Fig. 4

Model overview. It consists of four modules. The main feature extraction module (MFE module) and the position auxiliary extraction module (PAE module) simultaneously extract the features from the input WBS image. The features extracted from MFE are the main feature sources of this model. The auxiliary features and entropy obtained from PAE are the sources of lesion texture and ___location information. The main feature and auxiliary feature will be merged in feature fusion module (FF module). The fused features are first analyzed by Swin Transformer Encoder (ST_Encoder) for correlation, and then the ___location information contained in entropy is fused by Transformer Decoder (T_Decoder). Finally, features of different sizes are entered into the detection head module (DH module) for detection. The largest features containing the most details will be fed into the added small target detection head in order to obtain more small lesions.