2021/11/08 by Priscille de Dumast, de Dumast, Priscille, Hamza Kebiri +9
Medicine · Computer Science · #Fetal and Pediatric Neurological Disorders #Domain Adaptation and Few-Shot Learning #Neonatal and fetal brain pathology
paper · pdf · doi:10.48550/arxiv.2111.04737
The quantitative assessment of the developing human brain in utero is crucial\nto fully understand neurodevelopment. Thus, automated multi-tissue fetal brain\nsegmentation algorithms are being developed, which in turn require annotated\ndata to be trained. However, the available annotated fetal brain datasets are\nlimited in number and heterogeneity, hampering domain adaptation strategies for\nrobust segmentation. In this context, we use FaBiAN, a Fetal Brain magnetic\nresonance Acquisition Numerical phantom, to simulate various realistic magnetic\nresonance images of the fetal brain along with its class labels. We demonstrate\nthat these multiple synthetic annotated data, generated at no cost and further\nreconstructed using the target super-resolution technique, can be successfully\nused for domain adaptation of a deep learning method that segments seven brain\ntissues. Overall, the accuracy of the segmentation is significantly enhanced,\nespecially in the cortical gray matter, the white matter, the cerebellum, the\ndeep gray matter and the brain stem.\n