abstract |
The invention discloses a multimodal migration method for automatically delineating a model, comprising: collecting multimodal data of a batch of patients, and selecting a small part of the samples for clinical target area or organ-at-risk analysis on one of the modalities M1. sketch. For the collected data, each modality is registered and aligned with the first modality M1, and then preprocessed into two batches of training sets. The first batch of training sets includes all modal images of all the collected samples, and the input is other Image data of modalities, gold marked as aligned M1 image data, used to train a multimodal reconstruction network. The second batch of training sets is a small number of selected samples, the input is the image data of all modalities, and the gold standard is the corresponding clinical target area or the binary image of the organ at risk, which is used for the transfer training of image segmentation for the reconstruction network. In this way, the present invention only needs to perform gold-standard delineation on the image data of one modality, so that the trained automatic delineation model can be adapted to the image data of multiple modalities. |