Interactivity aspects in medical image visualisation highlighted in multi-faceted workshop

Utrecht 18 October 2001IMIVA 2001, the Interactive Medical Image Visualization and Analysis workshop was organised as part of the Fourth International Conference on Medical Image Computing and Computer-Assisted Intervention. This MICCAI satellite event introduced no less than 16 research projects, five of which are carried out at Dutch universities, to an expert audience of some 60 attendants. IMIVA hosted six sessions with eight full and eight short lectures adressing topics such as image segmentation, visualisation and analysis, computer-aided surgery, software platforms for interactive systems, and volume rendering.


Brain and cardiac imaging were recurrent themes in the IMIVA workshop, set up by Dr. Viergever, Director of the Dutch Image Sciences Institute (ISI), and chaired by Dr. Gerritsen, Philips Medical Systems, Dr. Niessen, University Medical Center Utrecht, and Dr. Olabarriaga, Federal University of Rio Grande do Sul in Bresil. ISI-researcher Letteboer explained how in neuro-navigation, a lot of deformations can occur during tumour visualisation, due to brainshift. This can be solved with intra-operative ultrasound. A first step consists in manually segmenting the tumour in pre-operative MR images. To speed up the process, the team has now developed an interactive multi-scale watershed algorithm for semi-automatic segmentation. The advantage is that it maintains the influence of the surgeon.

At the ETRO Department of the Free University in Brussels, researcher Deklerck has designed a visualisation software to interactively delineate and label the brain sulci using flat cortical 2D brain maps supported by 3D orthogonal views and surface rendering. The method allows to view the brain surface in one time and is based on MRI patient data taken at the Erasmus Hospital. Localisation of Tuberous Sclerosis (TS) lesions in MR brain images acquired by means of Fluid Attenuated Inversion Recovery (FLAIR) scans remains difficult. Ms. Alderliesten showed how semi-automatic segmentation and quantification at ISI has improved the reproducibility of results for this neurological disorder by automating the threshold selection process. Maybe, the method will serve as a golden standard in the future but the accuracy still has to be optimised.

Epilepsy surgeons will benefit from a volume renderer which can display multi-modality images such as CT, MR, PET, SPECT, fMRI, and EEG, next to segmentation results and 3D interactive tools. ISI-researcher Noordmans described how the system is applied for the planning of neuro-surgical procedures instead of 3D graphics, in order to localise functional areas in fMRI and visualise objects in CT scans that are surrounded by other objects, such as subdurally implanted EEG electrodes in an epileptic patient. Dr. Li of the Australian CSIRO Institute presented a high-assurance system (HAS) for vessel modelling with minimum human intervention. The system is used for the planning of endoluminal procedures in abdominal aortic aneurysm (AAA) patients. It enables surgeons to progressively construct and simultaneously segment finer patient-specific vessel models from bi-level generic representations.

At the London Imperial College, a Cardiac Viewer was developed to visualise and analyse myocardial motion obtained from 4D tagged MR images. Dr. Chandrashekara explained the technique is based on a non-rigid image registration algorithm and allows the cardiologist to navigate through the patient data set using either animated bullseye plots or virtual tag grids to correlate abnormalities in the cardiac motion with the anatomy of the heart. In order to render crisp 3D images of blood vessels Jaap Smit from Twente University made an ardent plea for the iso-surface volume rendering algorithm because it combines high accuracy and speed, without the clinician having to deal with disturbing artifacts.

With a touch of irony, Terry Peters from the Montreal Neurological Institute in Canada introduced the modular ASP (Acronym Still Pending) platform for a manual segmentation of post-operative lesions in MR images. Meshes can be deformed by means of 3D thin-plate spline control points resulting in an ellipsoid phantom. Stereoscopic visualisation aids in the performance of the task and the user can interact with the data via a 2D mouse or a 3D tracking system. The French INRIA researcher Delingette presented the YAV++ platform for image processing and visualisation. The system has 2 distinct layers, namely a set of C++ libraries and a user interface for application development, and 3 main features including modularity, portability on Windows and UNIX systems and also scripting ability used for prototyping applications. It is applied for deformable model-based segmentation, multi-modal registration of deep brain structures, and the follow-up of lesions in time series of images.

Charl Botha of Delft University held an enthusiastic talk on the Delft Shoulder Computer Assisted Surgery (DSCAS1) platform, developed for the DIPEX project in which improved endoprostheses for the upper extremities are designed. The system is uniquely used for visualisation in the shoulder replacement surgery process with fast slicers, a data cube viewer, a rich set of widgets, and surface and direct volume rendering. It has a flexible and scalable architecture with pool managers and data entities, and will be integrated with predictive modelling applications. At Massachusetts Institute of Technology, a virtual endoscopy tool was designed to combine the strengths of conventional endoscopy, 2D CT and MR data, and 3D volume visualisation. Delphine Nain pictured the system as a part of the 3D Slicer programme, implemented at the Brigham and Women's Hospital to generate 3D models, apply quantitative analysis, and perform interactive virtual colonoscopy and cardiovascular endoscopy. The team created an automatic path planning algorithm to obtain centre points for a virtual fly-through trajectory.

Cesar Mendoza described the work performed at INRIA to simulate realistic and real time cutting with haptic interaction in the SHARP project to train surgery procedures using Virtual Reality. The 2D cutting algorithm separates the facets of a mesh instead of destroying or dividing them. An intermediate representation of the physicial model has to overcome the difference between the sampling rate requirements of the model and the haptic device and to maintain the stability of the latter. Currently, the 2D meshes and haptic device run on a SGI R10000 at 175 MHz. Future research will focus on 3D cutting. Eric Bittar of the University of Reims demonstrated how the evolution of objects in a series of volumes can be displayed and quantified in a blink via their surface and trajectory. The aim is to study tagged proteins in living cells. The iso-surface calculated with the Marching Cube algorithm is applied to initialise a deformable surface model based on the delta-snake model.

Proposed by Dr. Yoo, a National Library of Medicine researcher, was a surgical planning and template design workstation to create and control the trajectories of pedicle screws for spine procedures. The system includes a volume rendering programme with physical user interfaces, 3D stereo viewing, polygonal primitives and tools for constructive solid geometry (CSG) to model the templates. The aim is to remove the need for expensive and fragile navigation equipment from the operating theatre and reduce patient X-ray radiation dose during the procedure. In Bresil, the Santa Maria University Hospital runs a project to objectively base skeletal age on a digital signature obtained by analysing the longitudinal axis of the middle finger (LAMF) in hand-wrist radiographs. Dr. Olabarriaga explained the interactive LAMF segmentation method in which robustness is set against variation. To this end, a new cursor with a vertical ellipse and fixed size was designed for the user to reduce the initial variation in the results.

Dr. Bernhard Preim from the Bremer Center for Medical Diagnostic Systems and Visualization (MeVis), addressed the challenge of 3D distance measurements in medical visualisations, which are vital for an accurate diagnosis. Quantitative analysis of spatial relations between objects is used in therapy planning and in liver surgery planning in particular. The 3D measurements require an intuitive and precise 3D interaction, provide a relation between the measurement lines, and should not hide each other nor occlude the visualisation heavily. MeVis is striving to develop an automatic approach in order to support the most important measurement tasks. The IMIVA audience showed great interest in the topics that were highlighted by the different speakers and materialised this interest in a multitude of questions. The session chairs had a tough job respecting the time schedule but succeeded wonderfully well in rounding a very successful workshop.

Leslie Versweyveld

[Medical IT News][Calendar][Virtual Medical Worlds Community][News on Advanced IT]