Efficient Data Generation for Stroke Classification via Multilayer Perceptron

The aim of this paper is to overcome one of the main problems of machine learning when it faces the medical world: the need of a large amount of data. Through the distorted Born approximation, the scattering parameters and the dielectric contrast in the domain of interest are linked by a linearized integral operator. This method allows to generate a large dataset in a short time. In this work, machine learning is exploited to classify brain stroke presence, typology and position. The classifier model is based on the multilayer perceptron algorithm and it is used firstly for validation and then with a testing set composed by full-wave simulations. In both cases, the model reaches very high level of accuracy.