Data-driven clustering : new methods and applications

Abstract : This work is concerned with the development and application of novel unsupervised learning methods, having in mind two target applications: the analysis of forensic case data and the classification of remote sensing images. First, a method based on a symbolic optimization of the inter-sample distance measure is proposed to improve the flexibility of spectral clustering algorithms, and applied to the problem of forensic case data. This distance is optimized using a loss function related to the preservation of neighborhood structure between the input space and the space of principal components, and solutions are found using genetic programming. Results are compared to a variety of state-of--the-art clustering algorithms. Subsequently, a new large-scale clustering method based on a joint optimization of feature extraction and classification is proposed and applied to various databases, including two hyperspectral remote sensing images. The algorithm makes uses of a functional model (e.g., a neural network) for clustering which is trained by stochastic gradient descent. Results indicate that such a technique can easily scale to huge databases, can avoid the so-called out-of-sample problem, and can compete with or even outperform existing clustering algorithms on both artificial data and real remote sensing images. This is verified on small databases as well as very large problems. Résumé : Ce travail de recherche porte sur le développement et l'application de méthodes d'apprentissage dites non supervisées. Les applications visées par ces méthodes sont l'analyse de données forensiques et la classification d'images hyperspectrales en télédétection. Dans un premier temps, une méthodologie de classification non supervisée fondée sur l'optimisation symbolique d'une mesure de distance inter-échantillons est proposée. Cette mesure est obtenue en optimisant une fonction de coût reliée à la préservation de la structure de voisinage d'un point entre l'espace des variables initiales et l'espace des composantes principales. Cette méthode est appliquée à l'analyse de données forensiques et comparée à un éventail de méthodes déjà existantes. En second lieu, une méthode fondée sur une optimisation conjointe des tâches de sélection de variables et de classification est implémentée dans un réseau de neurones et appliquée à diverses bases de données, dont deux images hyperspectrales. Le réseau de neurones est entraîné à l'aide d'un algorithme de gradient stochastique, ce qui rend cette technique applicable à des images de très haute résolution. Les résultats de l'application de cette dernière montrent que l'utilisation d'une telle technique permet de classifier de très grandes bases de données sans difficulté et donne des résultats avantageusement comparables aux méthodes existantes.

Improved myocardial tagging contrast in cine balanced SSFP images.

PURPOSE: To improve the tag persistence throughout the whole cardiac cycle by providing a constant tag-contrast throughout all the cardiac phases when using balanced steady-state free precession (bSSFP) imaging. MATERIALS AND METHODS: The flip angles of the imaging radiofrequency pulses were optimized to compensate for the tagging contrast-to-noise ratio (Tag-CNR) fading at later cardiac phases in bSSFP imaging. Complementary spatial modulation of magnetization (CSPAMM) tagging was implemented to improve the Tag-CNR. Numerical simulations were performed to examine the behavior of the Tag-CNR with the proposed method, and to compare the resulting Tag-CNR with that obtained from the more commonly used spoiled gradient echo (SPGR) imaging. A gel phantom, as well as five healthy human volunteers, were scanned on a 1.5T scanner using bSSFP imaging with and without the proposed technique. The phantom was also scanned with SPGR imaging. RESULTS: With the proposed technique, the Tag-CNR remained almost constant during the whole cardiac cycle. Using bSSFP imaging, the Tag-CNR was about double that of SPGR. CONCLUSION: The tag persistence was significantly improved when the proposed method was applied, with better Tag-CNR during the diastolic cardiac phase. The improved Tag-CNR will support automated tagging analysis and quantification methods.

Spatiotemporal analysis of multichannel EEG: CARTOOL.

This paper describes methods to analyze the brain's electric fields recorded with multichannel Electroencephalogram (EEG) and demonstrates their implementation in the software CARTOOL. It focuses on the analysis of the spatial properties of these fields and on quantitative assessment of changes of field topographies across time, experimental conditions, or populations. Topographic analyses are advantageous because they are reference independents and thus render statistically unambiguous results. Neurophysiologically, differences in topography directly indicate changes in the configuration of the active neuronal sources in the brain. We describe global measures of field strength and field similarities, temporal segmentation based on topographic variations, topographic analysis in the frequency domain, topographic statistical analysis, and source imaging based on distributed inverse solutions. All analysis methods are implemented in a freely available academic software package called CARTOOL. Besides providing these analysis tools, CARTOOL is particularly designed to visualize the data and the analysis results using 3-dimensional display routines that allow rapid manipulation and animation of 3D images. CARTOOL therefore is a helpful tool for researchers as well as for clinicians to interpret multichannel EEG and evoked potentials in a global, comprehensive, and unambiguous way.

