A Survey of Active Learning Algorithms for Supervised Remote Sensing Image Classification

Defining an efficient training set is one of the most delicate phases for the success of remote sensing image classification routines. The complexity of the problem, the limited temporal and financial resources, as well as the high intraclass variance can make an algorithm fail if it is trained with a suboptimal dataset. Active learning aims at building efficient training sets by iteratively improving the model performance through sampling. A user-defined heuristic ranks the unlabeled pixels according to a function of the uncertainty of their class membership and then the user is asked to provide labels for the most uncertain pixels. This paper reviews and tests the main families of active learning algorithms: committee, large margin, and posterior probability-based. For each of them, the most recent advances in the remote sensing community are discussed and some heuristics are detailed and tested. Several challenging remote sensing scenarios are considered, including very high spatial resolution and hyperspectral image classification. Finally, guidelines for choosing the good architecture are provided for new and/or unexperienced user.

A well-posedness theory in measures for some kinetic models of collective motion

We present existence, uniqueness and continuous dependence results for some kinetic equations motivated by models for the collective behavior of large groups of individuals. Models of this kind have been recently proposed to study the behavior of large groups of animals, such as flocks of birds, swarms, or schools of fish. Our aim is to give a well-posedness theory for general models which possibly include a variety of effects: an interaction through a potential, such as a short-range repulsion and long-range attraction; a velocity-averaging effect where individuals try to adapt their own velocity to that of other individuals in their surroundings; and self-propulsion effects, which take into account effects on one individual that are independent of the others. We develop our theory in a space of measures, using mass transportation distances. As consequences of our theory we show also the convergence of particle systems to their corresponding kinetic equations, and the local-in-time convergence to the hydrodynamic limit for one of the models.

Intravoxel incoherent motion perfusion imaging in acute stroke: initial clinical experience.

INTRODUCTION: Intravoxel incoherent motion (IVIM) imaging is an MRI perfusion technique that uses a diffusion-weighted sequence with multiple b values and a bi-compartmental signal model to measure the so-called pseudo-diffusion of blood caused by its passage through the microvascular network. The goal of the current study was to assess the feasibility of IVIM perfusion fraction imaging in patients with acute stroke. METHODS: Images were collected in 17 patients with acute stroke. Exclusion criteria were onset of symptoms to imaging >5 days, hemorrhagic transformation, infratentorial lesions, small lesions <0.5 cm in minimal diameter and hemodynamic instability. IVIM imaging was performed at 3 T, using a standard spin-echo Stejskal-Tanner pulsed gradients diffusion-weighted sequence, using 16 b values from 0 to 900 s/mm(2). Image quality was assessed by two radiologists, and quantitative analysis was performed in regions of interest placed in the stroke area, defined by thresholding the apparent diffusion coefficient maps, as well as in the contralateral region. RESULTS: IVIM perfusion fraction maps showed an area of decreased perfusion fraction f in the region of decreased apparent diffusion coefficient. Quantitative analysis showed a statistically significant decrease in both IVIM perfusion fraction f (0.026 ± 0.019 vs. 0.056 ± 0.025, p = 2.2 · 10(-6)) and diffusion coefficient D compared with the contralateral side (3.9 ± 0.79 · 10(-4) vs. 7.5 ± 0.86 · 10(-4) mm(2)/s, p = 1.3 · 10(-20)). CONCLUSION: IVIM perfusion fraction imaging is feasible in acute stroke. IVIM perfusion fraction is significantly reduced in the visible infarct. Further studies should evaluate the potential for IVIM to predict clinical outcome and treatment response.

Motion compensation strategies in magnetic resonance imaging.

Image quality in magnetic resonance imaging (MRI) is considerably affected by motion. Therefore, motion is one of the most common sources of artifacts in contemporary cardiovascular MRI. Such artifacts in turn may easily lead to misinterpretations in the images and a subsequent loss in diagnostic quality. Hence, there is considerable research interest in strategies that help to overcome these limitations at minimal cost in time, spatial resolution, temporal resolution, and signal-to-noise ratio. This review summarizes and discusses the three principal sources of motion: the beating heart, the breathing lungs, and bulk patient movement. This is followed by a comprehensive overview of commonly used compensation strategies for these different types of motion. Finally, a summary and an outlook are provided.

The use of classification and regression trees to predict the likelihood of seasonal influenza.

