ProGAN per a la generació de ressonància magnètica de mama

Abstract

[en] Currently, multiple machine learning models have been implemented to assist radiology staff in the decision-making process during the analysis of medical images for cancerous lesions, both benign and malignant. Despite this, such models mostly cannot be trained with sufficient images, due to the ethical and legal complications in obtaining the images. For this purpose, the objective of this study has been defined as the implementation of a deep learning model system, more specifically a Progressively Growing GAN, by means of which two-dimensional medical images of breast MRIs can be generated from a set of other two-dimensional medical images that were provided by the Consorci Corporació Sanitària Parc Taulí. Despite the lack of radiological knowledge and computational capacity, the results of the model can be seen as satisfactory, because the structure of the breasts is preserved, together with an approximation of the breast tissues. However, if these images are intended to be used in another project, the model should be trained on a machine with higher computational capacity by which the model can be trained for a longer time.