Deep Learning-Based Perfusion Parameter Generation from DCE-MRI (SIAT, CAS)
Fine-tuningCollaborated to collect and preprocess DCE-MRI data from hospital partners, then trained and iteratively improved pix2pix-based models to generate perfusion parameter maps from imaging inputs. Used the dataset pairing pipeline to enable medical image translation, replacing traditional voxel-wise least-square fitting. Conducted ongoing model and training strategy improvements to meet clinical auxiliary analysis requirements. • Built robust paired DCE-MRI datasets with corresponding perfusion targets from clinical inputs. • Trained pix2pix for direct Ktrans/perfusion map generation as an AI-based replacement for fitting. • Iteratively tuned model architecture and training strategies for improved generation quality. • Supported ongoing validation and integration into collaborative DCE-MRI analysis workflows.