This repository implements BCDU-Net, a deep autoencoder-decoder architecture for semantic segmentation of medical images. It combines bidirectional convolutional LSTM layers with U-Net structure to capture both semantic and high-resolution information. The model also uses densely connected convolutions and batch normalization for improved feature representation and convergence. It achieves state-of-the-art results on skin lesion, lung, and retinal blood vessel segmentation tasks.