Three-dimensional abdomen CT image multi-organ registration method based on full convolutional network

The invention discloses a three-dimensional abdomen CT image multi-organ registration method based on a full convolutional network. First, an abdominal CT image data set is obtained. Then, an attention mechanism module is introduced, a network model based on full convolution is constructed, the network model is trained, a preliminary segmentation result is obtained, a full-connection conditional random field is introduced, the segmentation result is further optimized, and extraction of multiple organ regions of the abdominal CT sequence is achieved. Secondly, constructing abdominal CT sequencemulti-organ region image pairs and a similarity measurement function based on a local correlation coefficient data item and a spatial regularization item; and finally, minimizing a similarity measurement function by adopting an alternating direction multiplier method to realize registration of the abdominal multi-organ region CT images. According to the method, a two-step strategy of segmentationand registration is adopt