Background Individual signs and symptoms are of limited value for the diagnosis of influenza. Objective To develop a decision tree for the diagnosis of influenza based on a classification and regression tree (CART) analysis. Methods Data from two previous similar cohort studies were assembled into a single dataset. The data were randomly divided into a development set (70%) and a validation set (30%). We used CART analysis to develop three models that maximize the number of patients who do not require diagnostic testing prior to treatment decisions. The validation set was used to evaluate overfitting of the model to the training set. Results Model 1 has seven terminal nodes based on temperature, the onset of symptoms and the presence of chills, cough and myalgia. Model 2 was a simpler tree with only two splits based on temperature and the presence of chills. Model 3 was developed with temperature as a dichotomous variable (≥38°C) and had only two splits based on the presence of fever and myalgia. The area under the receiver operating characteristic curves (AUROCC) for the development and validation sets, respectively, were 0.82 and 0.80 for Model 1, 0.75 and 0.76 for Model 2 and 0.76 and 0.77 for Model 3. Model 2 classified 67% of patients in the validation group into a high- or low-risk group compared with only 38% for Model 1 and 54% for Model 3. Conclusions A simple decision tree (Model 2) classified two-thirds of patients as low or high risk and had an AUROCC of 0.76. After further validation in an independent population, this CART model could support clinical decision making regarding influenza, with low-risk patients requiring no further evaluation for influenza and high-risk patients being candidates for empiric symptomatic or drug therapy.

Head motion detection using FID navigators.

This work explores a concept for motion detection in brain MR examinations using high channel-count RF coil arrays. It applies ultrashort (<100 μsec) free induction decay signals, making use of the knowledge that motion induces variations in these signals when compared to a reference free induction decay signal. As a proof-of-concept, the method was implemented in a standard structural MRI sequence. The stability of the free induction decay-signal was verified in phantom experiments. Human experiments demonstrated that the observed variations in the navigator data provide a sensitive measure for detection of relevant and common subject motion patterns. The proposed methodology provides a means to monitor subject motion throughout a MRI scan while causing little or no impact on the sequence timing and image contrast. It could hence complement available motion detection and correction methods, thus further reducing motion sensitivity in MR applications.

Classification of immature stage habitats of Culicidae (Diptera) collected in Córdoba, Argentina

In order to classify mosquito immature stage habitats, samples were taken in 42 localities of Córdoba Province, Argentina, representing the phytogeographic regions of Chaco, Espinal and Pampa. Immature stage habitats were described and classified according to the following criteria: natural or artificial; size; location related to light and neighboring houses; vegetation; water: permanence, movement, turbidity and pH. Four groups of species were associated based on the habitat similarity by means of cluster analysis: Aedes albifasciatus, Culex saltanensis, Cx. mollis, Cx. brethesi, Psorophora ciliata, Anopheles albitarsis, and Uranotaenia lowii (Group A); Cx. acharistus, Cx. quinquefasciatus, Cx. bidens, Cx. dolosus, Cx. maxi and Cx. apicinus (Group B); Cx. coronator, Cx. chidesteri, Mansonia titillans and Ps. ferox (Group C); Ae. fluviatilis and Ae. milleri (Group D). The principal component analysis (ordination method) pointed out that the different types of habitats, their nature (natural or artificial), plant species, water movement and depth are the main characters explaining the observed variation among the mosquito species. The distribution of mosquito species by phytogeographic region did not affect the species groups, since species belonging to different groups were collected in the same region.

Bayesian classification criterion for forensic multivariate data

This study presents a classification criteria for two-class Cannabis seedlings. As the cultivation of drug type cannabis is forbidden in Switzerland, law enforcement authorities regularly ask laboratories to determine cannabis plant's chemotype from seized material in order to ascertain that the plantation is legal or not. In this study, the classification analysis is based on data obtained from the relative proportion of three major leaf compounds measured by gas-chromatography interfaced with mass spectrometry (GC-MS). The aim is to discriminate between drug type (illegal) and fiber type (legal) cannabis at an early stage of the growth. A Bayesian procedure is proposed: a Bayes factor is computed and classification is performed on the basis of the decision maker specifications (i.e. prior probability distributions on cannabis type and consequences of classification measured by losses). Classification rates are computed with two statistical models and results are compared. Sensitivity analysis is then performed to analyze the robustness of classification criteria.

