SuPreM: 9,262 CT Volumes with 25 Annotated Classes for 3D Medical Imaging

SuPreM contains 9,262 CT volumes and 25 annotated classes designed for supervised pre-training of 3D medical imaging models. Created by MrGiovanni and featured as an ICLR 2024 Oral presentation, it serves as a foundational resource for transfer learning in radiology.

Use Cases

• Transfer learning for 3D medical image segmentation using the 25 annotated classes
• Pre-training foundation models for radiology tasks using 9,262 CT volumes
• Benchmarking supervised representation learning algorithms on volumetric medical data

Strengths

• 9,262 3D CT volumes
• 25 annotated classes for supervised learning
• ICLR 2024 Oral peer-reviewed status

Limitations

• Specific anatomical labels for the 25 classes are not detailed in the provided metadata
• Restricted to CT modality which may limit generalization to MRI or Ultrasound

Provenance

Source

ICLR 2024 Oral publication by MrGiovanni

Collection Method

Supervised annotation of 3D CT volumes

Freshness

Last updated January 2026; based on research presented in 2024.