MétaCan
Menu
Back to cohort
Record W2897793588

Un nouveau descripteur iconique topologique pour l'appariement d'images de structures anatomiques

2017· article· fr· W2897793588 on OpenAlexfundno aff
Nicolas Piché

Bibliographic record

VenuePolyPublie (École Polytechnique de Montréal) · 2017
Typearticle
Languagefr
FieldComputer Science
TopicImage Retrieval and Classification Techniques
Canadian institutionsnot available
FundersUniversité de Sherbrooke
KeywordsHumanitiesArt
DOInot available

Abstract

fetched live from OpenAlex

L’objectif principal de notre projet de recherche est de creer un descripteur iconique topologique qui permet de faire des appariements de structures anatomiques en se basant sur l’organisation topologique des intensites dans une image. Ce descripteur doit definir un espace metrique afin de pouvoir effectuer des comparaisons efficaces. Les descripteurs actuellement disponibles ne font pas intervenir les composantes distantes d’images pour decrire un lieu dans l’image. De facon inherente, ils sont locaux dans leurs analyses des caracteristiques de l’image. Bien sur, certains d’entre eux incluent un certain voisinage, mais aucun des descripteurs que nous avons analyses incorpore l’information de l’image dans son ensemble. Pour arriver a nos fins, nous proposons une representation sous forme de graphe de l’espace image. Pour construire ce graphe nous proposons deux approches. La premiere considere l’image comme un graphe implicite dans lequel les noeuds sont les pixels de l’image et la connectivite des noeuds du graphe represente le voisinage des pixels. La deuxieme est, quant a elle, plus elaboree et est bâtie en analysant les regions definies par une sursegmentation de l’image. Dans celle-ci, les regions sont les noeuds du graphe et les aretes du graphe representent la connectivite des regions. Pour decrire un lieu dans l’image, nous proposons d’extraire un arbre du graphe image pour chacun des endroits d’interet. Pour extraire cet arbre, nous utilisons un processus de minimisation de la rencontre des gradients, autrement dit, les branches de cet arbre poussent de maniere a reduire les gradients traverses. Cet arbre est une vue de l’image telle que percue par le lieu etudie. L’arbre ainsi concu est notre descripteur sous sa forme la plus complete. Pour etre en mesure de comparer et de conserver ces arbres, nous en extrayons une representation en histogramme. Nous avons etudie plusieurs versions, sous forme d’histogrammes, des arbres, chaque version etant plus ou moins riche en information. Le passage a une telle representation nous permet une sauvegarde efficace et configurable. En effet, si nous desirons inclure plus d’information relative a l’arbre dans l’histogramme le representant, nous pouvons utiliser un histogramme ayant plus de classes. Ce passage a une representation en histogramme nous permet egalement l’utilisation de mesures de metriques fiables pour quantifier la distance entre deux de nos descripteurs. Les resultats obtenus montrent que nos descripteurs sont en mesure de faire l’appariement d’organes presents dans des acquisitions de tomographie axiale (CT), et ce, dans la meme image ou dans des images distinctes. Pour mesurer la capacite d’appariement des descripteurs que l’on propose, nous avons propose un algorithme base sur les rangs medians de distances. Cet algorithme nous permet d’evaluer la capacite de correspondance en tenant compte de tous les descripteurs des images. Les resultats que nous avons obtenus sont tres prometteurs si on tient compte des difficultes de la tâche, liees, entre autres, au fait que l’appariement soit base sur une seule instance d’un descripteur et au manque de structure apparente dans les images 2D. En effet, compte tenu des temps de calculs considerables, nous avons restreint notre analyse des appariements des descripteurs a des images 2D de thorax. Les structures anatomiques etant comprehensibles seulement lorsque l’on effectue une analyse 3D, la qualite des resultats obtenus est impressionnante. Le passage a des descripteurs 3D ne demande aucune modification des algorithmes utilises et devrait nous fournir des resultats d’appariement des structures anatomiques encore meilleurs.----------ABSTRACT: The main objective of our research project is to create a topological iconic descriptor that allows for the matching of anatomical structures based on the topological organization of intensities in an image. This descriptor must define a metric space in order to be able to make efficient comparisons. Currently available descriptors do not involve remote image components to describe a location in the image. Inherently, they are local in their analyzes of the characteristics of the image. Of course, some of them include some neighborhood, but none of the descriptors we have analyzed incorporate the image information as a whole. To arrive at our ends, we propose a representation in the form of a graph of the image space. To build this graph we propose two approaches. The first considers the image as an implicit graph in which the nodes are the pixels of the image and the connectivity of the nodes of the graph represents the neighborhood of the pixels. The second is, in turn, more elaborate and is built by analyzing the regions defined by an over-segmentation of the image. In this one, the regions are the nodes of the graph and the edges of the graph represent the connectivity of the regions. To describe a region in the image, we propose to extract a tree from the image graph for each of the regions of interest. To extract this tree, we use a process of minimizing the gradients encounter, that is, the branches of this tree grow in order to reduce the gradients crossed. This tree is a view of the image as perceived by the region studied. The tree thus conceived is our descriptor in its most complete form. To be able to compare and preserve these trees, we extract a histogram representation. We have studied several versions, in the form of histograms, of the trees, each version being more or less rich in information. Moving to such a representation allows us an efficient and configurable storage. Indeed, if we want to include more information about the tree in the histogram representing it, we can use a histogram with more classes. This shift to a histogram representation also allows us to use reliable metric measurements to quantify the distance between two of our descriptors. The results show that our descriptors are able to match organs present in axial tomography (CT) acquisitions, in the same image or in separate images. To measure the ability to match the descriptors proposed, we proposed an algorithm based on the median ranks of distances. This algorithm allows us to evaluate the matching capability taking into account all the descriptors of the images. The results that we obtained are very promising if we take into account the difficulties of the task. Indeed, given the considerable computation time, we have restricted our analysis of descriptor matches to 2D images of thorax. The anatomical structures are understandable only when performing a 3D analysis, the quality of the results obtained is impressive. The shift to 3D descriptors requires no modification of the algorithms used and should provide us with even better anatomical structure matching results.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.

How this classification was reachedexpand

Full frame distilled prediction

Teacher imitation

Not calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.

metaresearch head score (Codex)0.003
metaresearch head score (Gemma)0.001
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesMeta-epidemiology (narrow), Science and technology studies, Scholarly communication, Research integrity
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Theoretical or conceptual · Consensus signal: none
GenreCandidate signal: Methods · Consensus signal: none
Teacher disagreement score0.637
Threshold uncertainty score1.000

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0030.001
Meta-epidemiology (narrow)0.0010.001
Meta-epidemiology (broad)0.0010.001
Bibliometrics0.0010.000
Science and technology studies0.0020.001
Scholarly communication0.0040.003
Open science0.0050.002
Research integrity0.0010.002
Insufficient payload (model declined to judge)0.0000.000

Machine scores (provisional)

The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.

Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.

Opus teacher head0.023
GPT teacher head0.266
Teacher spread0.243 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one teacher head, not a consensus.

Study designTheoretical or conceptual
Domainnot available
GenreMethods

How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".

Quick stats

Citations0
Published2017
Admission routes1
Has abstractyes

Explore more

Same venuePolyPublie (École Polytechnique de Montréal)Same topicImage Retrieval and Classification TechniquesFrench-language works237,207