John S. H. Baxter

Hierarchical max-flow segmentation framework for multi-atlas segmentation with Kohonen self-organizing map based Gaussian mixture modeling

The incorporation of intensity, spatial, and topological information into large-scale multi-region segmentation has been a topic of ongoing research in medical image analysis. Multi-region segmentation problems, such as segmentation of brain structures, pose unique challenges in image segmentation in which regions may not have a defined intensity, spatial, or topological distinction, but rely on a combination of the three. We propose a novel framework within the Advanced segmentation tools (ASETS)(2), which combines large-scale Gaussian mixture models trained via Kohonen self-organizing maps, with deformable registration, and a convex max-flow optimization algorithm incorporating region topology as a hierarchy or tree. Our framework is validated on two publicly available neuroimaging datasets, the OASIS and MRBrainS13 databases, against the more conventional Potts model, achieving more accurate segmentations. Each component is accelerated using general-purpose programming on graphics processing Units to ensure computational feasibility.

Stationary wavelet transform for under-sampled MRI reconstruction.

In addition to coil sensitivity data (parallel imaging), sparsity constraints are often used as an additional lp-penalty for under-sampled MRI reconstruction (compressed sensing). Penalizing the traditional decimated wavelet transform (DWT) coefficients, however, results in visual pseudo-Gibbs artifacts, some of which are attributed to the lack of translation invariance of the wavelet basis. We show that these artifacts can be greatly reduced by penalizing the translation-invariant stationary wavelet transform (SWT) coefficients. This holds with various additional reconstruction constraints, including coil sensitivity profiles and total variation. Additionally, SWT reconstructions result in lower error values and faster convergence compared to DWT. These concepts are illustrated with extensive experiments on in vivo MRI data with particular emphasis on multiple-channel acquisitions.

Registration of 3D shapes under anisotropic scaling

PurposeSeveral medical imaging modalities exhibit inherent scaling among the acquired data: The scale in an ultrasound image varies with the speed of sound and the scale of the range data used to reconstruct organ surfaces is subject to the scanner distance. In the context of surface-based registration, these scaling factors are often assumed to be isotropic, or as a known prior. Accounting for such anisotropies in scale can potentially dramatically improve registration and calibrations procedures that are essential for robust image-guided interventions.MethodsWe introduce an extension to the ordinary iterative closest point (ICP) algorithm, solving for the similarity transformation between point-sets comprising anisotropic scaling followed by rotation and translation. The proposed anisotropic-scaled ICP (ASICP) incorporate a novel use of Mahalanobis distance to establish correspondence and a new solution for the underlying registration problem. The derivation and convergence properties of ASICP are presented, and practical implementation details are discussed. Because the ASICP algorithm is independent of shape representation and feature extraction, it is generalizable for registrations involving scaling.ResultsExperimental results involving the ultrasound calibration, registration of partially overlapping range data, whole surfaces, as well as multi-modality surface data (intraoperative ultrasound to preoperative MR) show dramatic improvement in fiducial registration error.ConclusionWe present a generalization of the ICP algorithm, solving for a similarity transform between two point-sets by means of anisotropic scales, followed by rotation and translation. Our anisotropic-scaled ICP algorithm shares many traits with the ordinary ICP, including guaranteed convergence, independence of shape representation, and general applicability.

Motion magnification for endoscopic surgery

Endoscopic and laparoscopic surgeries are used for many minimally invasive procedures but limit the visual and haptic feedback available to the surgeon. This can make vessel sparing procedures particularly challenging to perform. Previous approaches have focused on hardware intensive intraoperative imaging or augmented reality systems that are difficult to integrate into the operating room. This paper presents a simple approach in which motion is visually enhanced in the endoscopic video to reveal pulsating arteries. This is accomplished by amplifying subtle, periodic changes in intensity coinciding with the patient’s pulse. This method is then applied to two procedures to illustrate its potential. The first, endoscopic third ventriculostomy, is a neurosurgical procedure where the floor of the third ventricle must be fenestrated without injury to the basilar artery. The second, nerve-sparing robotic prostatectomy, involves removing the prostate while limiting damage to the neurovascular bundles. In both procedures, motion magnification can enhance subtle pulsation in these structures to aid in identifying and avoiding them.

The Role of Augmented Reality in Training the Planning of Brain Tumor Resection

The environment in which a surgeons is trained profoundly effects their preferred method for visualizing patient images. While classical 2D viewing might be preferred by some older experts, the new generation of residents and novices has been raised navigating in 3D through video games, and are accustomed to seeing 3D reconstructions of the human anatomy. In this study, we evaluate the performance of different groups of users in 4 different visualization modalities (2D planes, orthogonal planes, 3D reconstruction and augmented reality). We hypothesize that this system will facilitate the spatio-visual abilities of individuals in terms of assessing patient-specific data, an essential requirement of many neurosurgical applications such as tumour resection. We also hypothesize that the difference between AR and the other modalities will be greater in the novice group. Our preliminary results indicate that AR is better or as good as other modalities in terms of performance.