Methods for Determining Frequency- and Region-Dependent Relationships Between Estimated LFPs and BOLD Responses in Humans

The relationship between electrophysiological and functional magnetic resonance imaging (fMRI) signals remains poorly understood. To date, studies have required invasive methods and have been limited to single functional regions and thus cannot account for possible variations across brain regions. Here we present a method that uses fMRI data and singe-trial electroencephalography (EEG) analyses to assess the spatial and spectral dependencies between the blood-oxygenation-level-dependent (BOLD) responses and the noninvasively estimated local field potentials (eLFPs) over a wide range of frequencies (0-256 Hz) throughout the entire brain volume. This method was applied in a study where human subjects completed separate fMRI and EEG sessions while performing a passive visual task. Intracranial LFPs were estimated from the scalp-recorded data using the ELECTRA source model. We compared statistical images from BOLD signals with statistical images of each frequency of the eLFPs. In agreement with previous studies in animals, we found a significant correspondence between LFP and BOLD statistical images in the gamma band (44-78 Hz) within primary visual cortices. In addition, significant correspondence was observed at low frequencies (<14 Hz) and also at very high frequencies (>100 Hz). Effects within extrastriate visual areas showed a different correspondence that not only included those frequency ranges observed in primary cortices but also additional frequencies. Results therefore suggest that the relationship between electrophysiological and hemodynamic signals thus might vary both as a function of frequency and anatomical region.

Automatic prostate segmentation in cone-beam computed tomography images using rigid registration.

We propose to evaluate automatic three-dimensional gray-value rigid registration (RR) methods for prostate localization on cone-beam computed tomography (CBCT) scans. In total, 103 CBCT scans of 9 prostate patients have been analyzed. Each one was registered to the planning CT scan using different methods: (a) global RR, (b) pelvis bone structure RR, (c) bone RR refined by local soft-tissue RR using the CT clinical target volume (CTV) expanded with a 1, 3, 5, 8, 10, 12, 15 or 20-mm margin. To evaluate results, a radiation oncologist was asked to manually delineate the CTV on the CBCT scans. The Dice coefficients between each automatic CBCT segmentation - derived from the transformation of the manual CT segmentation - and the manual CBCT segmentation were calculated. Global or bone CT/CBCT RR has been shown to yield insufficient results in average. Local RR with an 8-mm margin around the CTV after bone RR was found to be the best candidate for systematically significantly improving prostate localization.

Data-driven topo-climatic mapping with machine learning methods

Automatic environmental monitoring networks enforced by wireless communication technologies provide large and ever increasing volumes of data nowadays. The use of this information in natural hazard research is an important issue. Particularly useful for risk assessment and decision making are the spatial maps of hazard-related parameters produced from point observations and available auxiliary information. The purpose of this article is to present and explore the appropriate tools to process large amounts of available data and produce predictions at fine spatial scales. These are the algorithms of machine learning, which are aimed at non-parametric robust modelling of non-linear dependencies from empirical data. The computational efficiency of the data-driven methods allows producing the prediction maps in real time which makes them superior to physical models for the operational use in risk assessment and mitigation. Particularly, this situation encounters in spatial prediction of climatic variables (topo-climatic mapping). In complex topographies of the mountainous regions, the meteorological processes are highly influenced by the relief. The article shows how these relations, possibly regionalized and non-linear, can be modelled from data using the information from digital elevation models. The particular illustration of the developed methodology concerns the mapping of temperatures (including the situations of Föhn and temperature inversion) given the measurements taken from the Swiss meteorological monitoring network. The range of the methods used in the study includes data-driven feature selection, support vector algorithms and artificial neural networks.

A review of atlas-based segmentation for magnetic resonance brain images.

Normal and abnormal brains can be segmented by registering the target image with an atlas. Here, an atlas is defined as the combination of an intensity image (template) and its segmented image (the atlas labels). After registering the atlas template and the target image, the atlas labels are propagated to the target image. We define this process as atlas-based segmentation. In recent years, researchers have investigated registration algorithms to match atlases to query subjects and also strategies for atlas construction. In this paper we present a review of the automated approaches for atlas-based segmentation of magnetic resonance brain images. We aim to point out the strengths and weaknesses of atlas-based methods and suggest new research directions. We use two different criteria to present the methods. First, we refer to the algorithms according to their atlas-based strategy: label propagation, multi-atlas methods, and probabilistic techniques. Subsequently, we classify the methods according to their medical target: the brain and its internal structures, tissue segmentation in healthy subjects, tissue segmentation in fetus, neonates and elderly subjects, and segmentation of damaged brains. A quantitative comparison of the results reported in the literature is also presented.

