Lymphoma PET-CT fusion segmentation method based on small target position probability

The invention discloses a lymphoma PET-CT fusion segmentation method based on small target position probability, and the method comprises the steps: preheating an organ segmentation network of a CT image, and freezing the network after preheating; generating a random learnable parameter as a general lymphoma position prototype, combining the position prototype with the predicted organ mask of the frozen part, and generating a position prototype emphasizing small target lymphoma and a position prototype constraining an interference target; a new nn-UNet convolutional network is initialized to segment the whole body lymphoma, and the small target lymphoma position prototype and the interference target position prototype assist the 3D PET input image to segment the whole body lymphoma; and the nn-UNet convolutional neural network is used for carrying out segmentation by using a combined loss formed by Dice loss and Cross-Entropy Loss. According to the invention, the performance of automatic segmentation of syste