Fast pseudo-CT synthesis from MRI T1-weighted images using a patch-based approach

• Main Authors: Alcain, E. (Author), Montemayor, A. S. (Author), Rozenholc, Y. (Author), Lopez Herraiz, Joaquin (Contributor), Hernandez-Tamames, J. A. (Contributor), Adalsteinsson, Elfar (Contributor), Wald, Lawrence (Contributor), Malpica, N. (Contributor), Torrado-Carvajal, A. (Contributor)
• Other Authors: Madrid-MIT M+Visión Consortium (Contributor), Harvard-MIT Program in Health Sciences and Technology (Contributor), Massachusetts Institute of Technology. Research Laboratory of Electronics (Contributor)
• Format: Article
• Language: English
• Published: Society of Photo-Optical Instrumentation Engineers (SPIE), 2017-07-17T15:35:54Z.
• Subjects: Article
• Online Access: Get fulltext

 MRI-based bone segmentation is a challenging task because bone tissue and air both present low signal intensity on MR images, making it difficult to accurately delimit the bone boundaries. However, estimating bone from MRI images may allow decreasing patient ionization by removing the need of patient-specific CT acquisition in several applications. In this work, we propose a fast GPU-based pseudo-CT generation from a patient-specific MRI T1-weighted image using a group-wise patch-based approach and a limited MRI and CT atlas dictionary. For every voxel in the input MR image, we compute the similarity of the patch containing that voxel with the patches of all MR images in the database, which lie in a certain anatomical neighborhood. The pseudo-CT is obtained as a local weighted linear combination of the CT values of the corresponding patches. The algorithm was implemented in a GPU. The use of patch-based techniques allows a fast and accurate estimation of the pseudo-CT from MR T1-weighted images, with a similar accuracy as the patient-specific CT. The experimental normalized cross correlation reaches 0.9324±0.0048 for an atlas with 10 datasets. The high NCC values indicate how our method can accurately approximate the patient-specific CT. The GPU implementation led to a substantial decrease in computational time making the approach suitable for real applications.