Validation of tissue modelization and classification techniques in T1-weighted MR brain images

We propose a deep study on tissue modelization andclassification Techniques on T1-weighted MR images. Threeapproaches have been taken into account to perform thisvalidation study. Two of them are based on FiniteGaussian Mixture (FGM) model. The first one consists onlyin pure gaussian distributions (FGM-EM). The second oneuses a different model for partial volume (PV) (FGM-GA).The third one is based on a Hidden Markov Random Field(HMRF) model. All methods have been tested on a DigitalBrain Phantom image considered as the ground truth. Noiseand intensity non-uniformities have been added tosimulate real image conditions. Also the effect of ananisotropic filter is considered. Results demonstratethat methods relying in both intensity and spatialinformation are in general more robust to noise andinhomogeneities. However, in some cases there is nosignificant differences between all presented methods.

The M-Coffee web server: a meta-method for computing multiple sequence alignments by combining alternative alignment methods.

The M-Coffee server is a web server that makes it possible to compute multiple sequence alignments (MSAs) by running several MSA methods and combining their output into one single model. This allows the user to simultaneously run all his methods of choice without having to arbitrarily choose one of them. The MSA is delivered along with a local estimation of its consistency with the individual MSAs it was derived from. The computation of the consensus multiple alignment is carried out using a special mode of the T-Coffee package [Notredame, Higgins and Heringa (T-Coffee: a novel method for fast and accurate multiple sequence alignment. J. Mol. Biol. 2000; 302: 205-217); Wallace, O'Sullivan, Higgins and Notredame (M-Coffee: combining multiple sequence alignment methods with T-Coffee. Nucleic Acids Res. 2006; 34: 1692-1699)] Given a set of sequences (DNA or proteins) in FASTA format, M-Coffee delivers a multiple alignment in the most common formats. M-Coffee is a freeware open source package distributed under a GPL license and it is available either as a standalone package or as a web service from www.tcoffee.org.

Impact of the Advisa MRI pacing system on the diagnostic quality of cardiac MR images and contraction patterns of cardiac muscle during scans: Advisa MRI randomized clinical multicenter study results.

BACKGROUND: The Advisa MRI system is designed to safely undergo magnetic resonance imaging (MRI). Its influence on image quality is not well known. OBJECTIVE: To evaluate cardiac magnetic resonance (CMR) image quality and to characterize myocardial contraction patterns by using the Advisa MRI system. METHODS: In this international trial with 35 participating centers, an Advisa MRI system was implanted in 263 patients. Of those, 177 were randomized to the MRI group and 150 underwent MRI scans at the 9-12-week visit. Left ventricular (LV) and right ventricular (RV) cine long-axis steady-state free precession MR images were graded for quality. Signal loss along the implantable pulse generator and leads was measured. The tagging CMR data quality was assessed as the percentage of trackable tagging points on complementary spatial modulation of magnetization acquisitions (n=16) and segmental circumferential fiber shortening was quantified. RESULTS: Of all cine long-axis steady-state free precession acquisitions, 95% of LV and 98% of RV acquisitions were of diagnostic quality, with 84% and 93%, respectively, being of good or excellent quality. Tagging points were trackable from systole into early diastole (360-648 ms after the R-wave) in all segments. During RV pacing, tagging demonstrated a dyssynchronous contraction pattern, which was not observed in nonpaced (n = 4) and right atrial-paced (n = 8) patients. CONCLUSIONS: In the Advisa MRI study, high-quality CMR images for the assessment of cardiac anatomy and function were obtained in most patients with an implantable pacing system. In addition, this study demonstrated the feasibility of acquiring tagging data to study the LV function during pacing.

Cross validation of experts versus registration methods for target localization in deep brain stimulation.

In the last five years, Deep Brain Stimulation (DBS) has become the most popular and effective surgical technique for the treatent of Parkinson's disease (PD). The Subthalamic Nucleus (STN) is the usual target involved when applying DBS. Unfortunately, the STN is in general not visible in common medical imaging modalities. Therefore, atlas-based segmentation is commonly considered to locate it in the images. In this paper, we propose a scheme that allows both, to perform a comparison between different registration algorithms and to evaluate their ability to locate the STN automatically. Using this scheme we can evaluate the expert variability against the error of the algorithms and we demonstrate that automatic STN location is possible and as accurate as the methods currently used.

Effects of computing parameters and measurement locations on the estimation of 3D NPS in non-stationary MDCT images.

