© Springer International Publishing Switzerland 2016 326 E. Kyriacou et al. (eds.), XIV Mediterranean Conference on Medical and Biological Engineering and Computing 2016, IFMBE Proceedings 57, DOI: 10.1007/978-3-319-32703-7_64 3D Reconstruction of Tubular Structures from Three Orthogonal MRI Projections Junmo An 1 , Mahmut Unan 1 , Karen Chin 2 , Dipan J. Shah 2 , Andrew G. Webb 3 , Ioannis Seimenis 4 , and Nikolaos V. Tsekos 1 1 Medical Robotics Laboratory, University of Houston, Houston, TX, USA 2 Methodist DeBakey Cardiology Associates, Houston Methodist, Houston, TX, USA 3 C.J. Gorter Center for High Field MRI, Leiden University Medical Center, Leiden, Netherlands 4 Laboratory of Medical Physics, Department of Medicine, Democritus University of Thrace, Alexandroupolis, Greece Abstract— This study presents a novel method for 3D imag- ing of tubular structures, such as blood vessels and catheters. With this method, the 3D object is reconstructed from three mutually exclusive orthogonal projections of the same volume. This triplanar projection imaging (TPI) was evaluated on a phantom filled with T1-shortening, Gd-based contrast agent embedded into a matrix. The projected structures were seg- mented out on each one of the projections and a backprojection algorithm was used to generate a rendering of the tubular structure in 3D. The accuracy of the method was investigated by comparing the centerline of the 3D tubular structure generated from TPI with the centerline extracted from a multislice set of images of the same volume. The two tubular structures were well matched in 3D. With further optimization and reduction of acquisition time, this method can be used for 3D fast imaging of interventional tools or blood vessels with applications in interventional MRI. Keywords— Magnetic resonance imaging, contrast en- hanced, 3D reconstruction, volumetric imaging, projection imaging. I. INTRODUCTION Since the superb soft-tissue contrast of the MRI is unpar- alleled by other imaging modalities, many endeavors have been devoted to the development of MR-guided interven- tions [1, 2]. The multiplanar capabilities of MRI provide multislice and/or volumetric imaging for excellent visuali- zation of the interventional procedure region of interest. Although MRI is a very versatile and powerful diagnostic tool modality, it cannot provide, due to its inherent relatively low sensitivity, instantaneous speed and high spatial resolution afforded with x-ray based modalities such as fluoroscopy and angiography. 3D reconstruction from two projections is well studied in biplane fluoroscopy [3, 4]. In fluoroscopic imaging cases where the two projections that are not orthogonal to each other and the centers of the corresponding field of views (FOVs) are not identical, a series of transformations is applied to coregister the two projections to a common coordinate sys- tem of reference. With MRI, the multiple projections are al- ready inherently coregistered, and the center and size of the FOVs are identical. The inherent coregistration and coinci- dence of the FOV of the multiple projections result in a rather simple geometric processing. With 2D multislice MRI, how- ever, the speed of imaging and 3D visualization is rather slow. Multiple projection imaging methods were proposed for 3D visualization of active catheters [5-8]. In this study, in order for planning and guiding interventional devices, we investi- gate fast volumetric reconstruction of targeted contrast en- hanced (CE) tubular structures, such as catheters and blood vessels, with a simultaneously collecting method - triplanar projection imaging (TPI). II. MATERIALS AND METHODS A. Triplanar Projection Imaging (TPI) 3D reconstruction of an imaged structure entails two in- dependent but interrelated processes. First, the acquisition of the projections and, second, the reconstruction of the structure in the 3D space. Originating from its inherent ‘true-3D’ capabilities, MRI provides certain benefits in collecting data and facilitating the 3D reconstruction of the imaged structure from its projections: (a) The inherent coregistration of the spatial encoding on the three projec- tions makes axes assignment and matching straightforward. (b) The orientation of the imaged volume can be set to any arbitrary plane relative to the structure which can be used to better resolve it, thus reducing the number of needed projec- tions and computational resources required from the recon- struction algorithm. (c) The angle between the projection planes can be adjusted to any desired value and is not lim- ited to orthogonal. Figure 1(a) illustrates the operation of the method that collects three projections by setting the thickness of the slice in each acquisition (Fig. 1(b)) equal to the FOV encoded by the phase and readout gradients. B. TPI Acquisition Without loss of generality, we hypothesize that the inter- est is to visualize the tubular structure of choice without any background signal. This can further simplify the segmenta- tion and the 3D reconstruction of the structure. In the exper- imental part, therefore, we filled the tubular structure with T1-shortening Gd agent and used magnetization preparation to the implemented sequences to achieve long T1 species background suppression.