中,我们提出了一种无监督的域适应框架,其具有用于跨模态生物医学图像分割的对抗性学习。具体来说,我们的模型基于用于像素预测的扩散卷积网络。此外,我们构建了即插即用域适配模块(DAM)来映射与源域特征空间对齐的目标输入特征。建立域批评模块(DCM)以区分两个域的特征空间。在不使用任何targetdomain标签的情况下,通过对抗性损失优化DAM和DCM。我们提出的方法通过使用MRI图像训练的ConvNet适应心脏结构分割的非配对CT数据来验证,并且获得了非常有希望的结果。
translated by 谷歌翻译
肝癌是导致癌症死亡的主要原因之一。为了帮助医生进行肝细胞癌的诊断和治疗计划,在临床实践中需要一种准确的自动肝脏和肿瘤分割方法。最近,完全卷积神经网络(FCN),包括2D和3D FCN,在许多体积图像分割中充当了骨干。然而,2D卷积不能充分利用沿第三维的空间信息,而3D卷绕遭受高计算成本和GPU内存消耗。为了解决这些问题,我们提出了一种新颖的混合密集连接的UNet(H-DenseUNet),它由2D DenseUNet组成,用于有效地提取片内特征,以及3Dcounterpart,用于在肝脏和肿瘤的自动上下文算法的精神下分层聚集体积上下文。分割。我们以端到端的方式制定H-DenseUNet的学习过程,其中可以通过混合特征融合(HFF)层联合优化切片表示和片间特征。我们对MICCAI 2017肝肿瘤分割(LiTS)挑战和3DIRCADb数据集的数据集进行了广泛的评估。我们的方法在肿瘤的分割结果方面优于其他现有技术,并且即使使用单一模型也能实现非常有竞争力的分割效果。
translated by 谷歌翻译
Automatic detection of pulmonary nodules in thoracic computed tomography (CT)scans has been an active area of research for the last two decades. However,there have only been few studies that provide a comparative performanceevaluation of different systems on a common database. We have therefore set upthe LUNA16 challenge, an objective evaluation framework for automatic noduledetection algorithms using the largest publicly available reference database ofchest CT scans, the LIDC-IDRI data set. In LUNA16, participants develop theiralgorithm and upload their predictions on 888 CT scans in one of the twotracks: 1) the complete nodule detection track where a complete CAD systemshould be developed, or 2) the false positive reduction track where a providedset of nodule candidates should be classified. This paper describes the setupof LUNA16 and presents the results of the challenge so far. Moreover, theimpact of combining individual systems on the detection performance was alsoinvestigated. It was observed that the leading solutions employed convolutionalnetworks and used the provided set of nodule candidates. The combination ofthese solutions achieved an excellent sensitivity of over 95% at fewer than 1.0false positives per scan. This highlights the potential of combining algorithmsto improve the detection performance. Our observer study with four expertreaders has shown that the best system detects nodules that were missed byexpert readers who originally annotated the LIDC-IDRI data. We released thisset of additional nodules for further development of CAD systems.
translated by 谷歌翻译
Automatic liver segmentation from CT volumes is a crucial prerequisite yetchallenging task for computer-aided hepatic disease diagnosis and treatment. Inthis paper, we present a novel 3D deeply supervised network (3D DSN) to addressthis challenging task. The proposed 3D DSN takes advantage of a fullyconvolutional architecture which performs efficient end-to-end learning andinference. More importantly, we introduce a deep supervision mechanism duringthe learning process to combat potential optimization difficulties, and thusthe model can acquire a much faster convergence rate and more powerfuldiscrimination capability. On top of the high-quality score map produced by the3D DSN, a conditional random field model is further employed to obtain refinedsegmentation results. We evaluated our framework on the public MICCAI-SLiver07dataset. Extensive experiments demonstrated that our method achievescompetitive segmentation results to state-of-the-art approaches with a muchfaster processing speed.
translated by 谷歌翻译
由于不准确的检测和识别,自主车辆可能做出错误的决定。因此,智能车辆可以将自己的数据与其他车辆相结合,提高感知能力,从而提高检测精度和驾驶安全性。然而,多车协同感知要求现实世界场景的整合和原始传感器数据交换的流量远远超过现有车载网络的带宽。据我们所知,我们是第一个对原始数据级合作感知进行研究的人。提高自驾系统的检测能力。在这项工作中,依靠LiDAR 3D点云,我们完成了从连接车辆的不同位置和角度收集的传感器数据。提出了一种基于点云的三维物体检测方法,用于对齐点云的多样性。 KITTI和我们收集的数据集的实验结果表明,所提出的系统通过扩展感知区域优于感知,提高了检测精度并促进了增强结果。最重要的是,我们证明可以通过现有的车载网络技术传输用于协作感知的pointclouds数据。
translated by 谷歌翻译