Segmentation of bone metastases based on attention mechanism

Abstract

SPECT bone imaging is an important means to assist doctors in diagnosing diseases. The traditional processing method is that radiologists diagnose images. Manual diagnosis is not only cumbersome and time-consuming, but also different diagnosis results will be caused by the different diagnosis experience of doctors. In view of the above problems, this paper uses U-Net network as the basic model, and at the same time conducts model performance optimization research. Based on the U-Net network, the attention mechanism is integrated to segment the bone metastases in the pelvic area. Introducing the attention mechanism into the U-Net network can help improve the correlation of the pelvic region and reduce the interference caused by problems such as uneven brightness and low contrast to the model. Through multiple sets of experimental demonstrations, the U-Net network integrated with the attention mechanism can better segment bone metastases in the pelvic region based on SPECT images, and the model indicators have been significantly improved.