Information Processing in Computer-Assisted Interventions

The treatment of patients with head and neck cancers is a demanding medical field, due to the compact anatomy and complex functionality of the affected region. The planning process comprises issues regarding risk and applicability of intervention, extent of surgical removal, and the choice of appropriate access to the pathology. Required clinical information are obtained from different examinations, ranging from external visual and palpatory inspection, over preoperative panendoscopy and biopsy histology to radiological imaging.

The surgeon needs to process all available information in mind and virtually compile a mental patient model of the target anatomy. 3D visualizations of tomographic data may improve perception of spatial relationships. However, discussions with clinical practicians reveal that parameterization of advanced visual effects tend to be cumbersome and resulting visualizations are often too complex and not dedicated to specific diagnostic or treatment planning questions. Moreover, they will add valuable alternative views, but cannot replace all the other diagnostic sources.

We describe long-term experiences on developing and refining a software for ENT surgery planning and documentation. Regarding 3D visualizations, it turns out to be superior to generate sequences of rather simple 3D views directly supporting specific treatment questions, instead of presenting many anatomic structures simultaneously. Developing software for clinical practice thereby benefits from a thorough understanding of the target scenarios and the “visual questions” they raise. The second focus is on the seamless integration of the different diagnostic modalities, findings, and therapy decisions into a common electronic document. We report on the actual clinical use of the system and discuss how it fits into the surgical planning workflow.

We have designed an image guidance system to aid complex aortic aneurysm procedures. The system is based around an intensity-based 2D-3D rigid registration algorithm which accurately aligns a vertebra in the preoperative CT to interventional fluoroscopy. A 3D surface rendering of the aorta and relevant target vessels is then overlaid onto the fluoroscopy to aid guidance during the procedure. We report results from use of the system during twenty three procedures over a ten month period. Results showed the system to have an overall failure rate of 5.8%, which was mostly caused by errors in determining a starting position (misidentifying a vertebra). In 78% of cases our method was within our target accuracy of 3mm. Non-rigid deformation caused by the interventional instruments is believed to be the main reason for larger errors. In twenty one of the twenty three cases the surgeon rated the system as aiding the procedure.

Studying the usability of a system has become increasingly popular over the last decades and is seen as a crucial factor for the success of many systems and products. In a complex domain, like the Operating Room, creating systems with high usability is even more important, as deficiencies in the design can have serious impacts. However, evaluating the usability of systems in complex domains like the Operating Room (OR) is not an easy endeavor. In order to handle the complexity of the OR domain, usability specialists require deep domain knowledge, which they usually cannot acquire.

To support the usability study of intra-operative systems, we propose an OR specific domain model and inspect how this model could be used to reduce some of the usability evaluation problems. The proposed model consists of various aspects of the Operating Room, which affect the domain complexity like workflow, human role and their interconnections. As a proof of concept, we report from a usability study of an intra-operative device, which was performed based on the proposed approach.

A novel algorithm is presented to segment and reconstruct injected bone cement from a sparse set of X-Ray images acquired at arbitrary poses. The Sparse X-Ray Multi-view Active Contour (SxMAC – pronounced “smack”) can (1) reconstruct objects for which the background partially occludes the object in X-Ray images, (2) use X-Ray images acquired on a non-circular trajectory, and (3) incorporate prior CT information. The algorithm’s inputs are pre-processed X-Ray images, their associated pose information, and prior CT, if available. The algorithm initiates automated reconstruction using visual hull computation from a sparse number of x-ray images. It then improves the accuracy of the reconstruction by optimizing a geodesic active contour. A cadaver experiment demonstrates SxMAC’s ability to reconstruct high contrast bone cement that has been injected into a femur and achieve sub-millimeter accuracy with 4 images.

Spinal needle injections for back pain management are frequently carried out in hospitals and radiological clinics. Currently, these procedures are performed under fluoroscopy or CT guidance in specialized interventional radiology facilities. As an alternative, the use of inexpensive ultrasound image guidance promises to reduce the costs and increase the availability and safety of procedure. We propose to eliminate the need for ionizing radiation by creating a statistical shape model of the lumbar vertebrae and registering it to 3D ultrasound volumes of patient using a groupwise registration algorithm. From a total of 35 patient CT volumes, statistical shape models of the L2, L3 and L4 vertebrae are created, including the mean shapes and principal modes of variation. The statistical shape models are simultaneously registered to the 3D ultrasound by interchangeably optimizing the model parameters and their relative poses. We also use a biomechanical model to constrain the relative motion of the individual vertebra models throughout the registration process. The proposed method was validated on three phantoms with realistic spinal curvatures.

