Venous thrombosis supervised image indexing and fuzzy retrieval.

Clinical assessment of venous thrombosis (VT) is essential to evaluate the risk of size increase or embolism. Analyses like echogenecity and echostructure characterization, examine ancillary evidence to improve diagnosis. However, such analyses are inherently uncertain and operator dependent, adding enormous complexity to the task of indexing diagnosed images for medical practice support, by retrieving similar images, or to exploit electronic patient record repositories for data mining. This paper proposes a VT ultrasound image indexing and retrieval approach, which shows the suitability of neural network VT characterization, combined with a fuzzy similarity. Three types of image descriptors (sliding window, wavelet coefficients energy and co-occurrence matrix), are processed by three different neural networks, producing equivalent VT characterizations. Resulting values are projected on fuzzy membership functions and then compared with the fuzzy similarity. Compared to nominal and Euclidean distances, an experimental validation indicates that the fuzzy similarity increases image retrieval precision beyond the identification of images that belong to the same diagnostic class, taking into account the characterization result uncertainty, and allowing the user to privilege any particular feature.

Volume Calculation of Venous Thrombosis Using 2D Ultrasound Images.

Venous thrombosis screening exams use 2D ultrasound images, from which medical experts obtain a rough idea of the thrombosis aspect and infer an approximate volume. Such estimation is essential to follow up the thrombosis evolution. This paper proposes a method to calculate venous thrombosis volume from non-parallel 2D ultrasound images, taking advantage of a priori knowledge about the thrombosis shape. An interactive ellipse fitting contour segmentation extracts the 2D thrombosis contours. Then, a Delaunay triangulation is applied to the set of 2D segmented contours positioned in 3D, and the area that each contour defines, to obtain a global thrombosis 3D surface reconstruction, with a dense triangulation inside the contours. Volume is calculated from the obtained surface and contours triangulation, using a maximum unit normal component approach. Preliminary results obtained on 3 plastic phantoms and 3 in vitro venous thromboses, as well as one in vivo case are presented and discussed. An error rate of volume estimation inferior to 4,5% for the plastic phantoms, and 3,5% for the in vitro venous thromboses was obtained.

Information fusion: application to data and model fusion for ultrasound image segmentation.

Nowadays, information fusion constitutes a challenging research topic. Our study proposes to achieve the fusion of several knowledge sources. This, in order to detect the esophagus inner wall from ultrasound medical images. After a brief description of information fusion concepts, we propose a system architecture including both model and data fusion. The data fusion is accomplished using fuzzy modeling, which can be seen as a monosensor/multiple sources data fusion system. The model fusion is performed using a full-adapted snake theory, which projects the fuzzy decision into the binary decision space.

Toward the border between neural and Markovian paradigms.

A new tendency in the design of modern signal processing methods is the creation of hybrid algorithms. This paper gives an overview of different signal processing algorithms situated halfway between Markovian and neural paradigms. A new systematic way to classify these algorithms is proposed. Four specific classes of models are described. The first one is made up of algorithms based upon either one of the two paradigms, but including some parts of the other one. The second class includes algorithms proposing a parallel or sequential cooperation of two independent Markovian and neural parts. The third class tends to show Markov models (MMs) as a special case of neural networks (NNs), or conversely NNs as a special case of MMs. These algorithms concentrate mainly on bringing together respective learning methods. The fourth class of algorithms are hybrids, neither purely Markovian nor neural. They can be seen as belonging to a more general class of models, presenting features from both paradigms. The first two classes essentially include models with structural modifications, while two later classes propose algorithmic modifications. For the sake of clarity, only main mathematical formulas are given. Specific applications are intentionally avoided to give a wider view of the subject. The references provide more details for interested readers.