Statistics in science and in society: From a state-of-the-art to a new research agenda

Statistics occupies a prominent role in science and citizens' daily life. This article provides a state-of-the-art of the problems associated with statistics in science and in society, structured along the three paradigms defined by Bauer, Allum and Miller (2007). It explores in more detail medicine and public understanding of science on the one hand, and risks and surveys on the other. Statistics has received a good deal of attention; however, very often handled in terms of deficit - either of scientists or of citizens. Many tools have been proposed to improve statistical literacy, the image of and trust in statistics, but with little understanding of their roots, with little coordination among stakeholders and with few assessments of impacts. These deficiencies represent as many new and promising directions in which the PUS research agenda could be expanded.

Conditioning of multiple-point statistics facies simulations to tomographic images

Geophysical tomography captures the spatial distribution of the underlying geophysical property at a relatively high resolution, but the tomographic images tend to be blurred representations of reality and generally fail to reproduce sharp interfaces. Such models may cause significant bias when taken as a basis for predictive flow and transport modeling and are unsuitable for uncertainty assessment. We present a methodology in which tomograms are used to condition multiple-point statistics (MPS) simulations. A large set of geologically reasonable facies realizations and their corresponding synthetically calculated cross-hole radar tomograms are used as a training image. The training image is scanned with a direct sampling algorithm for patterns in the conditioning tomogram, while accounting for the spatially varying resolution of the tomograms. In a post-processing step, only those conditional simulations that predicted the radar traveltimes within the expected data error levels are accepted. The methodology is demonstrated on a two-facies example featuring channels and an aquifer analog of alluvial sedimentary structures with five facies. For both cases, MPS simulations exhibit the sharp interfaces and the geological patterns found in the training image. Compared to unconditioned MPS simulations, the uncertainty in transport predictions is markedly decreased for simulations conditioned to tomograms. As an improvement to other approaches relying on classical smoothness-constrained geophysical tomography, the proposed method allows for: (1) reproduction of sharp interfaces, (2) incorporation of realistic geological constraints and (3) generation of multiple realizations that enables uncertainty assessment.

Urban image classification with semisupervised multiscale cluster kernels

This paper presents a semisupervised support vector machine (SVM) that integrates the information of both labeled and unlabeled pixels efficiently. Method's performance is illustrated in the relevant problem of very high resolution image classification of urban areas. The SVM is trained with the linear combination of two kernels: a base kernel working only with labeled examples is deformed by a likelihood kernel encoding similarities between labeled and unlabeled examples. Results obtained on very high resolution (VHR) multispectral and hyperspectral images show the relevance of the method in the context of urban image classification. Also, its simplicity and the few parameters involved make the method versatile and workable by unexperienced users.

Scan statistics analysis of forest fire clusters

Spatio-temporal clusters in 1997?2003 fire sequences of Tuscany region (central Italy) have been identified and analysed by using the scan statistic, a method which was devised to evidence clusters in epidemiology. Results showed that the method is reliable to find clusters of events and to evaluate their significance via Monte Carlo replication. The evaluation of the presence of spatial and temporal patterns in fire occurrence and their significance could have a great impact in forthcoming studies on fire occurrences prediction.

Skeletal Muscle Mitochondrial and Lipid Droplet Volume Density: Validity of Electron Microscopy Point-counting Measurements

