京都大学 / 0048 / 新制・課程博士 / 博士(医学) / 甲第22325号 / 医博第4566号 / 新制||医||1041(附属図書館) / 京都大学大学院医学研究科医学専攻 / (主査)教授 妹尾 浩, 教授 萩原 正敏, 教授 小川 修 / 学位規則第4条第1項該当 / Doctor of Medical Science / Kyoto University / DFAM

Dans cette thèse, nous proposons une approche combinatoire novatrice fondée sur les matroïdes orientés pour l'étude quantitative de la forme de structures anatomiques 3D. Nous nous basons sur des points de repère qui ont été préalablement localisés par des experts sur la structure anatomique étudiée. La nouveauté de cette méthode provient de l'utilisation de matroïdes orientés. Ces outils mathématiques nous permettent de coder la position relative des points de repère de façon purement combinatoire, c'est-à-dire sans utiliser de notions d'angles ou de distances, en associant un signe (0, + ou -) à chaque sous-ensemble de (d+1) points de repère où d est la dimension de l'espace (dans notre cas 2 ou 3). Dans une première partie, nous supposons qu'il existe des contraintes d'ordres sur chaque axe de coordonnée pour les points de repère. Nous obtenons alors une caractérisation (en dimension 2 et 3) des sous-ensembles de points de repère dont le signe associé est constant, quelles que soient les valeurs des coordonnées satisfaisant les contraintes d'ordre. Dans une deuxième partie, nous cherchons à classifier un ensemble de modèles 3D, en les codant au préalable par ces listes de signes. Nous analysons d'abord comment s'appliquent les algorithmes de clustering classiques, puis nous décrivons comment caractériser des classes de façon directe, à l'aide des signes associés à quelques sous-ensembles de points de repère. Dans une troisième partie, nous détaillons les algorithmes et l'implémentation en machine de cette nouvelle méthode de morphométrie afin de pouvoir l'appliquer à des données réelles. Dans la dernière partie, nous appliquons la méthode sur trois bases de données composées chacune de plusieurs dizaines de points de repères relevés sur plusieurs dizaines à plusieurs centaines de structures crâniennes pour des applications en anatomie comparée, en orthodontie et sur des cas cliniques d'enfants présentant des déformations cranio-faciales. / In this thesis, we propose an innovative combinatorial method based on oriented matroids for the quantitative study of the shape of 3D anatomical structures. We rely on landmarks which were previously defined by experts on the studied anatomical structure. The novelty of this method results from the use of oriented matroids. These mathematical tools allow us to encode the relative position of landmarks in a purely combinatorial way, that is without using concepts of angles or distances, by associating a sign (0, + or -) for each subset of (d+1) landmarks where d is the dimension of space (in our case 2 or 3). In the first part, we assume that there exist constraints of orders on each coordinate axis for the landmarks. We obtain a characterization (in dimension 2 and 3) of the subsets of landmarks of which the associated sign is constant, regardless of the values of the coordinates satisfying the constraints of order. In a second part, we try to classify a set of 3D models, encoding in advance by these lists of signs. We first analyze how to apply classic clustering algorithms, and then describe how to characterize the classes directly, using signs associated with some subsets of landmarks. In the third part, we explain the algorithms and the implementation of this new morphometry method in order to apply it to real data. In the last part, we apply the method to three databases each consisting of several dozens of points defined on several dozens to several hundreds of cranial structures for applications in comparative anatomy, in orthodontics and on clinical cases of children with craniofacial deformities.

Computer-assisted three-dimensional (3D) mapping using stereo and multi-image (“softcopy”) photogrammetry is shown to enhance the visual interpretation of geomorphology in steep terrain with the direct benefit of greater locational accuracy than traditional manual mapping. This would benefit multi-parameter correlations between terrain attributes and landslide distribution in both direct and indirect forms of landslide hazard assessment. Case studies involve synthetic models of a landslide, and field studies of a rock slope and steep undeveloped hillsides with both recently formed and partly degraded, old landslide scars. Diagnostic 3D morphology was generated semi-automatically both using a terrain-following cursor under stereo-viewing and from high resolution digital elevation models created using area-based image correlation, further processed with curvature algorithms. Laboratory-based studies quantify limitations of area-based image correlation for measurement of 3D points on planar surfaces with varying camera orientations. The accuracy of point measurement is shown to be non-linear with limiting conditions created by both narrow and wide camera angles and moderate obliquity of the target plane. Analysis of the results with the planar surface highlighted problems with the controlling parameters of the area-based image correlation process when used for generating DEMs from images obtained with a low-cost digital camera. Although the specific cause of the phase-wrapped image artefacts identified was not found, the procedure would form a suitable method for testing image correlation software, as these artefacts may not be obvious in DEMs of non-planar surfaces. / Modelling of synthetic landslides shows that Fast Fourier Transforms are an efficient method for removing noise, as produced by errors in measurement of individual DEM points, enabling diagnostic morphological terrain elements to be extracted. Component landforms within landslides are complex entities and conversion of the automatically-defined morphology into geomorphology was only achieved with manual interpretation; however, this interpretation was facilitated by softcopy-driven stereo viewing of the morphological entities across the hillsides.In the final case study of a large landslide within a man-made slope, landslide displacements were measured using a photogrammetric model consisting of 79 images captured with a helicopter-borne, hand-held, small format digital camera. Displacement vectors and a thematic geomorphological map were superimposed over an animated, 3D photo-textured model to aid non-stereo visualisation and communication of results.