Advanced modeling of the childbirth system using different deep learning methods : from fetal skeleton segmentation to real-time soft tissue deformation ; Modélisation avancée du système d'accouchement à l'aide de différentes méthodes d'apprentissage profond : de la segmentation du squelette fœtal à la déformation des tissus mous en temps réel

The annual World Health Organization (WHO) report highlights a staggering statistic of 100 million childbirths occurring worldwide each year. However, a somber reality accompanies this high number, as the death rate reaches approximately half a million. A significant contributing factor to this alarming mortality rate is complicated obstructed labor. Childbirth simulations have been investigated in an attempt to forecast and prevent severe complications in both mothers and fetuses during delivery. Constructing a comprehensive model of the pregnant woman’s pelvic system and the fetal body is crucial in advancing the understanding of childbirth mechanics. However, scientific challenges remain in the realistic representation of the fetus and suitable computational cost and processing speed to deploy the childbirth simulations into the clinical routine practices.This PhD thesis has three original contributions to overcome these challenges: 1) automatic fetal skeleton segmentation into distinct parts using generative adversarial networks (GAN)-based model and 3D point cloud data; 2) prediction of real-time soft tissue deformation using recurrent neural networks (i.e. long short-term memory neural networks (LSTM)) coupled with principal component analysis (PCA)-based learning strategy; and 3) development and evaluation of an outstanding model to simulate real-time deformations of soft tissue using the physics-informed Neural Networks (PINN) and Neural Ordinary Differential Equations (NeuralODE).This thesis opens new avenues in the realistic modeling of the fetal representation and real-time soft tissue deformation toward a next-generation decision support tool for childbirth training and complication simulation ; Le rapport annuel de l'Organisation mondiale de la santé (OMS) met en évidence une statistique stupéfiante de 100 millions d'accouchements se produisant dans le monde chaque année. Cependant, une réalité sombre accompagne ce chiffre élevé, car le taux de mortalité atteint environ un demi-million. Un facteur ...