A Multi-model Organ Segmentation Method Based on Abdominal Ultrasound Image

A multi-model organ segmentation method was proposed in this paper. The modified U-Net deep neural network model and simplified DenseNet deep neural network model was used in our method to segment and classify abdominal organs. Preliminary semantic segmentation results were corrected combined with medical prior knowldege and the characteristics of video frame correlation, so that the fine segmentation of abdominal organs based on ultrasound images could be realized. TPR and SIR of our method reached 0.8172 and 0.7971, superior to similar algorithms in terms of both visual effect and evaluation indexes.