Analyse systématique et permanente de la délinquance sérielle : place des statistiques criminelles ; apport des approches situationnelles pour un système de classification ; perspectives en matière de coopération

Résumé de la thèse L'évolution des systèmes policiers donne une place prépondérante à l'information et au renseignement. Cette transformation implique de développer et de maintenir un ensemble de processus permanent d'analyse de la criminalité, en particulier pour traiter des événements répétitifs ou graves. Dans une organisation aux ressources limitées, le temps consacré au recueil des données, à leur codification et intégration, diminue le temps disponible pour l'analyse et la diffusion de renseignements. Les phases de collecte et d'intégration restent néanmoins indispensables, l'analyse n'étant pas possible sur des données volumineuses n'ayant aucune structure. Jusqu'à présent, ces problématiques d'analyse ont été abordées par des approches essentiellement spécialisées (calculs de hot-sports, data mining, ...) ou dirigées par un seul axe (par exemple, les sciences comportementales). Cette recherche s'inscrit sous un angle différent, une démarche interdisciplinaire a été adoptée. L'augmentation continuelle de la quantité de données à analyser tend à diminuer la capacité d'analyse des informations à disposition. Un bon découpage (classification) des problèmes rencontrés permet de délimiter les analyses sur des données pertinentes. Ces classes sont essentielles pour structurer la mémoire du système d'analyse. Les statistiques policières de la criminalité devraient déjà avoir répondu à ces questions de découpage de la délinquance (classification juridique). Cette décomposition a été comparée aux besoins d'un système de suivi permanent dans la criminalité. La recherche confirme que nos efforts pour comprendre la nature et la répartition du crime se butent à un obstacle, à savoir que la définition juridique des formes de criminalité n'est pas adaptée à son analyse, à son étude. Depuis près de vingt ans, les corps de police de Suisse romande utilisent et développent un système de classification basé sur l'expérience policière (découpage par phénomène). Cette recherche propose d'interpréter ce système dans le cadre des approches situationnelles (approche théorique) et de le confronter aux données « statistiques » disponibles pour vérifier sa capacité à distinguer les formes de criminalité. La recherche se limite aux cambriolages d'habitations, un délit répétitif fréquent. La théorie des opportunités soutien qu'il faut réunir dans le temps et dans l'espace au minimum les trois facteurs suivants : un délinquant potentiel, une cible intéressante et l'absence de gardien capable de prévenir ou d'empêcher le passage à l'acte. Ainsi, le délit n'est possible que dans certaines circonstances, c'est-à-dire dans un contexte bien précis. Identifier ces contextes permet catégoriser la criminalité. Chaque cas est unique, mais un groupe de cas montre des similitudes. Par exemple, certaines conditions avec certains environnements attirent certains types de cambrioleurs. Deux hypothèses ont été testées. La première est que les cambriolages d'habitations ne se répartissent pas uniformément dans les classes formées par des « paramètres situationnels » ; la deuxième que des niches apparaissent en recoupant les différents paramètres et qu'elles correspondent à la classification mise en place par la coordination judiciaire vaudoise et le CICOP. La base de données vaudoise des cambriolages enregistrés entre 1997 et 2006 par la police a été utilisée (25'369 cas). Des situations spécifiques ont été mises en évidence, elles correspondent aux classes définies empiriquement. Dans une deuxième phase, le lien entre une situation spécifique et d'activité d'un auteur au sein d'une même situation a été vérifié. Les observations réalisées dans cette recherche indiquent que les auteurs de cambriolages sont actifs dans des niches. Plusieurs auteurs sériels ont commis des délits qui ne sont pas dans leur niche, mais le nombre de ces infractions est faible par rapport au nombre de cas commis dans la niche. Un système de classification qui correspond à des réalités criminelles permet de décomposer les événements et de mettre en place un système d'alerte et de suivi « intelligent ». Une nouvelle série dans un phénomène sera détectée par une augmentation du nombre de cas de ce phénomène, en particulier dans une région et à une période donnée. Cette nouvelle série, mélangée parmi l'ensemble des délits, ne serait pas forcément détectable, en particulier si elle se déplace. Finalement, la coopération entre les structures de renseignement criminel opérationnel en Suisse romande a été améliorée par le développement d'une plateforme d'information commune et le système de classification y a été entièrement intégré.

Functional mapping of the human visual cortex with intravoxel incoherent motion MRI.

Functional imaging with intravoxel incoherent motion (IVIM) magnetic resonance imaging (MRI) is demonstrated. Images were acquired at 3 Tesla using a standard Stejskal-Tanner diffusion-weighted echo-planar imaging sequence with multiple b-values. Cerebro-spinal fluid signal, which is highly incoherent, was suppressed with an inversion recovery preparation pulse. IVIM microvascular perfusion parameters were calculated according to a two-compartment (vascular and non-vascular) diffusion model. The results obtained in 8 healthy human volunteers during visual stimulation are presented. The IVIM blood flow related parameter fD* increased 170% during stimulation in the visual cortex, and 70% in the underlying white matter.

