Descriptive and generative models of the spatial organisation of element-based textures. Towards medical image analysis tasks. – SPIRIT (SPatial InteRactions In Textures

On the descriptive side of the project, we propose implicit and explicit models. First at stationary low-scale, based on spatial satistics models. And second, we also propose multi-scale models, based on fuzzy description of spatial relationships.

On the generative side of the project, we propose two different kind of approaches: by-example approaches and programmable approaches.

We proposed up to now:
- a test method to characterize different tumor growth scenarios in the colorectal cancer context
- a multi-scale descriptive method for the grading of colorectal cancer based on histology images
- two detection methods of fundus pathologies (glaucoma and AMD) based on local texture description
- a descriptive method of visual point set using lattice structures in histology images
- a by-example approach for the synthesis of element-based vector textures
- a programmable approach for the synthesis of element-based vector textures
- a by-example approach for the synthesis of spatially structured drawn doodles

Perspectives are in several domains: computer-aided diagnosis, quantitative bioimaging, computer-aided vector textures design.

- Grading Cancer from Liver Histology Images using Inter and Intra Region Spatial Relations
M. Garnier, M. Alsheh Ali, J. Seguin, N. Mignet, T. Hurtut, L. Wendling
In Proc. IEEE International Conference on Image Analysis and Recognition (ICIAR), 2014

- Automatic Multiresolution Age-related Macular Degeneration Detection from Fundus Images
M. Garnier, T. Hurtut, H. Tahar, F. Cheriet
in Proc. SPIE Medical Imaging, Computer-Aided Diagnosis, 2014

- Glaucoma Detection based on Local Binary Patterns in Fundus Photographs
M. Alsheh Ali, T. Hurtut, T. Faucon, F. Cheriet
in Proc. SPIE Medical Imaging, Computer-Aided Diagnosis, 2014

- Comparison of the Spatial Organization in Colorectal Tumors using Second-Order Statistics and Functional ANOVA
M. Alsheh Ali, J. Seguin, Aurélie Fischer, N. Mignet, L. Wendling, T. Hurtut
in Proc. IEEE Intl. Symposium on Image and Signal Processing and Analysis (ISPA), 2013. (Best Paper Award)

- Visual Point Set Processing with Lattice Structures: Application to Parsimonious Representations of Digital Histopathology Images
N. Loménie
in Proc. Geometric Science of Information (GSI), 2013

- Discrete Texture Design Using a Programmable Approach
H. Loi, T. Hurtut, R. Vergne, J. Thollot
in SIGGRAPH Talks, 2013

- Caractéristiques du second ordre et ANOVA fonctionnelle pour l’étude de l’organisation spatiale de tumeurs colorectales
M. Alsheh Ali, J. Seguin, A. Fischer, N. Mignet, L. Wendling, T. Hurtut
in Proc. Congrés de la Société des Mathématiques Appliquées et Industrielles (SMAI), 2013.

- A Shape-Aware Model for Discrete Texture Synthesis
P.-E. Landes, B. Galerne, T. Hurtut
Computer Graphics Forum (Proc. EGSR'13), vol. 32 (4), 2013

- Synthesizing Structured Doodle Hybrids
T. Hurtut, P.-E. Landes
in SIGGRAPH Asia 2012 (Poster)

- Object Description Based on Spatial Relations between Level-Sets
M. Garnier, T. Hurtut, L. Wendling
in Proc. IEEE Digital Imaging Computing: Techniques and Applications (DICTA), 2012

Texture modeling is a key fundamental research problem in computer vision and graphics systems. It play an important role in many important medical imaging applications. Some of these medical imaging applications are based on a specific kind of textures, made of a limited number of distinct elements (cell, fiber structure, etc.). These elements are spatially arranged following patterns that often reflect pathological behaviors. At the present time, mostly pixel-based spatial relationship models have been proposed. Current models are thus not structured enough to provide high-level semantic cues. The motivation behind this project is to overcome these limitations by proposing texture models that are able to efficiently capture, describe and simulate the relationships existing between the elements in an image.