Simulated Brainweb PET/MR Data for Denoising and Deblurring

Georg Schramm from Stanford University provides a dataset of 20 simulated subjects based on normal anatomical models from the BrainWeb phantom. Each subject includes three simulated PET contrast images, an attenuation image, and a high-resolution T1 MR scan, all with a shape of (220,220,184) and 1mm isotropic voxel size in NIfTI format. The anatomical models and T1 scans used to create the ground truth PET images originate from the BrainWeb: Simulated Brain Database.

Use Cases

• Benchmarking PET image denoising algorithms based on the three simulated 'random' PET contrast images.
• Developing deblurring methods for PET/MR data using the provided high-resolution T1 MR scans as anatomical reference.
• Evaluating multimodal image registration techniques using the co-registered PET and MR data from the BrainWeb phantom.
• Training or validating supervised deep learning models for medical image enhancement with simulated ground truth data.

Strengths

• Dataset contains 20 distinct simulated subjects, providing multiple data points for algorithm testing.
• All images have a consistent, high-resolution shape of (220,220,184) with a 1mm x 1mm x 1mm voxel size.
• Data is derived from the established BrainWeb phantom, a known resource for medical imaging simulation.
• Includes multiple modalities per subject: three PET contrasts, an attenuation map, and a T1 MR scan.

Limitations

• Row count and total file size are unknown, which may limit suitability assessment for large-scale training.
• Column-level documentation is absent; field semantics for the simulated PET contrasts must be inferred from the description.
• Last update date is unknown; freshness unverified.

Provenance

Source

BrainWeb: Simulated Brain Database (McGill University)

Collection Method

Simulated data generated from anatomical models of the BrainWeb phantom.