Jane M. Blackall

A Generic Framework for Non-rigid Registration Based on Non-uniform Multi-level Free-Form Deformations

This work presents a framework for non-rigid registration which extends and generalizes a previously developed technique by Rueckert et al. [1]. We combine multi-resolution optimization with free-form deformations (FFDs) based on multi-level B-splines to simulate a non-uniform control point distribution. We have applied this to a number of different medical registration tasks to demonstrate its wide applicability, including interventional MRI brain tissue deformation compensation, breathing motion compensation in liver MRI, intra-modality inter-modality registration of pre-operative brain MRI to CT electrode implant data, and inter-subject registration of brain MRI. Our results demonstrate that the new algorithm can successfully register images with an improved performance, while achieving a significant reduction in run-time.

A study of the motion and deformation of the heart due to respiration

This paper describes a quantitative assessment of respiratory motion of the heart and the construction of a model of respiratory motion. Three-dimensional magnetic resonance scans were acquired on eight normal volunteers and ten patients. The volunteers were imaged at multiple positions in the breathing cycle between full exhalation and full inhalation while holding their breath. The exhalation volume was segmented and used as a template to which the other volumes were registered using an intensity-based rigid registration algorithm followed by nonrigid registration. The patients were imaged at inhale and exhale only. The registration results were validated by visual assessment and consistency measurements indicating subvoxel registration accuracy. For all subjects, we assessed the nonrigid motion of the heart at the right coronary artery, right atrium, and left ventricle. We show that the rigid-body motion of the heart is primarily in the craniocaudal direction with smaller displacements in the right-left and anterior-posterior directions; this is in agreement with previous studies. Deformation was greatest for the free wall of the right atrium and the left ventricle; typical deformations were 3-4 mm with deformations of up to 7 mm observed in some subjects. Using the registration results, landmarks on the template surface were mapped to their correct positions through the breathing cycle. Principal component analysis produced a statistical model of the motion and deformation of the heart. We discuss how this model could be used to assist motion correction.

An Image Registration Approach to Automated Calibration for Freehand 3D Ultrasound

This paper describes an image registration approach to calibration for freehand three-dimensional (3D) ultrasound. If a conventional ultrasound probe is tracked using a position sensor, and the relation between this sensor and the two-dimensional (2D) image plane is known, the resulting set of B-scans may be correctly compounded into an image volume. Calibration is the process of determining the transformation (rotation, translation, and optionally image scaling) that maps B-mode image slice coordinates to points in the coordinate system of the tracking sensor mounted on the ultrasound probe. A set of 2D ultrasound images of a calibration phantom is obtained using a tracked ultrasound probe. Calibration is performed by searching for the calibration parameters that maximise the similarity between a model of the calibration phantom, which can be an image volume or a geometrical model, and the ultrasound images transformed into the coordinate space of the phantom. Validation of this calibration method is performed using a gelatin phantom. Measures of the calibration reproducibility, reconstruction precision and reconstruction accuracy are presented for this technique, and compared to those obtained using a conventional cross-wire phantom. Registration-based calibration is shown to be a rapid and accurate method of automatic calibration for freehand 3D ultrasound.

A Statistical Model of Respiratory Motion and Deformation of the Liver

This paper presents a statistical model of the respiratory motion and deformation of the liver. Individual models were made for seven volunteers using a series of MR images taken throughout the breathing cycle. One image was selected as a template and the others were registered to this using an intensity-based non-rigid registration algorithm. The resulting free-form transformations allowed us to map landmarks defined on the template image into their correct positions throughout the breathing cycle. Principal component analysis of these landmarks was used to produce a statistical model of motion and deformation. Results showing typical motion and deformation for a single volunteer are presented.

A Stochastic Iterative Closest Point Algorithm (stochastICP)

We present a modification to the iterative closest point algorithm which improves the algorithms robustness and precision. At the start of each iteration, before point correspondence is calculated between the two feature sets, the algorithm randomly perturbs the point positions in one feature set. These perturbations allow the algorithm to move out of some local minima to find a minimum with a lower residual error. The size of this perturbation is reduced during the registration process. The algorithm has been tested using multiple starting positions to register three sets of data: a surface of a femur, a skull surface and a registration to hepatic vessels and a liver surface. Our results show that, if local minima are present, the stochastic ICP algorithm is more robust and is more precise than the standard ICP algorithm.

Bayesian Estimation of Intra-operative Deformation for Image-Guided Surgery Using 3-D Ultrasound

This paper describes the application of Bayesian theory to the problem of compensating for soft tissue deformation to improve the accuracy of image-guided surgery. A triangular surface mesh segmented from a pre-operative image is used as the input to the algorithm, and intra-operatively acquired ultrasound data compounded into a 3-D volume is used to guide the deformation process. Prior probabilities are defined for the boundary points of the segmented structure based on knowledge of the direction of gravity, the position of the surface of the surgical scene, and knowledge of the tissue properties. The posterior probabilities of the locations of each of the boundary points are then maximised according to Bayes’ theorem. A regularisation term is included to constrain deformation to the global structure of the object.

