No abstract

No abstract

This paper explored the hidden biomedical information from knee magnetic resonance (MR) images for osteoarthritis (OA) prediction. We have computed the cartilage damage index (CDI) information from 36 informative locations on tibiofemoral cartilage compartment from 3-D MR imaging and used principal component analysis (PCA) analysis to process the feature set. Four machine learning methods (artificial neural network (ANN), support vector machine, random forest, and naïve Bayes) were employed to predict the progression of OA, which was measured by the change of Kellgren and Lawrence (KL) grade, Joint Space Narrowing on Medial compartment (JSM) grade, and Joint Space Narrowing on Lateral compartment (JSL) grade. To examine the different effects of medial and lateral informative locations, we have divided the 36-D feature set into a 18-D medial feature set and a 18-D lateral feature set and run the experiment on four classifiers separately. Experiment results showed that the medial feature set generated better prediction performance than the lateral feature set, while using the total 36-D feature set generated the best. PCA analysis is helpful in feature space reduction and performance improvement. For KL grade prediction, the best performance was achieved by ANN with AUC = 0.761 and F-measure = 0.714. For JSM grade prediction, the best performance was achieved by random forest with AUC = 0.785 and F-measure = 0.743, while for JSL grade prediction, the best performance was achieved by ANN with AUC = 0.695 and F-measure = 0.796. As experiment results showing that the informative locations on medial compartment provide more distinguishing features than informative locations on the lateral compartment, it could be considered to select more points from the medial compartment while reducing the number of points from the lateral compartment to improve clinical CDI design.

No abstract

No abstract

No abstract

No abstract

We present a method for the automatic estimation of wind directions from synthetic aperture radar (SAR) images of the ocean. The method is based on a wavelet analysis and assumes that the wind direction aligns with boundary-layer atmospheric roll vortices, which often appear as streaks at kilometre scales in SAR images of the ocean, and measures the orientation of the streaks. Unlike estimation methods that use the discrete Fourier transform (DFT), the streaks in SAR images are described quantitatively as a natural output of this method. Furthermore, more optimal wind directions are obtained by comparing the directional orientation of the streaks at different spatial scales. Sub-scenes in which the streaks are too weak to determine wind direction do not return a direction, as governed by a user-selected threshold. Wind directions for these sub-scenes are based on those in neighbouring sub-scenes by using an adaptive smoothing technique. Quality control involves tuning the threshold level. We apply the method to two examples of RADARSAT-1 SAR images. The results are compared with those of a DFT-based wind direction analysis, and it is shown that a robust wind direction field is obtained. Mesoscale wind structures can be described by using a finer computing grid. The estimated wind directions still include a 180°direction ambiguity.Résumé. Nous présentons une méthode pour l'estimation automatique des directions de vent à partir d'images du radar à synthèse d'ouverture (RSO) de l'océan. La méthode est basée sur l'analyse en ondelettes et repose sur la prémisse que la direction du vent s'aligne suivant les vortex des rouleaux à la couche limite de l'atmosphère, qui se manifestent souvent sous forme de stries à l'échelle kilométrique dans les images RSO de l'océan, et mesure l'orientation de ces stries. Contrairement aux méthodes d'estimation utilisant la transformée de Fourier discrète (TFD), les stries dans les images RSO peuvent être décrites quantitativement comme un produit naturel de cette méthode. De plus, des directions plus optimales de vent sont obtenues en comparant l'orientation directionnelle des stries à différentes échelles spatiales. Les sous-scènes dans lesquelles les stries sont trop faibles pour permettre de déterminer la direction du vent ne retournent pas de direction tel que défini par le seuil choisi par l'utilisateur. Les directions de vent pour ces sous-scènes sont basées sur celles des sous-scènes avoisinantes en utilisant une technique adaptative de lissage. Le contrôle de la qualité implique un ajustement du niveau de seuillage. Nous appliquons la méthode à deux exemples d'images RSO de RADARSAT-1. Les résultats sont comparés aux résultats de l'analyse des directions de vent basée sur la TFD et il est démontré qu'il est possible d'obtenir un champ robuste de directions de vent. Des structures de vent à méso-échelle peuvent être décrites en utilisant une grille de calcul plus fine. Les directions de vent estimées comportent toujours une ambiguïté de direction de 180°. [Traduit par la Rédacti...