A PREDICTION MODEL FOR TURNER SYNDROME BASED ON FEW-SHOT LEARNING AND MULTISCALE RESIDUAL NETWOEK

A prediction model is proposed for improving the diagnosis efficiency of Turner syndrome (TS) based on a multiscale residual network (MRN) and few-shot learning. TS facial images were pre-processed to obtain the main facial areas. A multiscale residual block (MRB) with multilevel attention mechanisms (MAM) was designed. The MRB was implemented by integrating the residual structure of multi-scale convolution kernels, and the MAM was used to learn feature channel relationships and the importance of different convolution kernels. The MRN was built using the MRB. The few-shot learning was utilized to train the MRN. The experimental results demonstrate that the prediction model can improve the diagnostic accuracy of TS.