CT lymph node detection system based on spatial-temporal recurrent attention mechanism

摘要

The present disclosure discloses a CT lymph node detection system based on a spatial-temporal recurrent attention mechanism and specifically relates to the field of medical image analysis technologies. Based on a deep convolutional neural network and a recurrent attention mechanism, the present disclosure can construct an attention feature map adaptive to a lesion size in a slice direction and a spatial direction of a lymph node CT sequence. Firstly, a high-level spatial feature corresponding to the lymph node CT image is extracted by use of a pre-trained convolutional network; secondly, a recurrent attention mechanism based on a Gaussian Kernel Function is constructed with a slice at the center of the lymph node as a reference in a spatial domain; based on this, a temporal (slice direction) attention mechanism based on a Gaussian Mixture Model is performed; in addition, a predicted attention position is constrained based on the prior information of position distribution of the lymph node in the CT slice sequence; finally, in combination with the high-level features extracted by the two attention methods, the recurrent neural network performs classification to obtain a lymph node detection result.