Disperse multiphase flows such as bubbly flows have important applications in areas such as cavitation, micro-reactors and fuel cells. Such flows are highly complex, yet there are limited methods for obtaining experimental data in environments with limited optical access. This project aims to develop a novel diagnostic tool to obtain the flow topology as well as the bubble sizes in the flow. This is accomplished through the development of a hybrid image processing approach that combines deep learning and inverse optimization techniques based on the scattering physics of light on particles in the Lorentz-Mie theory.