A new robotic system for trans-rectal ultrasound (TRUS) imaging during robot-assisted laparoscopic radical prostatectomy is described. The system consists of three main parts: a robotic probe manipulator (robot), an ultrasound machine with a biplane TRUS probe, and control and image processing software. A review of prior use of TRUS during prostatectomy is provided in order to demonstrate the potential benefits of such intra-operative imaging. The ability of the system to capture two-dimensional and three-dimensional B-mode and elastography data is demonstrated using a prostate phantom. A registration method that can be used for instrument tracking in RALRP is described and tested. Initial patient images captured using the system are presented.

We present a novel and relatively simple method for magnifying forces perceived by an operator using a tool. A sensor measures the force between the tip of a tool and its handle held by the operator’s fingers. These measurements are used to create a proportionally greater force between the handle and a brace attached to the operator’s hand, providing an enhanced perception of forces between the tip of the tool and a target. We have designed and tested a prototype that is completely hand-held and thus can be easily manipulated to a wide variety of locations and orientations. Preliminary psychophysical evaluation demonstrates that the device improves the ability to detect and differentiate between small forces at the tip of the tool. Magnifying forces in this manner may provide an improved ability to perform delicate surgical procedures, while preserving the flexibility of a hand-held instrument.

In this paper we propose a new system that allows reliable acetabular cup placement when the THA is operated in lateral approach. Conceptually it combines the accuracy of computer-generated patient-specific morphology information with an easy-to-use mechanical guide, which effectively uses natural gravity as the angular reference. The former is achieved by using a statistical shape model-based 2D-3D reconstruction technique that can generate a scaled, patient-specific 3D shape model of the pelvis from a single conventional anteroposterior (AP) pelvic X-ray radiograph. The reconstructed 3D shape model facilitates a reliable and accurate co-registration of the mechanical guide with the patient’s anatomy in the operating theater. We validated the accuracy of our system by conducting experiments on placing seven cups to four pelvises with different morphologies. Taking the measurements from an image-free navigation system as the ground truth, our system showed an average accuracy of 2.1 ±0.7 ^o for inclination and an average accuracy of 1.2 ±1.4 ^o for anteversion.

Movement disorders in patients with Parkinson’s disease may require functional surgery, when medical therapy isn’t effective. In Deep Brain Stimulation (DBS), electrodes are implanted within the brain to stimulate deep structures such as SubThalamic Nucleus (STN). This paper describes successive steps for constructing digital atlases gathering patient’s location of electrode contacts and clinical scores. Three motor and three neuro-psychological scores were integrated in the study. Correlations between active contacts localization and clinical data were carried out using an adapted Hierarchical Ascendant Classification and have enabled the extraction of clusters aiming to suggest optimum sites for therapeutic STN DBS. The postero-superior region has been found to be effective for motor score improvement whereas the antero-inferior region revealed noticeable neuro-psychological scores deterioration. Comparison with existing results has shown that such atlases are very promising for understanding phenomena better.

We propose the use of a “pick-up” ultrasound transducer for robot-assisted minimally invasive surgeries. Unlike prior approaches, the ultrasound transducer is inserted before the procedure and remains in the abdominal cavity throughout. We present a new design for such an intra-abdominal ultrasound transducer with a handle that can be grasped in a repeatable manner using a da Vinci Pro-Grasp tool. The main application is mapping the vasculature, which is segmented from Doppler and B-mode images using a Kalman-filtering approach. Our goal is employ the vasculature to register pre-operative CT to intra-operative camera images. To demonstrate the feasibility of the approach, we use an ultrasound flow phantom to register a CT surface model to extracted ultrasound vessel center points using an iterative closest point method. The transducer was tracked with electromagnetic sensors and a target registration error of 3.2 mm was calculated. The initial application will be nephrectomy where vessel localization is paramount.

Fluoroscopic overlay images rendered from pre-operative volumetric data can provide additional guidance for physicians during catheter ablation procedures for treatment of atrial fibrillation (AFib). As these overlay images are compromised by cardiac and respiratory motion, motion compensation methods have been proposed. The approaches so far either require simultaneous biplane imaging for 3-D motion compensation or, in case of mono-plane X-ray imaging, provide only a limited 2-D functionality. To overcome the downsides of the previously suggested methods, we propose a new approach that facilitates full 3-D motion compensation even if only mono-plane X-ray views are available. To this end, we use constrained model-based 2-D/3-D registration to track a circumferential mapping catheter which is commonly used during AFib catheter ablation procedures. Our approach yields an average 2-D tracking error of 0.6 mm and an average 3-D tracking error of 2.1 mm.

