Antoine Abche

Image registration based on neural network and Fourier transform.

An image registration technique based on feed forward neural network and Fourier Transform is developed and presented. In the proposed scheme, the spectrums of the acquired images are computed, the Fourier coefficients within a selected central window of each spectrum are extracted and fed as inputs to the neural network. The feed forward neural network is implemented to estimate the transformation, defined in terms of the translation, rotation and magnification parameters, to align the corresponding images. This approach does not estimate the various registration parameters separately. They are estimated simultaneously leading to a better-optimized set of registration parameters. The approach is successful and yields better results than another Fourier based registration technique. The approach is validated on 2D images. However, it can be easily extended to 3-D application

Convex Hull in Medical Simulations: A New Hybrid Approach

Nowadays, virtual reality techniques have become widely used in different fields such as medical and architecture. Since a real object does not have a deterministic shape, it is impossible to define a geometric equation to model it. Thus, alternative approaches are the convex hull algorithms to form the convex envelopes of any object and to mimic realistic environment with exact collision detection between objects in the virtual world. In this paper, a hybrid approach to generate the convex hull is developed and presented. The new algorithm is validated by performing a comparison with three conventional algorithms namely the brute force, the gift wrapping and the QuickHull algorithms. The evaluation is achieved by generating the convex envelope of 3D wrist and knee bones using the four different approaches. The results show the improvement associated with the proposed approach

Computer-Based Training System for Simulating Wrist Arthroscopy

The minimally invasive approach of arthroscopy means less pain and faster recovery time for patients compared to open surgery. However, it implies a high difficulty of performance. In this paper, a functional prototype of a computer-based training simulator for wrist arthroscopy is introduced. A 3-D virtual representation of the bones constituting the wrist of a patient is shown. Objects are modeled using the convex hull approaches and an algorithm to simulate real time collision detection during the training on the operation is presented. In addition, a force feedback device is used as a haptic interface with the computer simulation system. This leads in the development of a low cost system that is used by trainees with the same benefits as professional devices. In this regard, the wrist arthroscopy can be simulated and medical students can learn the basic skills required with safety, flexibility and less cost.

Collision Detection in Computer Simulations for Wrist Arthroscopic Surgery Training

Computer-Based surgical simulators are one of the most recent topics in virtual reality development. They have become the training method and the tool to acquire valuable information for many medical students and medical practitioners. The real-time interactive collision detection is an important problem that must be addressed to make such simulators more realistic. In this regard, this paper addresses the issue of precise collision detection between virtual objects and proposes a new technique. First, the convex hull of each object is constructed. Then, the problem is formulated and a linear programming solution is obtained to determine whether a collision exists or not. The algorithm is tested on a medical application. The proposed collision detection approach is compared with two conventional algorithms namely the IVRI-CD and SWIFT techniques and validated using a 3D wrist model. A Haptic feedback system is implemented for evaluation purposes. The results show that the proposed approach is efficient, accurate, fast and robust in detecting collision between virtual objects during training and experimenting surgery.

An FPGA Implementation of a High Resolution Phase Shift Beamformer

In this work, an implementation of a high resolution phase shift beamformer on FPGA is presented. This implementation is based on the angle recording (AR) CORDIC algorithm and uses a fine precision 2-path floating point adder for an accurate computation of the phase shifts. A pipelined architecture is used to decrease the chip area requirements and to increase the throughput of the beamformer. The proposed approach is quantitatively evaluated by performing a comparison with am approach based on the Xilinx core generator modules. The results show that the proposed implementation outperforms the latter technique in terms of speed and area. The implemented beamformer could be used for acquiring 2D and 3D ultrasound medical images in real time.

MRI Reconstruction Using SVD in the Least Square Sense

Magnetic resonance imaging is a very powerful tool for imaging structures or organs of the human body in a non-invasive manner. Having collected the raw data and discretized in the k-space, Fourier methods are the natural choice to reconstruct the MRI images. In this paper, a technique to reconstruct MRI images is presented. It does not perform the inverse Fourier transform directly on the k-space images. It is based on the SVD decomposition in the least square sense. This approach is compared with other existing approaches using different criteria such as the root mean square and the performance test. The results show that the proposed approach is much better.

A Pseudo-Output Error Algorithm To Improve Global Convergence

In this work, a new off-line optimization approach is proposed to improve the global convergence. This algorithm, called pseudo-output error algorithm, is based on the introduction of a stationary filter in the sensitivity functions of the Newton algorithm. The convergence of the proposed algorithm is studied and analyzed. A comparison with Newton algorithm is performed to evaluate its performance. The results of the simulations show that the pseudo-output error algorithm converges to true parameters describing the system to be identified. Whereas, the Newton algorithm might converge to a secondary minimum for the same input-output data and initialized parameters

Robust informed split gradient NMF using αβ-divergence for source apportionment

Source apportionment is a very challenging topic for which non-negative source separation is well-suited. Recently, we proposed several informed Non-negative Matrix Factorization (NMF) for which some expert knowledge was introduced. These methods were all dealing with some set values of one factor together with the row sum-to-one property by either processing each constraint alternatingly or using a new parameterization which involves all of them. However, this last method was sensitive to the presence of outliers. In this paper, we thus propose a new robust informed Split Gradient NMF method which is based on a weighted αβ-divergence cost function. Experiments conducted for several input SNR with and without outliers on simulated mixtures of particulate matter sources show the relevance of the new approach.

Initialization of output-error identification methods – Comparison between ARX and RPM models

Abstract The main disadvantage of the Output-Error (OE) identification methods is that they may converge to a secondary optimum. A good initialization converges to the global optimum. The ARX model is often selected as initialization step to OE algorithms. However, the ARX model may be too biased and may not lead to a good initialization. This paper presents an approach based on the Reinitialized Partial Moment (RPM) to obtain a good initialization for OE methods. The RPM model presents an implicit filter that replaces the necessary explicit filter required by the ARX model. The result are encouraging and they have shown that the RPM based approach has to lead to a better initialization than the conventional techniques.

Informed Split Gradient Non-negative Matrix factorization using Huber cost function for source apportionment

Source apportionment is usually tackled with blind Positive/Non-negative Matrix factorization (PMF/NMF) methods. However, the obtained results may be poor due to the dependence between some rows of the second factor. We recently proposed to inform the estimation of this factor using some prior knowledge provided by chemists—some entries are set to some fixed values—and the sum-to-one property of each row. These constraints were recently taken into account by using a parameterization which gathers all of them. In this paper, a novel robust NMF approach able to cope with outliers is proposed. For that purpose, we consider the Huber loss function—a ℓ2-ℓ1 cost function—which is robust to outliers, contrary to the Frobenius norm classically met in NMF. We thus propose new update rules for the informed Huber NMF in the framework of the split gradient techniques. The choice of the adaptive cutoff parameter—which links both single cost functions—is discussed along this paper. The proposed approach is shown to outperform state-of-the-art methods on several source apportionment simulations involving various input SNRs and outliers.