Neonatal Hypoxic Ischemic Encephalopathy Severity Grading Using Multimodal Swin Transformer

Thibaud Brochet, Kangfu Han, Jiale Cheng, Fenqiang Zhao, Jérôme Lapuyade-Lahorgue, Su Ruan, Yi-Fang Tu, Sheng-Che Hung, Gang Li

Abstract: Neonatal Hypoxic Ischemic Encephalopathy (HIE) is a brain injury that occurs in case of a lack of oxygen in the blood flow, which can result in death, disabilities and long-term neurodevelopmental sequelae. Magnetic Resonance Imaging (MRI) is of great significance in clinical HIE diagnosis and grading, as early detection and properly moderate hypothermia treatment would increase the chances of the newborn having a better brain development. In this paper, to tackle this issue, we propose a novel Swin Transformer based network for automatic HIE severity grading using MRI data. Specifically, we integrate the spatial and channel attention modules into the network to extract informative and complementary features from the input multimodal MR images. Experiments on a dataset including both T1w and T2w MR images of 109 neonates suffering from three stages of HIE have shown the promising performance for the proposed network, when compared with state-of-the-art methods.