The document discusses the development of a new deep learning model called bata-unet for liver segmentation from medical images, specifically CT scans. This model enhances a previously proposed bata-convnet model by incorporating batch normalization after each convolution layer, achieving significant improvements in segmentation accuracy, with dice coefficients of 0.97% and 0.96% for two datasets. The paper highlights the advancements in deep learning techniques and their applications in medical image processing, emphasizing the importance of automatic liver segmentation for clinical analysis and treatment planning.