Combining attention mechanisms and residual networks for automatic multi-disease diagnosis of bone scan images

Whole body bone scanning technology is a common diagnostic method for bone diseases such as bone metastases and plays an important role in the early diagnosis and treatment of human bone diseases. In this paper, a deep classification model combining attention mechanism and residual network is proposed to realize automatic classification of various bone diseases. The model uses ResNet as the base network, which effectively alleviates the problem of gradient disappearance caused by excessive depth of network. By introducing multi-spectral channel attention method into the residual module, the classification network is more focused on the extraction of focal region features. The SPECT bone scan image classification network was constructed. The experimental results on a set of real SPECT bone scan images demonstrate the effectiveness of the model in this paper, and its accuracy, precision, recall and F-1 score are 0.7919, 0.8060, 0.7895 and 0.7909, respectively, indicating that the model proposed in this paper has good classification effect on a variety of diseases in SPECT images.