LoDoPaB-CT: Benchmark for Low-Dose Computed Tomography Reconstruction

LoDoPaB-CT provides 42,895 sample pairs for training, validation, and testing low-dose CT reconstruction algorithms. The dataset was created by researchers at the University of Bremen and published in 2021. It is based on reconstructions from the publicly available LIDC/IDRI database.

Use Cases

• Benchmarking low-dose CT reconstruction algorithms based on the provided observation and ground truth image pairs.
• Training supervised deep learning models for denoising based on simulated low-dose observations.
• Developing and evaluating image quality metrics for CT reconstructions based on the high-quality ground truth references.

Strengths

• Contains 42,895 total sample pairs (35,820 train, 3,522 validation, 3,553 test) providing substantial data for model development.
• Includes paired low-dose observation and high-quality ground truth images, enabling supervised learning approaches.
• Provides randomized patient IDs and clear file structure, facilitating reproducible data access and patient-level analysis.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

LIDC/IDRI Database, processed by University of Bremen researchers.

Collection Method

Simulated low-dose CT projections from high-quality reconstructions.