The susceptibility of IMRT dose distributions to intrafraction organ motion: An investigation into smoothing filters derived from four dimensional computed tomography data

This study investigated the sensitivity of static planning of intensity-modulated beams (IMBs) to intrafraction deformable organ motion and assessed whether smoothing of the IMBs at the treatment-planning stage can reduce this sensitivity. The study was performed with a 4D computed tomography (CT) data set for an IMRT treatment of a patient with liver cancer. Fluence profiles obtained from inverse-planning calculations on a standard reference CT scan were redelivered on a CT scan from the 4D data set at a different part of the breathing cycle. The use of a nonrigid registration model on the 4D data set additionally enabled detailed analysis of the overall intrafraction motion effects on the IMRT delivery during free breathing. Smoothing filters were then applied to the beam profiles within the optimization process to investigate whether this could reduce the sensitivity of IMBs to intrafraction organ motion. In addition, optimal fluence profiles from calculations on each individual phase of the breathing cycle were averaged to mimic the convolution of a static dose distribution with a motion probability kernel and assess its usefulness. Results from nonrigid registrations of the CT scan data showed a maximum liver motion of 7mm in superior-inferior direction for this patient. Dose-volume histogram (DVH) comparison indicated a systematic shift when planning treatment on a motion-frozen, standard CT scan but delivering over a full breathing cycle. The ratio of the dose to 50% of the normal liver to 50% of the planning target volume (PTV) changed up to 28% between different phases. Smoothing beam profiles with a median-window filter did not overcome the substantial shift in dose due to a difference in breathing phase between planning and delivery of treatment. Averaging of optimal beam profiles at different phases of the breathing cycle mainly resulted in an increase in dose to the organs at risk (OAR) and did not seem beneficial to compensate for organ motion compared with using a large margin. Additionally, the results emphasized the need for 4D CT scans when aiming to reduce the internal margin (IM). Using only a single planning scan introduces a systematic shift in the dose distribution during delivery. Smoothing beam profiles either based on a single scan or over the different breathing phases was not beneficial for reducing this shift.

Using combined x-ray and MR imaging for prostate I-125 post-implant dosimetry : phantom validation and preliminary patient work

Post-implantation dosimetry is an important element of permanent prostate brachytherapy. This process relies on accurate localization of implanted seeds relative to the surrounding organs. Localization is commonly achieved using CT images, which provide suboptimal prostate delineation. On MR images, conversely, prostate visualization is excellent but seed localization is imprecise due to distortion and susceptibility artefacts. This paper presents a method based on fused MR and x-ray images acquired consecutively in a combined x-ray and MRI interventional suite. The method does not rely on any explicit registration step but on a combination of system calibration and tracking. A purpose-built phantom was imaged using MRI and x-rays, and the images were successfully registered. The same protocol was applied to three patients where combining soft tissue information from MRI with stereoscopic seed identification from x-ray imaging facilitated post-implant dosimetry. This technique has the potential to improve on dosimetry using either CT or MR alone.

A comparison of internal target volume definition by limited four-dimensional computed tomography, the addition of patient-specific margins, or the addition of generic margins when planning radical radiotherapy for lymph node-positive non-small cell lung cancer

AIMS Radical radiotherapy for stage II/III non-small cell lung cancer (NSCLC) includes the primary tumour and positive mediastinal lymph nodes in the clinical target volume (CTV). These move independently of each other in magnitude and direction during respiration. To prevent a geographical miss, a generic margin is usually added to the CTV to create an internal target volume (ITV). Previous studies have investigated the use of additional breath-hold computed tomography to generate patient-specific ITVs for primary tumours alone. We used a similar technique to investigate the generation of patient-specific and generic ITVs for CTVs that include mediastinal lymph nodes. MATERIALS AND METHODS Thirteen patients with node-positive NSCLC had two limited end-tidal breath-hold computed tomography scans in addition to their planning computed tomography. The CTV was segmented in each scan and a rigid registration was carried out on the vertebral columns to align them. Different methods for generating an ITV were then analysed. RESULTS Generic margins provided >95% mean coverage of the reference ITV. However, with the exception of 1cm expansion margins, there were cases of inadequate coverage (<95%) for each ITV. With increasing ITV margins there was a small increase in reference ITV coverage, but at the expense of a large increase in the volume of normal tissue within the ITV. DISCUSSION For stage II/III NSCLC, ITV generation by the addition of a generic margin is not optimal. It can result in both geographical miss and excessive irradiation of normal tissue in the same treatment plan. A simple method for producing a patient-specific ITV is to co-register end-tidal breath-hold computed tomography scans to the planning scan. CONCLUSIONS Further work is required to determine whether end-tidal breath-hold scans are representative of the anatomy at the limits of tidal respiration. Planning strategies are also needed to account for breathing cycle variation during a course of radiotherapy.

Computational Models In Image Guided Interventions

In image-guided surgery and image-directed therapy a plan based on pre-procedure imaging is registered to the patient in the operating or treatment room using a 3D spatial localizer. The plan can be used as long as the transformation between plan and patient remains valid. Most systems use a rigid-body transformation restricting guidance to bony structures (e.g. orthopaedic surgery or maxillo-facial surgery) or structures that are rigidly related to bone (e.g. neurosurgery). Fully 3D intra-operative imaging such as interventional MR allows image guidance to be extended to structures that move or deform during an intervention. However, this technology is expensive, interferes significantly with standard surgical protocols and requires computationally expensive non-rigid registration of the plan to the current patient scan. This talk will describe four examples where computational models of motion and anatomy are combined with 2D intra-operative imaging to extend the scope of image directed methods. In the first, image guided neurosurgery, we show how intra-operative imaging may account for distortion caused by the intervention itself. In two further applications - percutaneous ablation of metastatic liver disease and external beam radiotherapy of the lung - we show how computational models of motion might be used in conjunction with a therapy plan to guide the intervention. In the final example, selected from orthopaedic surgery, we show recent advances that demonstrate how a statistical shape model generated from example 3D images, can be used to provide image guidance without any pre-operative 3D imaging