Mitochondrial (M) and lipid droplet (L) volume density (vd) are often used in exercise research. Vd is the volume of muscle occupied by M and L. The means of calculating these percents are accomplished by applying a grid to a 2D image taken with transmission electron microscopy; however, it is not known which grid best predicts these values. PURPOSE: To determine the grid with the least variability of Mvd and Lvd in human skeletal muscle. METHODS: Muscle biopsies were taken from vastus lateralis of 10 healthy adults, trained (N=6) and untrained (N=4). Samples of 5-10mg were fixed in 2.5% glutaraldehyde and embedded in EPON. Longitudinal sections of 60 nm were cut and 20 images were taken at random at 33,000x magnification. Vd was calculated as the number of times M or L touched two intersecting grid lines (called a point) divided by the total number of points using 3 different sizes of grids with squares of 1000x1000nm sides (corresponding to 1µm2), 500x500nm (0.25µm2) and 250x250nm (0.0625µm2). Statistics included coefficient of variation (CV), 1 way-BS ANOVA and spearman correlations. RESULTS: Mean age was 67 ± 4 yo, mean VO2peak 2.29 ± 0.70 L/min and mean BMI 25.1 ± 3.7 kg/m2. Mean Mvd was 6.39% ± 0.71 for the 1000nm squares, 6.01% ± 0.70 for the 500nm and 6.37% ± 0.80 for the 250nm. Lvd was 1.28% ± 0.03 for the 1000nm, 1.41% ± 0.02 for the 500nm and 1.38% ± 0.02 for the 250nm. The mean CV of the three grids was 6.65% ±1.15 for Mvd with no significant differences between grids (P>0.05). Mean CV for Lvd was 13.83% ± 3.51, with a significant difference between the 1000nm squares and the two other grids (P<0.05). The 500nm squares grid showed the least variability between subjects. Mvd showed a positive correlation with VO2peak (r = 0.89, p < 0.05) but not with weight, height, or age. No correlations were found with Lvd. CONCLUSION: Different size grids have different variability in assessing skeletal muscle Mvd and Lvd. The grid size of 500x500nm (240 points) was more reliable than 1000x1000nm (56 points). 250x250nm (1023 points) did not show better reliability compared with the 500x500nm, but was more time consuming. Thus, choosing a grid with square size of 500x500nm seems the best option. This is particularly relevant as most grids used in the literature are either 100 points or 400 points without clear information on their square size.

How to set up and apply reference levels in fluoroscopy at a national level.

A nationwide survey was launched to investigate the use of fluoroscopy and establish national reference levels (RL) for dose-intensive procedures. The 2-year investigation covered five radiology and nine cardiology departments in public hospitals and private clinics, and focused on 12 examination types: 6 diagnostic and 6 interventional. A total of 1,000 examinations was registered. Information including the fluoroscopy time (T), the number of frames (N) and the dose-area product (DAP) was provided. The data set was used to establish the distributions of T, N and the DAP and the associated RL values. The examinations were pooled to improve the statistics. A wide variation in dose and image quality in fixed geometry was observed. As an example, the skin dose rate for abdominal examinations varied in the range of 10 to 45 mGy/min for comparable image quality. A wide variability was found for several types of examinations, mainly complex ones. DAP RLs of 210, 125, 80, 240, 440 and 110 Gy cm2 were established for lower limb and iliac angiography, cerebral angiography, coronary angiography, biliary drainage and stenting, cerebral embolization and PTCA, respectively. The RL values established are compared to the data published in the literature.

Generating controlled image sets in cognitive neuroscience research.

The investigation of perceptual and cognitive functions with non-invasive brain imaging methods critically depends on the careful selection of stimuli for use in experiments. For example, it must be verified that any observed effects follow from the parameter of interest (e.g. semantic category) rather than other low-level physical features (e.g. luminance, or spectral properties). Otherwise, interpretation of results is confounded. Often, researchers circumvent this issue by including additional control conditions or tasks, both of which are flawed and also prolong experiments. Here, we present some new approaches for controlling classes of stimuli intended for use in cognitive neuroscience, however these methods can be readily extrapolated to other applications and stimulus modalities. Our approach is comprised of two levels. The first level aims at equalizing individual stimuli in terms of their mean luminance. Each data point in the stimulus is adjusted to a standardized value based on a standard value across the stimulus battery. The second level analyzes two populations of stimuli along their spectral properties (i.e. spatial frequency) using a dissimilarity metric that equals the root mean square of the distance between two populations of objects as a function of spatial frequency along x- and y-dimensions of the image. Randomized permutations are used to obtain a minimal value between the populations to minimize, in a completely data-driven manner, the spectral differences between image sets. While another paper in this issue applies these methods in the case of acoustic stimuli (Aeschlimann et al., Brain Topogr 2008), we illustrate this approach here in detail for complex visual stimuli.

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.