TLMB-GAN: A Network Model for Left Ventricular Image Segmentation

Aiming at the problem of small data sets and insufficient data sets in the field of medical image segmentation, a hybrid transfer learning generative adversarial network (TLMB-GAN) segmentation model is proposed. Using VGG16 network as the pre-training model, the pre-training model is trained on Image Net, and the parameters of the pre-training network are transferred to the target network for secondary training using the hybrid transfer learning method of transfer learning. By adjusting the weight of each convolutional layer, the optimal segmentation network is obtained. Compared with MCA-GAN segmentation model, the Dice coefficient of left ventricle MRI is increased by 4.34% and 5.47% in small data. The Jaccard coefficient increased by 2.76% and 4.85%; The sensitivity coefficient increased by 4.91% and 2.13%. All performance index are significantly better than other models.