Quantitative Measurement of Brain Perfusion with Intravoxel Incoherent Motion MR Imaging.

Purpose: To evaluate the sensitivity of the perfusion parameters derived from Intravoxel Incoherent Motion (IVIM) MR imaging to hypercapnia-induced vasodilatation and hyperoxygenation-induced vasoconstriction in the human brain. Materials and Methods: This study was approved by the local ethics committee and informed consent was obtained from all participants. Images were acquired with a standard pulsed-gradient spin-echo sequence (Stejskal-Tanner) in a clinical 3-T system by using 16 b values ranging from 0 to 900 sec/mm(2). Seven healthy volunteers were examined while they inhaled four different gas mixtures known to modify brain perfusion (pure oxygen, ambient air, 5% CO(2) in ambient air, and 8% CO(2) in ambient air). Diffusion coefficient (D), pseudodiffusion coefficient (D*), perfusion fraction (f), and blood flow-related parameter (fD*) maps were calculated on the basis of the IVIM biexponential model, and the parametric maps were compared among the four different gas mixtures. Paired, one-tailed Student t tests were performed to assess for statistically significant differences. Results: Signal decay curves were biexponential in the brain parenchyma of all volunteers. When compared with inhaled ambient air, the IVIM perfusion parameters D*, f, and fD* increased as the concentration of inhaled CO(2) was increased (for the entire brain, P = .01 for f, D*, and fD* for CO(2) 5%; P = .02 for f, and P = .01 for D* and fD* for CO(2) 8%), and a trend toward a reduction was observed when participants inhaled pure oxygen (although P > .05). D remained globally stable. Conclusion: The IVIM perfusion parameters were reactive to hyperoxygenation-induced vasoconstriction and hypercapnia-induced vasodilatation. Accordingly, IVIM imaging was found to be a valid and promising method to quantify brain perfusion in humans. © RSNA, 2012.

Lipid composition of fingermark residue and donor classification using GC/MS

Lipids available in fingermark residue represent important targets for enhancement and dating techniques. While it is well known that lipid composition varies among fingermarks of the same donor (intra-variability) and between fingermarks of different donors (inter-variability), the extent of this variability remains uncharacterised. Thus, this worked aimed at studying qualitatively and quantitatively the initial lipid composition of fingermark residue of 25 different donors. Among the 104 detected lipids, 43 were reported for the first time in the literature. Furthermore, palmitic acid, squalene, cholesterol, myristyl myristate and myristyl myristoleate were quantified and their correlation within fingermark residue was highlighted. Ten compounds were then selected and further studied as potential targets for dating or enhancement techniques. It was shown that their relative standard deviation was significantly lower for the intra-variability than for the inter-variability. Moreover, the use of data pretreatments could significantly reduce this variability. Based on these observations, an objective donor classification model was proposed. Hierarchical cluster analysis was conducted on the pre-treated data and the fingermarks of the 25 donors were classified into two main groups, corresponding to "poor" and "rich" lipid donors. The robustness of this classification was tested using fingermark replicates of selected donors. 86% of these replicates were correctly classified, showing the potential of such a donor classification model for research purposes in order to select representative donors based on compounds of interest.

Central and Cortical Gray Mater Segmentation of Magnetic Resonance Images of the Fetal Brain

