The application relates to the technical field of medical
image processing, and discloses a three-dimensional
cerebral blood volume image generation method,
system, device and medium. The method comprises the following steps: acquiring MRI three-dimensional body data and real CBV three-dimensional body data of a plurality of objects; gradually adding
noise to each real CBV three-dimensional body data based on a
diffusion model, and gradually denoising joint input formed by each
noise-added sample and corresponding MRI three-dimensional body data; taking a denoising network of the
diffusion model as a generator, acquiring all generated CBV three-dimensional body data output by the generator, discriminating each generated CBV three-dimensional body data according to a
feature matching degree of the generated CBV three-dimensional body data and corresponding real CBV three-dimensional body data through a
discriminator, and training the
discriminator and the generator through a discrimination result of the
discriminator and the
feature matching degree; and inputting MRI three-dimensional body data of a target object into a model trained through the above steps, to generate a corresponding three-dimensional CBV image.