The advent of medical imaging and new operating techniques has revolutionized the working methods of medical professionals. This change requires physicians and surgeons to undergo additional training. This is why the development of appropriate tools, like medical simulators, is of paramount importance. Within this context, we focus on physical modeling of soft tissue deformations and collision detection in virtual environments. First, we propose an alternative model based on Pascal''s principle for soft tissue simulation, by successively presenting the aspects related to the formulation of the model, the numerical resolution of the model and the treatment of the physical interactions. Following this, we present some improved collision detection algorithms developed in our work to treat virtual interactions. These algorithms have better numerical robustness, are optimized and share a unified underlying data structure. We apply our results within the framework of an echographic simulator of the human thigh and a simulator for the arthroscopy reconstruction of the anterior cruciate ligament of the knee.