3-D Reconstruction in Canonical Co-Ordinate Space From Arbitrarily Oriented 2-D Images

Abstract

Limited capture range, and the requirement to provide high quality initialization for optimization-based 2-D/3-D image registration methods, can significantly degrade the performance of 3-D image reconstruction and motion compensation pipelines. Challenging clinical imaging scenarios, which contain significant subject motion, such as fetal in-utero imaging, complicate the 3-D image and volume reconstruction process. In this paper, we present a learning-based image registration method capable of predicting 3-D rigid transformations of arbitrarily oriented 2-D image slices, with respect to a learned canonical atlas co-ordinate system. Only image slice intensity information is used to perform registration and canonical alignment, no spatial transform initialization is required. To find image transformations, we utilize a convolutional neural network architecture to learn the regression function capable of mapping 2-D image slices to a 3-D canonical atlas space. We extensively evaluate the effectiveness of our approach quantitatively on simulated magnetic resonance imaging (MRI), fetal brain imagery with synthetic motion and further demonstrate qualitative results on real fetal MRI data where our method is integrated into a full reconstruction and motion compensation pipeline. Our learning based registration achieves an average spatial prediction error of 7 mm on simulated data and produces qualitatively improved reconstructions for heavily moving fetuses with gestational ages of approximately 20 weeks. Our model provides a general and computationally efficient solution to the 2-D/3-D registration initialization problem and is suitable for real-time scenarios.

Computing CNN Loss and Gradients for Pose Estimation with Riemannian Geometry

“…This is demonstrated on the current state-of-the-art for pose estimation, PoseNet, where we show that our method achieves similar performance as the carefully tuned approximation used in [7]. We also show significant improvements for medical image pose estimation and outperform the state-of-the-art in this domain [4,3]. Acknowledgements: Supported by the Wellcome Trust IEH Award [102431] and Nvidia.…”

“…We derive appropriate gradients that are required for CNN back propagation. Our method couples the translation and rotation parameters, and regresses them simultaneously as one parameter on the Lie algebra se (3). We show that our loss function is agnostic to the architecture by training different CNNs and can effectively predict poses that are comparable to state-of-the-art methods.…”

“…Alternatively, [8,16] reparameterised the pose parameters as projected co-ordinates on a 2D view plane. This was similarly performed by [4,3] with Anchor Points, where three arbitrary selected reference points on a 2D plane define the plane's location in 3D space. Using the L2-norm to calculate the Euclidean distance between a predicted projection co-ordinate and the ground truth projection co-ordinate is appropriate, as the L2-norm is the appropriate metric.…”

Evaluation of MRI to Ultrasound Registration Methods for Brain Shift Correction: The CuRIOUS2018 Challenge

“…Deep learning has been successfully applied to image registration [29,30], but in nonlinear image registration tasks with high accuracy requirement, such as brain shift correction, further exploration is still needed [31]. The two submissions that used DL in this challenge were from Sun et al [11], who did not participate to the second phase, and Zhong et al [17] (team FAX), who ranked first in the results reported on the training database.…”

Real-Time Deep Pose Estimation With Geodesic Loss for Image-to-Template Rigid Registration

“…Prior work on slice-to-volume registration in fetal MRI has shown a strong need for regularization and initialization of slice transformations through hierarchical registration [23], [24] or state-space motion modeling [25]. Learning-based methods have been recently used to improve prediction of slice locations in fetal MRI [26], [27] and fetal ultrasound [28]. In [26], [27] anchor-point slice parametrization was used along with the Euclidean loss function based on [29] to predict slice positions and reconstruct fetal MRI in canonical space.…”