The goal of this study was to investigate the impact of computing parameters and the location of volumes of interest (VOI) on the calculation of 3D noise power spectrum (NPS) in order to determine an optimal set of computing parameters and propose a robust method for evaluating the noise properties of imaging systems. Noise stationarity in noise volumes acquired with a water phantom on a 128-MDCT and a 320-MDCT scanner were analyzed in the spatial domain in order to define locally stationary VOIs. The influence of the computing parameters in the 3D NPS measurement: the sampling distances bx,y,z and the VOI lengths Lx,y,z, the number of VOIs NVOI and the structured noise were investigated to minimize measurement errors. The effect of the VOI locations on the NPS was also investigated. Results showed that the noise (standard deviation) varies more in the r-direction (phantom radius) than z-direction plane. A 25 × 25 × 40 mm(3) VOI associated with DFOV = 200 mm (Lx,y,z = 64, bx,y = 0.391 mm with 512 × 512 matrix) and a first-order detrending method to reduce structured noise led to an accurate NPS estimation. NPS estimated from off centered small VOIs had a directional dependency contrary to NPS obtained from large VOIs located in the center of the volume or from small VOIs located on a concentric circle. This showed that the VOI size and location play a major role in the determination of NPS when images are not stationary. This study emphasizes the need for consistent measurement methods to assess and compare image quality in CT.

Implicit methods for qualitative modeling of gene regulatory networks.

Advancements in high-throughput technologies to measure increasingly complex biological phenomena at the genomic level are rapidly changing the face of biological research from the single-gene single-protein experimental approach to studying the behavior of a gene in the context of the entire genome (and proteome). This shift in research methodologies has resulted in a new field of network biology that deals with modeling cellular behavior in terms of network structures such as signaling pathways and gene regulatory networks. In these networks, different biological entities such as genes, proteins, and metabolites interact with each other, giving rise to a dynamical system. Even though there exists a mature field of dynamical systems theory to model such network structures, some technical challenges are unique to biology such as the inability to measure precise kinetic information on gene-gene or gene-protein interactions and the need to model increasingly large networks comprising thousands of nodes. These challenges have renewed interest in developing new computational techniques for modeling complex biological systems. This chapter presents a modeling framework based on Boolean algebra and finite-state machines that are reminiscent of the approach used for digital circuit synthesis and simulation in the field of very-large-scale integration (VLSI). The proposed formalism enables a common mathematical framework to develop computational techniques for modeling different aspects of the regulatory networks such as steady-state behavior, stochasticity, and gene perturbation experiments.

Qualitative grading of severity of lumbar spinal stenosis based on the morphology of the dural sac on magnetic resonance images.

STUDY DESIGN.: Retrospective radiologic study on a prospective patient cohort. OBJECTIVE.: To devise a qualitative grading of lumbar spinal stenosis (LSS), study its reliability and clinical relevance. SUMMARY OF BACKGROUND DATA.: Radiologic stenosis is assessed commonly by measuring dural sac cross-sectional area (DSCA). Great variation is observed though in surfaces recorded between symptomatic and asymptomatic individuals. METHODS.: We describe a 7-grade classification based on the morphology of the dural sac as observed on T2 axial magnetic resonance images based on the rootlet/cerebrospinal fluid ratio. Grades A and B show cerebrospinal fluid presence while grades C and D show none at all. The grading was applied to magnetic resonance images of 95 subjects divided in 3 groups as follows: 37 symptomatic LSS surgically treated patients; 31 symptomatic LSS conservatively treated patients (average follow-up, 2.5 and 3.1 years); and 27 low back pain (LBP) sufferers. DSCA was also digitally measured. We studied intra- and interobserver reliability, distribution of grades, relation between morphologic grading and DSCA, as well relation between grades, DSCA, and Oswestry Disability Index. RESULTS.: Average intra- and interobserver agreement was substantial and moderate, respectively (k = 0.65 and 0.44), whereas they were substantial for physicians working in the study originating unit. Surgical patients had the smallest DSCA. A larger proportion of C and D grades was observed in the surgical group. Surface measurementsresulted in overdiagnosis of stenosis in 35 patients and under diagnosis in 12. No relation could be found between stenosis grade or DSCA and baseline Oswestry Disability Index or surgical result. C and D grade patients were more likely to fail conservative treatment, whereas grades A and B were less likely to warrant surgery. CONCLUSION.: The grading defines stenosis in different subjects than surface measurements alone. Since it mainly considers impingement of neural tissue it might be a more appropriate clinical and research tool as well as carrying a prognostic value.