Segmentation of Multi-Organ Functional Tissue Units Using UNet-EfficientNet-B8
Aoran Shen, Ruxin Chen, Yueze Zhu, Ruohan Hu · 2023
Segmentation of functional tissue units aims to discriminate tissue units of epithelium, glandular cavity, and fiber from an image, which helps to enhance humans’ understanding from the cell level and can fundamentally and strongly support future research in the field of Biotechnology. However, there are no easily discernible boundaries between functional and non-functional units in tissue images, making it difficult for machines to do such tasks. Based on common ideas from convolutional neural networks, the structural advantages of UNet and EficientNet are combined to create a new segmentation model for organ functional tissue units. In our experiment, UNet and EfficientNet are fused into a new model that extracts features with the help of the pre-trained EficientNet parameters to improve the ability of feature extraction. Besides, the combination of multi-scale features in the network is realized via the skip connection, and the segmentation accuracy of the model is thus improved. Finally, our proposed model is compared with other models by using the metrics of the Dice Similarity Coefficient (DSC). The result shows our UNet-EfficientNet-B8 owns the highest DSC of 0.721 among UNet-ResNet-50, UNet-ResNet-101, and UNet-Se_ResNet-101.