Osteosarcoma CT Image Segmentation Based on OSCA-TransUnet Model

This paper mainly conducts research on the segmentation of osteosarcoma CT images by using deep learning methods. Due to the lack of a public dataset for osteosarcoma CT images, we collaborated with Peking University People’s Hospital to construct the osteosarcoma CT dataset CT-OSD. Meanwhile, considering the limited accuracy of existing medical image segmentation models for osteosarcoma CT segmentation, we propose the OSCA-TransUnet model based on the fusion of spatial and channel attention. On the osteosarcoma CT dataset CT-OSD, the DSC index of the OSCA-TransUnet model is 2.44% higher than that of the TransUnet model, and the HD95 index is 2.05 mm lower than that of the TransUnet model. This effectively improves the segmentation accuracy of osteosarcoma CT images. Additionally, we also conducted generalization experiments of the OSCA-TransUnet model on the public medical dataset Synapse. The results show that the DSC index reaches 80.42% and the HD95 index reaches 18.04 mm. The performance is better than that of existing models, demonstrating the good generalization performance of the OSCA-TransUnet model.