Multi-modal Fusion with Dense Connection for Acute Bilirubin Encephalopathy Classification.

Bilirubin is a metabolite of red blood cells. When the bilirubin content exceeds the normal value, it will cause acute bilirubin encephalopathy (ABE), and rapidly deteriorate to kernicterus if without intervene, which has a case fatality rate as high as 10%. Recently, deep learning approaches have been widely applied in structural magnetic resonance image analysis. This study aims to develop a method based on deep learning to effectively classify ABE. Since most current medical image research only relies on single modal, which leads to the model unable to learn complementary features from multi-modality MRIs. Motivated by this issue, a convolutional neural network incorporating multi-modal fusion with dense connection is designed to learn the multi-level features of MRI for ABE classification. Moreover, the attention module was introduced into the dense connection to address the issue of class imbalance in the medical image. By exploring several dense connection strategies and designing ablation experiments, the rationality of the proposed method was demonstrated.