Motivation. The study of human brain development in itsearly stage is today possible thanks to in vivo fetalmagnetic resonance imaging (MRI) techniques. Aquantitative analysis of fetal cortical surfacerepresents a new approach which can be used as a markerof the cerebral maturation (as gyration) and also forstudying central nervous system pathologies [1]. However,this quantitative approach is a major challenge forseveral reasons. First, movement of the fetus inside theamniotic cavity requires very fast MRI sequences tominimize motion artifacts, resulting in a poor spatialresolution and/or lower SNR. Second, due to the ongoingmyelination and cortical maturation, the appearance ofthe developing brain differs very much from thehomogenous tissue types found in adults. Third, due tolow resolution, fetal MR images considerably suffer ofpartial volume (PV) effect, sometimes in large areas.Today extensive efforts are made to deal with thereconstruction of high resolution 3D fetal volumes[2,3,4] to cope with intra-volume motion and low SNR.However, few studies exist related to the automatedsegmentation of MR fetal imaging. [5] and [6] work on thesegmentation of specific areas of the fetal brain such asposterior fossa, brainstem or germinal matrix. Firstattempt for automated brain tissue segmentation has beenpresented in [7] and in our previous work [8]. Bothmethods apply the Expectation-Maximization Markov RandomField (EM-MRF) framework but contrary to [7] we do notneed from any anatomical atlas prior. Data set &Methods. Prenatal MR imaging was performed with a 1-Tsystem (GE Medical Systems, Milwaukee) using single shotfast spin echo (ssFSE) sequences (TR 7000 ms, TE 180 ms,FOV 40 x 40 cm, slice thickness 5.4mm, in plane spatialresolution 1.09mm). Each fetus has 6 axial volumes(around 15 slices per volume), each of them acquired inabout 1 min. Each volume is shifted by 1 mm with respectto the previous one. Gestational age (GA) ranges from 29to 32 weeks. Mother is under sedation. Each volume ismanually segmented to extract fetal brain fromsurrounding maternal tissues. Then, in-homogeneityintensity correction is performed using [9] and linearintensity normalization is performed to have intensityvalues that range from 0 to 255. Note that due tointra-tissue variability of developing brain someintensity variability still remains. For each fetus, ahigh spatial resolution image of isotropic voxel size of1.09 mm is created applying [2] and using B-splines forthe scattered data interpolation [10] (see Fig. 1). Then,basal ganglia (BS) segmentation is performed on thissuper reconstructed volume. Active contour framework witha Level Set (LS) implementation is used. Our LS follows aslightly different formulation from well-known Chan-Vese[11] formulation. In our case, the LS evolves forcing themean of the inside of the curve to be the mean intensityof basal ganglia. Moreover, we add local spatial priorthrough a probabilistic map created by fitting anellipsoid onto the basal ganglia region. Some userinteraction is needed to set the mean intensity of BG(green dots in Fig. 2) and the initial fitting points forthe probabilistic prior map (blue points in Fig. 2). Oncebasal ganglia are removed from the image, brain tissuesegmentation is performed as described in [8]. Results.The case study presented here has 29 weeks of GA. Thehigh resolution reconstructed volume is presented in Fig.1. The steps of BG segmentation are shown in Fig. 2.Overlap in comparison with manual segmentation isquantified by the Dice similarity index (DSI) equal to0.829 (values above 0.7 are considered a very goodagreement). Such BG segmentation has been applied on 3other subjects ranging for 29 to 32 GA and the DSI hasbeen of 0.856, 0.794 and 0.785. Our segmentation of theinner (red and blue contours) and outer cortical surface(green contour) is presented in Fig. 3. Finally, torefine the results we include our WM segmentation in theFreesurfer software [12] and some manual corrections toobtain Fig.4. Discussion. Precise cortical surfaceextraction of fetal brain is needed for quantitativestudies of early human brain development. Our workcombines the well known statistical classificationframework with the active contour segmentation forcentral gray mater extraction. A main advantage of thepresented procedure for fetal brain surface extraction isthat we do not include any spatial prior coming fromanatomical atlases. The results presented here arepreliminary but promising. Our efforts are now in testingsuch approach on a wider range of gestational ages thatwe will include in the final version of this work andstudying as well its generalization to different scannersand different type of MRI sequences. References. [1]Guibaud, Prenatal Diagnosis 29(4) (2009). [2] Rousseau,Acad. Rad. 13(9), 2006, [3] Jiang, IEEE TMI 2007. [4]Warfield IADB, MICCAI 2009. [5] Claude, IEEE Trans. Bio.Eng. 51(4) (2004). [6] Habas, MICCAI (Pt. 1) 2008. [7]Bertelsen, ISMRM 2009 [8] Bach Cuadra, IADB, MICCAI 2009.[9] Styner, IEEE TMI 19(39 (2000). [10] Lee, IEEE Trans.Visual. And Comp. Graph. 3(3), 1997, [11] Chan, IEEETrans. Img. Proc, 10(2), 2001 [12] Freesurfer,http://surfer.nmr.mgh.harvard.edu.

Structured output SVM for remote sensing image classification

In the recent years, kernel methods have revealed very powerful tools in many application domains in general and in remote sensing image classification in particular. The special characteristics of remote sensing images (high dimension, few labeled samples and different noise sources) are efficiently dealt with kernel machines. In this paper, we propose the use of structured output learning to improve remote sensing image classification based on kernels. Structured output learning is concerned with the design of machine learning algorithms that not only implement input-output mapping, but also take into account the relations between output labels, thus generalizing unstructured kernel methods. We analyze the framework and introduce it to the remote sensing community. Output similarity is here encoded into SVM classifiers by modifying the model loss function and the kernel function either independently or jointly. Experiments on a very high resolution (VHR) image classification problem shows promising results and opens a wide field of research with structured output kernel methods.