Detection and visualization of dural pulsation for spine needle interventions

PurposeEpidural and spinal anesthesia are common procedures that require a needle to be inserted into the patient’s spine to deliver an anesthetic. Traditionally, these procedures were performed without image guidance, using only palpation to identify the correct vertebral interspace. More recently, ultrasound has seen widespread use in guiding spinal needle interventions. Dural pulsation is a valuable cue for finding a path through the vertebral interspace and for determining needle insertion depth. However, dural pulsation is challenging to detect and not perceptible in many cases. Here, a method for automatically detecting very subtle dural pulsation from live ultrasound video is presented.MethodsA periodic model is fit to the B-mode intenstity values through extended Kalman filtering. The fitted frequencies and amplitudes are used to detect and visualize dural pulsation. The method is validated retrospectively on synthetic and human video and used in real time on an interventional spinal phantom.ResultsThis method was capable of quickly identifying subtle dural pulsation and was robust to background noise and motion. The pulsation visualization reduced both the normalized path length and number of attempts required in a mock epidural procedure.ConclusionThis technique is able to localize the dura and help find a clear needle trajectory to the epidural space. It can be run in real time on commercial ultrasound systems and has the potential to improve ultrasound guidance of spine needle interventions.

Directed Acyclic Graph Continuous Max-Flow Image Segmentation for Unconstrained Label Orderings

Label ordering, the specification of subset–superset relationships for segmentation labels, has been of increasing interest in image segmentation as they allow for complex regions to be represented as a collection of simple parts. Recent advances in continuous max-flow segmentation have widely expanded the possible label orderings from binary background/foreground problems to extendable frameworks in which the label ordering can be specified. This article presents Directed Acyclic Graph Max-Flow image segmentation which is flexible enough to incorporate any label ordering without constraints. This framework uses augmented Lagrangian multipliers and primal–dual optimization to develop a highly parallelized solver implemented using GPGPU. This framework is validated on synthetic, natural, and medical images illustrating its general applicability.

Line fiducial material and thickness considerations for ultrasound calibration

Ultrasound calibration is a necessary procedure in many image-guided interventions, relating the position of tools and anatomical structures in the ultrasound image to a common coordinate system. This is a necessary component of augmented reality environments in image-guided interventions as it allows for a 3D visualization where other surgical tools outside the imaging plane can be found. Accuracy of ultrasound calibration fundamentally affects the total accuracy of this interventional guidance system. Many ultrasound calibration procedures have been proposed based on a variety of phantom materials and geometries. These differences lead to differences in representation of the phantom on the ultrasound image which subsequently affect the ability to accurately and automatically segment the phantom. For example, taut wires are commonly used as line fiducials in ultrasound calibration. However, at large depths or oblique angles, the fiducials appear blurred and smeared in ultrasound images making it hard to localize their cross-section with the ultrasound image plane. Intuitively, larger diameter phantoms with lower echogenicity are more accurately segmented in ultrasound images in comparison to highly reflective thin phantoms. In this work, an evaluation of a variety of calibration phantoms with different geometrical and material properties for the phantomless calibration procedure was performed. The phantoms used in this study include braided wire, plastic straws, and polyvinyl alcohol cryogel tubes with different diameters. Conventional B-mode and synthetic aperture images of the phantoms at different positions were obtained. The phantoms were automatically segmented from the ultrasound images using an ellipse fitting algorithm, the centroid of which is subsequently used as a fiducial for calibration. Calibration accuracy was evaluated for these procedures based on the leave-one-out target registration error. It was shown that larger diameter phantoms with lower echogenicity are more accurately segmented in comparison to highly reflective thin phantoms. This improvement in segmentation accuracy leads to a lower fiducial localization error, which ultimately results in low target registration error. This would have a profound effect on calibration procedures and the feasibility of different calibration procedures in the context of image-guided procedures.

Visual Enhancement of MR Angiography Images to Facilitate Planning of Arteriovenous Malformation Interventions

The primary purpose of medical image visualization is to improve patient outcomes by facilitating the inspection, analysis, and interpretation of patient data. This is only possible if the users’ perceptual and cognitive limitations are taken into account during every step of design, implementation, and evaluation of interactive displays. Visualization of medical images, if executed effectively and efficiently, can empower physicians to explore patient data rapidly and accurately with minimal cognitive effort. This article describes a specific case study in biomedical visualization system design and evaluation, which is the visualization of MR angiography images for planning arteriovenous malformation (AVM) interventions. The success of an AVM intervention greatly depends on the surgeon gaining a full understanding of the anatomy of the malformation and its surrounding structures. Accordingly, the purpose of this study was to investigate the usability of visualization modalities involving contour enhancement and stereopsis in the identification and localization of vascular structures using objective user studies. Our preliminary results indicate that contour enhancement, particularly when combined with stereopsis, results in improved performance enhancement of the perception of connectivity and relative depth between different structures.

Variational Time-Implicit Multiphase Level-Sets

We propose a new principle, the variational region competi- tion, to simultaneously propagate multiple disjoint level-sets in a fully time-implicit manner, minimizing the total cost w.r.t. region changes. We demonstrate, that the problem of multiphase level-set evolution can be reformulated in terms of a Potts problem, for which fast optimization algorithms are available using recent developments in convex relaxation. Further, we use an efficient recently proposed duality-based continuous max-flow method (1) implemented using massively parallel computing on GPUs for high computational performance. In contrast to conven- tional multi-phase level-set evolution approaches, ours allows for large time steps accelerating the evolution procedure. Further, the proposed method propagates all regions simultaneously, as opposed to the one-by- one phase movement of current time-implicit implementations. Promis- ing experiment results demonstrate substantial improvements in a wide spectrum of practical applications.