X-ray fluoroscopically guided cardiac ablation procedures are commonly carried out for the treatment of cardiac arrhythmias, such as atrial fibrillation (AF). X-ray images have poor soft tissue contrast and, for this reason, overlay of a 3D roadmap derived from pre-procedural volumetric image data can be used to add anatomical information. It is a requirement to determine and record the 3D positions of the ablation catheter tip in the 3D road map during AF ablation. This feature can be used as a guidance and post-procedure analysis tool. The 3D positions of the catheter tip can be calculated from biplane X-ray images and mapped to the 3D roadmap. However, the registration between the 3D roadmap and the 2D X-ray data can be compromised by patient respiratory and cardiac motions. As the coronary sinus (CS) catheter is not routinely altered during the procedure, tracking the CS catheter in real-time can be used as means of motion correction to improve the accuracy of registration between live X-ray images and a 3D roadmap. To achieve a fast and automatic ablation point tagging system from biplane images, we developed a novel tracking method for real-time simultaneous detection of the ablation catheter and the CS catheter from fluoroscopic X-ray images. We tested our tracking method on 1083 fluoroscopy frames from 16 patients and achieved a success rate of 97.5% and an average 2D tracking error of 0.5 mm ± 0.3 mm. In order to achieve tagging using a mono-plane X-ray image system, we proposed a novel motion gating method to select a pair of images from two short image sequences acquired from two different views. Both respiratory and cardiac motion phases are matched by selecting the pair of images with the minimum reconstruction error of the CS catheter electrodes. Finally, the 3D position of the ablation catheter tip was calculated using the epipolar constraint from the multiview images. We validated our automatic ablation point tagging strategy by computing the reconstruction error of the ablation catheter tip and achieved an error of 1.1 mm ± 0.5 mm.

Organ motion and deformation are major obstacles hindering the introduction of image guidance into soft tissue surgery. Due to challenges with update rate, invasiveness and intra-operative complexity, there is currently no clinically established solution for deformation measurements on soft tissues. We present a soft tissue tracking approach based on single passive markers as part of a navigation system for open liver surgery. Such markers are minimally-invasive and allow for real-time organ motion measurements using available tracking systems. The absence of correspondence between position measurements over time and the sensitivity to other reflectors present within the workspace inhibit the direct clinical implementation of such technology. Hence, we remove measurement artifacts, establish marker correspondence over time, and achieve robustness against occlusions and presence of other reflecting objects. A thorough experimental evaluation demonstrates reliable motion tracking and motivates its use for deformation detection and respiratory gating in open liver surgery.

Integration of information from complementary imaging mo-dalities in medical image registration schemes potentially improves the registration accuracy. MRI is now being used for guidance of various neurosurgical procedures like anterior temporal lobe resection in patients with refractory temporal lobe epilepsy. Accurate localisation of critical white matter tracts, such as the optic radiation, during neurosurgery is critical in ensuring good patient outcome. Current commercial interventional MR scanners are capable of performing anatomical and diffusion weighted imaging. We propose a near real-time multivariate registration scheme that uses both anatomical and diffusion images from the pre and intraoperative imaging sessions. The registration framework is optimized for use on graphical processing units and we perform a full multivariate non-rigid registration in under three minutes making the proposed framework suitable for use under the stringent time constraints of neurosurgical procedures. We assess the accuracy of our algorithm using a numerical phantom and demonstrate accurate localisation of the optic radiation in clinical datasets. This work could be of significant utility in image guided interventions and facilitate effective surgical treatments.

Epidural anesthesia is a common but challenging procedure in obstetrics and surgery, especially for the obese patient. An ultrasound guidance system is proposed using a transducer-mounted camera to create 3D panorama images of the spine relative to markings on the skin. Guidance will include identification of individual vertebrae, and selection of a suitable puncture site, trajectory and depth of needle insertion. This study describes the panorama creation and preliminary testing. The camera tracks the transducer movement using a specialized strip of markers attached to the skin surface, which enables absolute position estimation of the transducer with respect to the patient over the full range of the spine. The 3D panorama image can then be resliced in various parasagittal planes to show either the target epidural spaces or the laminae. The geometric accuracy of the panoramas are validated against an optical tracking system and independent measurements by a sonographer.