The segmentation of organs at risk is a crucial and time-consuming step in radiotherapy planning. Good automatic methods can significantly reduce the time clinicians have to spend on this task. Due to its variability in shape and low contrast to surrounding structures, segmenting the parotid gland is challenging. Motivated by the recent success of deep learning, we study the use of two-dimensional (2-D), 2-D ensemble, and three-dimensional (3-D) U-Nets for segmentation. The mean Dice similarity to ground truth is ∼0.83 for all three models. A patch-based approach for class balancing seems promising for false-positive reduction. The 2-D ensemble and 3-D U-Net are applied to the test data of the 2015 MICCAI challenge on head and neck autosegmentation. Both deep learning methods generalize well onto independent data (Dice 0.865 and 0.88) and are superior to a selection of model- and atlas-based methods with respect to the Dice coefficient. Since appropriate reference annotations are essential for training but often difficult and expensive to obtain, it is important to know how many samples are needed for training. We evaluate the performance after training with different-sized training sets and observe no significant increase in the Dice coefficient for more than 250 training cases.
The segmentation of target structures and organs at risk is a crucial and very time-consuming step in radiotherapy planning. Good automatic methods can significantly reduce the time clinicians have to spend on this task. Due to its variability in shape and often low contrast to surrounding structures, segmentation of the parotid gland is especially challenging. Motivated by the recent success of deep learning, we study different deep learning approaches for parotid gland segmentation. Particularly, we compare 2D, 2D ensemble and 3D U-Net approaches and find that the 2D U-Net ensemble yields the best results with a mean Dice score of 0.817 on our test data. The ensemble approach reduces false positives without the need for an automatic region of interest detection. We also apply our trained 2D U-Net ensemble to segment the test data of the 2015 MICCAI head and neck auto-segmentation challenge. With a mean Dice score of 0.861, our classifier exceeds the highest mean score in the challenge. This shows that the method generalizes well onto data from independent sites. Since appropriate reference annotations are essential for training but often difficult and expensive to obtain, it is important to know how many samples are needed to properly train a neural network. We evaluate the classifier performance after training with differently sized training sets (50–450) and find that 250 cases (without using extensive data augmentation) are sufficient to obtain good results with the 2D ensemble. Adding more samples does not significantly improve the Dice score of the segmentations.
KEYWORDS: Image registration, Image segmentation, Error analysis, Magnetic resonance imaging, 3D image processing, 3D acquisition, Control systems, Motion estimation, Detection and tracking algorithms, Computed tomography
We present a technique to rectify nonrigid registrations by improving their group-wise consistency, which is a widely used unsupervised measure to assess pair-wise registration quality. While pair-wise registration methods cannot guarantee any group-wise consistency, group-wise approaches typically enforce perfect consistency by registering all images to a common reference. However, errors in individual registrations to the reference then propagate, distorting the mean and accumulating in the pair-wise registrations inferred via the reference. Furthermore, the assumption that perfect correspondences exist is not always true, e.g., for interpatient registration. The proposed consistency-based registration rectification (CBRR) method addresses these issues by minimizing the group-wise inconsistency of all pair-wise registrations using a regularized least-squares algorithm. The regularization controls the adherence to the original registration, which is additionally weighted by the local postregistration similarity. This allows CBRR to adaptively improve consistency while locally preserving accurate pair-wise registrations. We show that the resulting registrations are not only more consistent, but also have lower average transformation error when compared to known transformations in simulated data. On clinical data, we show improvements of up to 50% target registration error in breathing motion estimation from four-dimensional MRI and improvements in atlas-based segmentation quality of up to 65% in terms of mean surface distance in three-dimensional (3-D) CT. Such improvement was observed consistently using different registration algorithms, dimensionality (two-dimensional/3-D), and modalities (MRI/CT).
In this paper we present a novel post-processing technique to detect and correct inconsistency-based errors in non-rigid registration. While deformable registration is ubiquitous in medical image computing, assessing its quality has yet been an open problem. We propose a method that predicts local registration errors of existing pairwise registrations between a set of images, while simultaneously estimating corrected registrations. In the solution the error is constrained to be small in areas of high post-registration image similarity, while local registrations are constrained to be consistent between direct and indirect registration paths. The latter is a critical property of an ideal registration process, and has been frequently used to asses the performance of <i>registration algorithms</i>. In our work, the consistency is used as a <i>target criterion</i>, for which we efficiently find a solution using a linear least-squares model on a coarse grid of registration control points. We show experimentally that the local errors estimated by our algorithm correlate strongly with true registration errors in experiments with known, dense ground-truth deformations. Additionally, the estimated corrected registrations consistently improve over the initial registrations in terms of average deformation error or TRE for different registration algorithms on both simulated and clinical data, independent of modality (MRI/CT), dimensionality (2D/3D) and employed primary registration method (demons/Markov-randomfield).
This paper studies improving joint segmentation and registration by introducing auxiliary labels for anatomy that has similar appearance to the target anatomy while not being part of that target. Such auxiliary labels help avoid false positive labelling of non-target anatomy by resolving ambiguity. A known registration of a segmented atlas can help identify where a target segmentation should lie. Conversely, segmentations of anatomy in two images can help them be better registered. Joint segmentation and registration is then a method that can leverage information from both registration and segmentation to help one another. It has received increasing attention recently in the literature. Often, merely a single organ of interest is labelled in the atlas. In the presense of other anatomical structures with similar appearance, this leads to ambiguity in intensity based segmentation; for example, when segmenting individual bones in CT images where other bones share the same intensity profile. To alleviate this problem, we introduce automatic generation of additional labels in atlas segmentations, by marking similar-appearance non-target anatomy with an auxiliary label. Information from the auxiliary-labeled atlas segmentation is then incorporated by using a novel coherence potential, which penalizes differences between the deformed atlas segmentation and the target segmentation estimate. We validated this on a joint segmentation-registration approach that iteratively alternates between registering an atlas and segmenting the target image to find a final anatomical segmentation. The results show that automatic auxiliary labelling outperforms the same approach using a single label atlasses, for both mandibular bone segmentation in 3D-CT and corpus callosum segmentation in 2D-MRI.