Editorial Universidad Tecnológica de Pereira

SINOPSIS DEL LIBRO

This work presents a prototype based on advanced techniques for semantic segmentation and classification of chest radiographs using deep learning. An architecture based on a variant of the DenseNet network is proposed, enabling the simultaneous and accurate segmentation of lung structures and classification of pathologies present in chest radiographs. The architecture is based on an encoder–decoder configuration similar to a U-Net architecture. In addition, a customized loss layer is implemented, applying the concepts of function weighting and considering the homoscedastic uncertainty of each of the tasks. This allows for jointly optimizing both tasks, considering the different evaluation metrics used in each one, to serve as a tool to support the diagnosis of pulmonary pathologies.

METADATOS

URL: https://hdl.handle.net/11059/16975

DOI: https://doi.org/10.22517/9786285011184

e-ISBN: 978-628-501-118-4

CARACTERÍSTICAS:

Número de páginas: 138