Barbara Romaniuk

Shape variability and spatial relationships modeling in statistical pattern recognition

We focus on the problem of shape variability modeling in statistical pattern recognition. We present a nonlinear statistical model invariant to affine transformations. This model is learned on an ordinate set of points. The concept of relations between model components is also taken in account. This model is used to find curves and points partially occulted in the image. We present its application on medical imaging in cephalometry.

3D medical image segmentation approach based on multi-label front propagation

Many practical applications in the field of medical image processing require robust and valid 3D image segmentation results. In this paper, we present a semi-automatic iterative segmentation approach for 3D medical image by combining a 2D boundary tracking algorithm and a boundary mapping process. Upon each of the consecutive slice, the boundary tracking process is accomplished in an alternate procedure of the morphological dilatation and the multi-label front propagation. The multi-label front propagation method is developed based on the minimal path theory and fast sweeping evolution method to ensure the efficiency, and speed of the boundary tracking algorithm. This 3D image segmentation approach can easily extract the close and smooth boundary of the desired object from a 2D medical image series. This approach is efficient and reliable, and requires very limited user intervention. Some experimental results are also presented to demonstrate the efficiency of this approach.

Computer assisted landmarking of cephalometric radiographs

We address the problem of finding an initial estimation of the location of landmarks on an image, when the landmarks are difficult to distinguish on the image and when the locations are dependent together from external forces such as growth. Our method solves the problem using an adaptive coordinate space where locations are registered. In this space, variability is greatly reduced. A training set is observed to build automatically a mean and a variability model of the landmarks. This model is used to predict the initial estimation on a new image. This method is applied to the difficult problem of the interpretation of cephalograms, with good results.

Missing data estimation using polynomial kernels

In this paper, we deal with the problem of partially observed objects. These objects are defined by a set of points and their shape variations are represented by a statistical model. We present two models in this paper: a linear model based on PCA and a non-linear model based on KPCA. The present work attempts to localize of non visible parts of an object, from the visible part and from the model, using the variability represented by the models. Both are applied to synthesis data and to cephalometric data with good results.

Contour tracking by minimal cost path approach: application to cephalometry

In this paper, a minimal cost approach is used for contour tracking with a good robustness. Dynamic programming was chosen for its efficiency. This general method is applied to the extraction of the cranial contour on high-resolution X-Ray images. As a first step for automated localization of cephalometric points, an ellipse is then fitted on the extracted contour. This method was tested on 424 X-Ray images, with different acquisition parameters.

First steps toward automatic location of landmarks on X-ray images

We address the problem of locating some anatomical bone structures on lateral cranial X-ray images. These structures are landmarks used in orthodontic therapy. The main problem in this pattern recognition application is that the landmarks are difficult to distinguish on images even for the human expert, because of lateral projection of the X-ray process. We propose a 3 steps approach: the first step provides a statistical estimation of the landmarks, using an adaptive coordinates space; the second step computes a region of interest around the estimated landmark; and in the third step, each landmark is precisely located using its anatomical definition. This paper describes the two first generic steps and gives examples of the last step for two anatomical points.

Multicriteria 3D PET image segmentation

The analysis of images acquired with Positron Emission Tomography (PET) is challenging. In particular, there is no consensus on the best criterion to quantify the metabolic activity for lesion detection and segmentation purposes. Based on this consideration, we propose a versatile knowledge-based segmentation methodology for 3D PET imaging. In contrast to previous methods, an arbitrary number of quantitative criteria can be involved and the experts behaviour learned and reproduced in order to guide the segmentation process. The classification part of the scheme relies on example-based learning strategies, allowing interactive example definition and more generally incremental refinement. The image processing part relies on hierarchical segmentation, allowing vectorial attribute handling. Preliminary results on synthetic and real images confirm the relevance of this methodology, both as a segmentation approach and as an experimental framework for criteria evaluation.

Hierarchical forest attributes for multimodal tumor segmentation on FDG-PET/contrast-enhanced CT

Accurate tumor volume delineation is a crucial step for disease assessment, treatment planning and monitoring of several kinds of cancers. However, this process is complex due to variations in tumors properties. Recently, some methods have been proposed for taking advantage of the spatial and spectral information carried by coupled modalities (e.g., PET-CT, MRI-PET). Simultaneously, the development of attributebased approaches has contributed to improve PET image analysis. In this work, we aim at developing a coupled multimodal / attribute-based approach for image segmentation. Our proposal is to take advantage of hierarchical image models for determining relevant and specific attribute from each modality. These attributes then allow us to define a unique, semantic vectorial image. Sequentially, this image can be processed by a standard segmentation method, in our case a random-walker approach, for segmenting tumors based on their intrinsic attribute-based properties. Experimental results emphasize the relevance of computing region-based attributes from both modalities.

Incremental plane-based reconstruction and exploration of buildings in old postcards

We present an approach for 3D reconstruction of a city model over the time from a collection of old postcards of the city of Reims. The planar structure of the buildings facades constraints the dense reconstruction of the city. We use a feature matching technique while proposing the registration of facades in the images and there use for the reasoning about there visibility in the images. This system is semi-automatic, it requires a user control in the complicated case where no matches are found to link an image to the dataset. The image data set is sparse and the urban space evolves over time.

Partially observed objects localization with PCA and KPCA models

We deal with the problem of partially observed objects. These objects are defined by sets of points and their shape variations are represented by a statistical model. We present two models: a linear model based on PCA and a non-linear model based on KPCA (kernel PCA). The present work attempts to localize non visible parts of an object from visible parts and from the model, explicitly. using the variability represented by the model. Both are applied to the cephalometric problem with good results.