An improved optical flow tracking technique for real-time MR-guided beam therapies in moving organs

Abstract

Magnetic resonance (MR) guided high intensity focused ultrasound and external beam radiotherapy interventions, which we shall refer to as beam therapies/interventions, are promising techniques for the non-invasive ablation of tumours in abdominal organs. However, therapeutic energy delivery in these areas becomes challenging due to the continuous displacement of the organs with respiration. Previous studies have addressed this problem by coupling high-framerate MR-imaging with a tracking technique based on the algorithm proposed by Horn and Schunck (H and S), which was chosen due to its fast convergence rate and highly parallelisable numerical scheme. Such characteristics were shown to be indispensable for the real-time guidance of beam therapies. In its original form, however, the algorithm is sensitive to local grey-level intensity variations not attributed to motion such as those that occur, for example, in the proximity of pulsating arteries.In this study, an improved motion estimation strategy which reduces the impact of such effects is proposed. Displacements are estimated through the minimisation of a variation of the H and S functional for which the quadratic data fidelity term was replaced with a term based on the linear L(1)norm, resulting in what we have called an L(2)-L(1) functional.The proposed method was tested in the livers and kidneys of two healthy volunteers under free-breathing conditions, on a data set comprising 3000 images equally divided between the volunteers. The results show that, compared to the existing approaches, our method demonstrates a greater robustness to local grey-level intensity variations introduced by arterial pulsations. Additionally, the computational time required by our implementation make it compatible with the work-flow of real-time MR-guided beam interventions.To the best of our knowledge this study was the first to analyse the behaviour of an L(1)-based optical flow functional in an applicative context: real-time MR-guidance of beam therapies in moving organs.

A framework for continuous target tracking during MR-guided high intensity focused ultrasound thermal ablations in the abdomen

“…This decision was made due to its prior successful in-vivo validation in previous independent studies [47–50]. However, the drawback of such an approach is that the optical flow algorithm has its own shortcomings which affect its performance (discussed at length in [37]). Thus, the in-vivo errors reported for the multi-modal algorithm stem from both its own mis-registrations and the errors in the silver standard itself.…”

“…For this reason, image registration was performed using two classes of algorithms: mono- and multi-modal [36]. In the scope of this study, a mono-modal registration method, namely the optical flow algorithm [37], was employed when the compared images had the same contrast weighting (more specifically during (RS # 1) and (RS # 2)). For images acquired with different contrasts, a multi-modal method based on the modality independent neighborhood descriptors (MIND) [38] was used (more specifically during (RS # 3)).…”

“…As previously mentioned, depending on whether the images being registered were acquired with the same MR-contrast weighting or not, one of two registration algorithms were employed: the optical flow [37] or the MIND [38] algorithm. The methods were chosen due to their fast numerical schemes, low number of input parameters and their capability to provide deformations on a voxel-by-voxel/pixel-by-pixel basis, characteristics which make them particularly attractive for medical image registration.…”

“…The optical flow algorithm followed the implementation described by Zachiu et al [37]. The approach provides the deformation between two images I and J , as the minimizer of the following functional: …”

Fluid filling of the digestive tract for improved proton resonance frequency shift‐based MR thermometry in the pancreas

“…For both referenceless and multibaseline MRT, a registration step is necessary. For this step an optical flow‐based technique was used . The incoming images were first registered to a gated reference image.…”