We propose a new approach for the data assimilation and simulation of anatomically correct left ventricle fluid dynamics based on cardiac magnetic resonance images. The movement of the ventricle is captured by a set of tracking points on the endocardium identified from a time series of magnetic resonance images. The displacement of the endocardium is interpolated using radial basis functions to produce a smooth global deformation field in both space and time. To regularize the motion of the fluid inside the ventricle we impose the displacement on a fictitious elastic structure surrounding the fluid domain, and then solve the problem in a fluid-structure interaction formulation. This allows to provide physiological flow and pressure levels inside the left ventricle. In order to have physically reasonable outflow conditions, we couple the left ventricle to a network of one-dimensional models, where we can simulate the pressure and flow rate waveforms in the major arteries, thus obtaining a geometrical multiscale model. The end result is a baseline model for the left ventricle that captures the basic fluidic phenomena and that can in the future be extended to consider clinical patient-specific studies by integrating more complex models into the multiscale framework.