Establishment and Research of Liver Medical Image Online Database Based on Deep Learning

Liver cancer is one of the most common types of cancer with a high mortality rate worldwide. Accurate segmentation of liver tumor CT images is essential for the diagnosis and treatment planning of liver cancer. Among various medical image segmentation tasks, U-Net variants have been widely applied to liver tumor segmentation. Considering the significant variation in tumor shape and size, this paper proposes a liver tumor segmentation model based on a U-Net network integrated with a CBAM and BN. The CBAM employs sequential channel and spatial attention modules, instead of skip connections, to reduce the semantic gap between the encoder and decoder, while the inclusion of BN helps prevent overfitting. Additionally, a database-linked website based on SpringBoot, MySQL, MyBatis, and SpringMVC was developed to collect experimental datasets for deep learning experiments and research. Testing on the public LiTS dataset showed that the model achieved an accuracy exceeding 80.0% for most CT image segmentations, with an average segmentation accuracy of 85.5%. The proposed method can assist doctors in marking tumor locations in medical images, thereby enhancing diagnostic efficiency and accuracy.