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.

Extent of hypoattenuation on CT angiography source images in basilar artery occlusion: prognostic value in the Basilar Artery International Cooperation Study.

BACKGROUND AND PURPOSE: The posterior circulation Acute Stroke Prognosis Early CT Score (pc-ASPECTS) quantifies the extent of early ischemic changes in the posterior circulation with a 10-point grading system. We hypothesized that pc-ASPECTS applied to CT angiography source images predicts functional outcome of patients in the Basilar Artery International Cooperation Study (BASICS). METHODS: BASICS was a prospective, observational registry of consecutive patients with acute symptomatic basilar artery occlusion. Functional outcome was assessed at 1 month. We applied pc-ASPECTS to CT angiography source images of patients with CT angiography for confirmation of basilar artery occlusion. We calculated unadjusted and adjusted risk ratios (RRs) of pc-ASPECTS dichotomized at ≥8 versus <8. Primary outcome measure was favorable outcome (modified Rankin Scale scores 0-3). Secondary outcome measures were mortality and functional independence (modified Rankin Scale scores 0-2). RESULTS: Of 158 patients included, 78 patients had a CT angiography source images pc-ASPECTS≥8. Patients with a pc-ASPECTS≥8 more often had a favorable outcome than patients with a pc-ASPECTS<8 (crude RR, 1.7; 95% CI, 0.98-3.0). After adjustment for age, baseline National Institutes of Health Stroke Scale score, and thrombolysis, pc-ASPECTS≥8 was not related to favorable outcome (RR, 1.3; 95% CI, 0.8-2.2), but it was related to reduced mortality (RR, 0.7; 95% CI, 0.5-0.98) and functional independence (RR, 2.0; 95% CI, 1.1-3.8). In post hoc analysis, pc-ASPECTS dichotomized at ≥6 versus <6 predicted a favorable outcome (adjusted RR, 3.1; 95% CI, 1.2-7.5). CONCLUSIONS: pc-ASPECTS on CT angiography source images independently predicted death and functional independence at 1 month in the CT angiography subgroup of patients in the BASICS registry.

Multimodal evaluation for medical image segmentation

This paper is a joint effort between five institutionsthat introduces several novel similarity measures andcombines them to carry out a multimodal segmentationevaluation. The new similarity measures proposed arebased on the location and the intensity values of themisclassified voxels as well as on the connectivity andthe boundaries of the segmented data. We showexperimentally that the combination of these measuresimprove the quality of the evaluation. The study that weshow here has been carried out using four differentsegmentation methods from four different labs applied toa MRI simulated dataset of the brain. We claim that ournew measures improve the robustness of the evaluation andprovides better understanding about the differencebetween segmentation methods.

Mirror image atrial dilatation in adult patients with atrial fibrillation and congenital heart disease.

BACKGROUND: Atrial fibrillation (AF) is largely regarded to be initiated from left atrial (LA) dilatation, with subsequent dilatation of the right atrium (RA) in those who progress to chronic AF. We hypothesized that in adult patients with right-sided congenital heart disease (CHD) and AF, RA dilatation will predominate with subsequent dilatation of the left atrium, as a mirror image. METHODS: Adult patients with diagnosis of right-sided, ASD or left-sided CHD who had undergone an echocardiographic study and electrocardiographic recording in 2007 were included. RA and LA area were measured from the apical view. AF was diagnosed from a 12-lead electrocardiogram or Holter recording. A multivariate logistic regression model was used to identify predictors of AF and linear regression models were performed to measure relationship between RA and LA area and AF. RESULTS: A total of 291 patients were included in the study. Multivariate analysis showed that age (p=0.0001), RA (p=0.025) and LA area (p=0.0016) were significantly related to AF. In patients with pure left-sided pathologies, there was progressive and predominant LA dilatation that paralleled the development of AF from none to paroxysmal to chronic AF. In patients with pure right-sided pathologies, there was a mirror image of progressive and predominant RA dilatation with the development of AF. CONCLUSION: We observed a mirror image atrial dilatation in patients with right sided disease and AF. This may provide novel mechanistic insight as to the origin of AF in these patients and deserves further studying in the form of targeted electrophysiological studies.

Learning spatial filters for multispectral image segmentation

We present a novel filtering method for multispectral satellite image classification. The proposed method learns a set of spatial filters that maximize class separability of binary support vector machine (SVM) through a gradient descent approach. Regularization issues are discussed in detail and a Frobenius-norm regularization is proposed to efficiently exclude uninformative filters coefficients. Experiments carried out on multiclass one-against-all classification and target detection show the capabilities of the learned spatial filters.

Monitoring of illicit pill distribution networks using an image collection exploration framework

This paper proposes a novel approach for the analysis of illicit tablets based on their visual characteristics. In particular, the paper concentrates on the problem of ecstasy pill seizure profiling and monitoring. The presented method extracts the visual information from pill images and builds a representation of it, i.e. it builds a pill profile based on the pill visual appearance. Different visual features are used to build different image similarity measures, which are the basis for a pill monitoring strategy based on both discriminative and clustering models. The discriminative model permits to infer whether two pills come from the same seizure, while the clustering models groups of pills that share similar visual characteristics. The resulting clustering structure allows to perform a visual identification of the relationships between different seizures. The proposed approach was evaluated using a data set of 621 Ecstasy pill pictures. The results demonstrate that this is a feasible and cost effective method for performing pill profiling and monitoring.

Weight status, body image and bullying among adolescents in the Seychelles.

