- Cet évènement est passé
Séminaire IMAGE : « Digital topology constraints in computational anatomy models of embryonic human brains » (Akinobu Shimizu, Tokyo University of Agriculture and Technology), et « On the distribution of texture in the nuclei of follicular cells in malignant lymphoma » (Hidekata Hontani, Nogoya Institute of Technology)
14 septembre 2023 / 14:00
1 ) Digital topology constraints in computational anatomy models of embryonic human brains
Akinobu Shimizu (Tokyo University of Agriculture and Technology)
The human body exhibits nested structures, including ventricles that envelop chorioid plexuses. Employing topological constraints proves valuable in building a computational anatomy model that captures statistical variations within these nested constraints. Diffeomorphism represents a common method for managing such constraints, but it falls short when it comes to depicting the appearance and vanishing of anatomical structures in embryonic human brains, like chorioid plexuses, which emerge post-Carnegie stage 19.
This presentation introduces an approach for depicting the statistical variations in anatomical structures while considering both nested and neighboring constraints. The utilization of a signed distance-based approach enables us to describing the appearance and disappearance of anatomical structures within these constraints. We apply this proposed method to construct a spatio-temporal statistical model encompassing the surfaces of the brain, ventricles, and chorioid plexuses in human embryos.
2) Dynamic PET Image Reconstruction Using Nonnegative Matrix Factorization Incorporated with Deep Image Prior
Hidekata Hontani (Nogoya Institute of Technology)
We propose a method that reconstructs dynamic positron emission tomography (PET) images from given sinograms by using non-negative matrix factorization incorporated with a deep image prior for appropriately constraining the spatial patterns of resultant images. The proposed method can reconstruct dynamic PET images with higher signal- to-noise ratio and blindly decompose an image matrix into pairs of spatial and temporal factors. The former represents homogeneous tissues with different kinetic parameters and the latter represent the time activity curves that are observed in the corresponding homogeneous tissues. We employ U-Nets combined in parallel for deep image prior and each of the U-Nets is used to extract each spatial factor decomposed from the data matrix. Experimental results show that the proposed method outperforms conventional methods and can extract spatial factors that represent the homogeneous tissues.