Neonatal Brain MRI Segmentation Using Fine-Tuned Convolutional Neural Networks

抄録

The early treatment improves the prognosis of child with developmental disabilities. A brain shape analysis based system is required to achieve the very early detection of the developmental disorder in neonatal period but the neonatal brain shape evaluation is difficult due to a large variety of brain shape deformation caused by brain development and developmental disorder.

A large neonatal brain dataset and their brain region mask are required to construct statistical model to evaluate neonatal brain shape differences. The brain segmentation is a time consuming task; therefore, this paper proposes the neonatal brain segmentation method using convolutional neural networks and fine-tuning using adult brain dataset. The experimental results showed that the neonatal brain was successfully extracted with high accuracy (dice coefficient was 0.942). In addition, this paper revealed that the fine-tuning using adult brain magnetic resonance images is effective to the neonatal brain segmentation using convolutional neural networks.