Data augmentation study for rare diseases assessment with Deep Learning: Confocal Imaging analysis of Congenital Muscular Dystrophy
- Frías, M.
- Jiménez Mallebrera, C.
- Badosa, C.
- Porta, J.M.
- Roldán, M.
- Joaquín Roca González (coord.)
- Dolores Ojados González (coord.)
- Juan Suardíaz Muro (coord.)
Publisher: Universidad Politécnica de Cartagena
ISBN: 978-84-17853-76-1
Year of publication: 2023
Pages: 460-463
Congress: Congreso Anual de la Sociedad Española de Ingeniería Biomédica. CASEIB (41. 2023. Cartagena)
Type: Conference paper
Sustainable development goals
Abstract
Artificial Intelligence (AI) algorithms are widely used in healthcare nowadays. However, there are still fields where the application of these technologies could be challenging, such as rare diseases. In these cases, the main challenge arises from the reduced size of the available data sets. This paper proposes a data augmentation pipeline to address this challenge when using a Deep Learning (DL) algorithm to assess fibroblast cultures from skin biopsies to diagnose Collagen VI-related Congenital Muscular Dystrophy (COL6-CMD). Different data augmentation schemes are described in the literature. However, they must be used cautiously since they might result in overfitting. The results presented in this paper demonstrate that the right combination of data augmentation techniques results in a high diagnostic accuracy (up to 75.35% for the best approach) even with a scarce amount of data.