We investigated the relationship between being bullied and measured body weight and perceived body weight among adolescents of a middle-income sub Saharan African country. Our data originated from the Global School-based Health Survey, which targets adolescents aged 13-15 years. Student weights and heights were measured before administrating the questionnaire which included questions about personal data, health behaviors and being bullied. Standard criteria were used to assess thinness, overweight and obesity. Among 1,006 participants who had complete data, 16.5% (95%CI 13.3-20.2) reported being bullied ≥ 3 days during the past 30 days; 13.4% were thin, 16.8% were overweight and 7.6% were obese. Categories of actual weight and of perceived weight correlated only moderately (Spearman correlation coefficient 0.37 for boys and 0.57 for girls; p < 0.001). In univariate analysis, both actual obesity (OR 1.76; p = 0.051) and perception of high weight (OR 1.63 for "slightly overweight"; OR 2.74 for "very overweight", both p < 0.05) were associated with being bullied. In multivariate analysis, ORs for categories of perceived overweight were virtually unchanged while ORs for actual overweight and obesity were substantially attenuated, suggesting a substantial role of perceived weight in the association with being bullied. Actual underweight and perceived thinness also tended to be associated with being bullied, although not significantly. Our findings suggest that more research attention be given to disentangling the significant association between body image, overweight and bullying among adolescents. Further studies in diverse populations are warranted.

Single-trial topographic analysis of human EEG: A new image of event-related potentials

We present a novel approach for analyzing single-trial electroencephalography (EEG) data, using topographic information. The method allows for visualizing event-related potentials using all the electrodes of recordings overcoming the problem of previous approaches that required electrode selection and waveforms filtering. We apply this method to EEG data from an auditory object recognition experiment that we have previously analyzed at an ERP level. Temporally structured periods were statistically identified wherein a given topography predominated without any prior information about the temporal behavior. In addition to providing novel methods for EEG analysis, the data indicate that ERPs are reliably observable at a single-trial level when examined topographically.

Online teaching of inflammatory skin pathology by a French-speaking international university network

Introduction: Developments in technology, webbased teaching and whole slide imaging have broadened the teaching horizon in anatomic pathology. Creating online learning material including many types of media like radiologic images, videos, clinical and macroscopic photographs and whole slides imaging is now accessible to almost every university. Unfortunately, a major limiting factor to maintain and update the learning material is the amount of work, time and resources needed. In this perspective, a French national university network was initiated in 2011 to build mutualised online teaching pathology modules with clinical cases and tests. This network has been extended to an international level in 2012-2014 (Quebec, Switzerland and Ivory Coast). Method: One of the first steps of the international project was to build a learning module on inflammatory skin pathology intended for interns and residents of pathology and dermatology. A pathology resident from Quebec spent 6 weeks in France and Switzerland to develop the contents and build the module on an e-learning Moodle platform (http: //moodle.sorbonne-paris-cite.fr) under the supervision of two dermatopathologists (BV, MB). The learning module contains text, interactive clinical cases, tests with feedback, whole slides images (WSI), images and clinical photographs. For that module, the virtual slides are decentralized in 2 universities (Bordeaux and Paris 7). Each university is responsible of its own slide scanning, image storage and online display with virtual slide viewers. Results: The module on inflammatory skin pathology includes more than 50 web pages with French original content, tests and clinical cases, links to over 45 WSI and more than 50 micro and clinical photographs. The whole learning module is currently being revised by four dermatopathologists and two senior pathologists. It will be accessible to interns and residents in spring 2014. The experience and knowledge gained from that work will be transferred to the next international fellowship intern whose work will be aimed at creating lung and breast pathology learning modules. Conclusion: The challenges of sustaining a project of this scope are numerous. The technical aspect of whole-slide imaging and storage needs to be developed by each university or group. The content needs to be regularly updated, completed and its use and existence needs to be promoted by the different actors in pathology. Of the great benefits of that kind of project are the international partnerships and connections that have been established between numerous Frenchspeaking universities and pathologists with the common goals of promoting education in pathology and the use of technology including whole slide imaging. * The Moodle website is hosted by PRES Sorbonne Paris Cité, and financial supports for hardware have been obtained from UNF3S (http://www.unf3s.org/) and PRES Sorbonne Paris Cité. Financial support for international fellowships has been obtained from CFQCU (http://www.cfqcu.org/).

Statistical assessment of dataset shift and model portability in multi-angle in-track image acquisitions

In this study we propose an evaluation of the angular effects altering the spectral response of the land-cover over multi-angle remote sensing image acquisitions. The shift in the statistical distribution of the pixels observed in an in-track sequence of WorldView-2 images is analyzed by means of a kernel-based measure of distance between probability distributions. Afterwards, the portability of supervised classifiers across the sequence is investigated by looking at the evolution of the classification accuracy with respect to the changing observation angle. In this context, the efficiency of various physically and statistically based preprocessing methods in obtaining angle-invariant data spaces is compared and possible synergies are discussed.

Recent advances in remote sensing image processing

Remote sensing image processing is nowadays a mature research area. The techniques developed in the field allow many real-life applications with great societal value. For instance, urban monitoring, fire detection or flood prediction can have a great impact on economical and environmental issues. To attain such objectives, the remote sensing community has turned into a multidisciplinary field of science that embraces physics, signal theory, computer science, electronics, and communications. From a machine learning and signal/image processing point of view, all the applications are tackled under specific formalisms, such as classification and clustering, regression and function approximation, image coding, restoration and enhancement, source unmixing, data fusion or feature selection and extraction. This paper serves as a survey of methods and applications, and reviews the last methodological advances in remote sensing image processing.