Spatial Diffusion: A Scalable Self-Supervised Deep Learning Framework for Robust 3D Medical Imaging Registration

Pretrained model weights for "Spatial Diffusion: A Scalable Self-Supervised Deep Learning Framework for Robust 3D Medical Imaging Registration." To access the source code, please visit the GitHub repository for the project. Please refer to the citation section if you find our models useful. The training configuration and description of each pre-trained model is included in the table below: File name Training Dataset(s) Training Steps Description oct_ema_ckpt_115000.pth OCT 115,000 Trained unconditionally octa_ema_ckpt_115000.pth OCTA 115,000 Trained unconditionally oct-octa_ema_ckpt_165000.pth OCT and OCTA 165,000 Modality-conditional class key: 0 = Modality-invarient (use for cross-modality) 1 = OCT 2 = OCTA brain-atlas_ema_ckpt_120000.pth T1 Brain MRI (OASIS-1) 120,000 Trained unconditionally brain-template_ema_ckpt_120000.pth T1 Brain MRI (OASIS-1), MNI152 Brain Template 120,000 Trained unconditionally head-and-neck_venous_ema_ckpt_115000.pth T2 Head & Neck MRI 115,000 Trained unconditionally foundation_ema_ckpt_265000.pth OCT, OCTA, T1 Brain MRI (OASIS-1), MNI152 Brain Template, T2 Head & Neck MRI 265,000 Modality-conditional class key: 0 = Modality-invarient (use for cross-modality) 1 = OCT 2 = OCTA 3 = T1 Brain MRI 4 = T2 Head & Neck MRI