Vincent Bismuth

A comparison of line enhancement techniques: applications to guide-wire detection and respiratory motion tracking

During interventional radiology procedures, guide-wires are usually inserted into the patients vascular tree for diagnosis or healing purpose. These procedures are monitored with an Xray interventional system providing images of the interventional devices navigating through the patients body. The automatic detection of such tools by image processing means has gained maturity over the past years and enables applications ranging from image enhancement to multimodal image fusion. Sophisticated detection methods are emerging, which rely on a variety of device enhancement techniques. In this article we reviewed and classified these techniques into three families. We chose a state of the art approach in each of them and built a rigorous framework to compare their detection capability and their computational complexity. Through simulations and the intensive use of ROC curves we demonstrated that the Hessian based methods are the most robust to strong curvature of the devices and that the family of rotated filters technique is the most suited for detecting low CNR and low curvature devices. The steerable filter approach demonstrated less interesting detection capabilities and appears to be the most expensive one to compute. Finally we demonstrated the interest of automatic guide-wire detection on a clinical topic: the compensation of respiratory motion in multimodal image fusion.

Curvilinear structure enhancement with the polygonal path image - application to guide-wire segmentation in x-ray fluoroscopy

Curvilinear structures are common in medical imaging, which typically require dedicated processing techniques. We present a new structure to process these, that we call the polygonal path image, denoted (see text for symbol). We derive from (see text for symbol) some curvilinear structure enhancement and analysis algorithms. We show that (see text for symbol) has some interesting properties: it generalizes several concepts found in other methods; it makes it possible to control the smoothness and length of the structures under study; and it can be computed efficiently. We estimate quantitatively its performance in the context of interventional cardiology for the detection of guide-wires in Xray images. We show that (see text for symbol) is particularly well suited for this task where it appears to outperform previous state of the art techniques.

Elastic registration for stent enhancement in X-ray image sequences

Curing coronary artery disease typically involves deploying a stent (a fine mesh of wire) supported by a balloon inside the patient artery. This procedure is monitored by X-ray fluoroscopy where stent visibility is challenging. Since 2000, stent enhancement techniques for X-ray fluoroscopy image sequences have become popular. They are based on motion compensated temporal integration of the stent images. The motion of the stent is classically inferred from the motion of the highly contrasted balloon marker balls. We propose a technique that compensates the motion of the whole guide-wire in the neighborhood of the marker balls. We demonstrate that the guide wire-based registration enables better enhanced images in 28% of the cases whereas marker ball based registration is better only in 4% of the cases. We provide examples, qualitative and quantitative analysis on a database of 144 clinical cases.

A comprehensive study of stent visualization enhancement in X-ray images by image processing means

In this work we propose a comprehensive study of Digital Stent Enhancement (DSE), from the analysis of the requirements to the validation of the proposed solution. First, we derive the stent visualization requirements in the context of the clinical application and workflow. Then, we propose a DSE algorithm combining automatic detection, tracking, registration and contrast enhancement. The most original parts of our solution: landmark segmentation and non-linear image registration are detailed. Finally, we validate the algorithm on a large number of synthetic and clinical cases. Performance is characterized in terms of automation, image quality and execution time. This work is, to the best of our knowledge, the first comprehensive article on DSE, covering problem statement, proposed solution, and validation strategies.

A device enhancing and denoising algorithm for X-ray cardiac fluoroscopy

In X-ray interventional imaging, improved denoising enables lower X-ray dose and better visualization, resulting in increased confidence in the therapeutic act. This article focuses on the denoising of fluoroscopic image sequences in interventional cardiology. The challenge with these images is the need to preserve fine details in terms of size and contrast, representing the tools manipulated by the operator. These tools superimpose over an anatomical background. This context drives us to propose an algorithm based on spatial-temporal filtering conditioned by a feature of interest map. Based on the confidence in the feature detection our algorithm will either filter, preserve or enhance the image content. Our method is fast and compares very favorably with state of the art methods both quantitatively on synthetic data and perceptually on clinical data.

An image-based catheter segmentation algorithm for optimized electrophysiology procedure workflow

Electrophysiology ablation procedures are performed in an interventional lab. The therapy is delivered through several catheters introduced in cardiac chambers under x-ray guidance. They are also be used to measure some local electrical properties which can be color-coded. A kind of color-map is then established and it can be overlaid to images of the anatomy obtained with fluoroscopy. A potential improvement in the workflow of the procedure may be reached by tracking the location of the tip of the catheter performing the measurement. We propose here an image-based strategy to detect it and we report the results obtained on a large clinical database. We segment the object of interest by selecting contrasted objects and we characterize them by taking into account all possible co founding factors. A selection strategy has been defined from the distribution of the found values for the true positive and false positive elements in a first clinical database (3000 images from a single site). We got a success rate for the detection of the target object of 86% on a larger database formed of about 4500 images coming from 7 different sites. We also developed an active learning strategy for improving the performance of the algorithm and its stability in the field. The principle is to take into account the users manual correction made on a given frame when processing the following ones, which is adapted to the clinical workflow: the segmentation result is assessed and corrected by an operator for each frame. We then gained additional 6% up to 91% on the success rate: the number of algorithm mistakes to be corrected by the operator is reduced to an acceptable level.