MGRegBench: A Novel Benchmark Dataset with Anatomical Landmarks for Mammography Image Registration

Researchers have released MGRegBench, the first large-scale public dataset for mammography image registration with over 5,000 image pairs and 100 manually annotated landmarks. This addresses a critical gap in medical AI research by enabling standardized, reproducible benchmarking of registration methods across classical, learning-based, and deep learning approaches.

The medical imaging field has long struggled with fragmented research efforts due to reliance on private datasets and inconsistent evaluation protocols. MGRegBench directly addresses this bottleneck by establishing the first transparent, patient-disjoint benchmark specifically designed for mammography registration—a clinically critical application for tracking disease progression in breast tissue. The dataset's 5,000+ image pairs with segmentation masks and 100 pairs featuring manual anatomical landmarks create a foundation for meaningful comparative analysis across diverse methodologies.

Historically, progress in medical image registration has been impeded by researchers using proprietary data and non-standardized evaluation frameworks, making it impossible to determine which approaches genuinely outperform others. The researchers benchmarked multiple registration methods—from classical approaches like ANTs to contemporary deep learning solutions including VoxelMorph, TransMorph, IDIR, and MammoRegNet—under identical conditions. External validation on the independent SDM-MCs dataset tests whether findings generalize beyond the primary benchmark, strengthening confidence in comparative results.

For the medical AI ecosystem, this release establishes new standards for reproducibility and transparency. Academic institutions, commercial medical imaging companies, and healthcare AI developers now access a reference standard for evaluating innovations in registration technology. The public release of code and data reduces barriers to entry for research teams and startups developing mammography analysis tools, potentially accelerating innovation cycles.

Looking forward, the dataset's impact depends on adoption across the research community. Similar benchmarking initiatives in other medical imaging modalities could follow this model, progressively standardizing AI validation practices in healthcare.

• →MGRegBench provides the first large-scale public benchmark for mammography registration with 5,000+ image pairs and manual landmarks, addressing a critical gap in reproducible medical imaging research.
• →The benchmark enables direct comparison of classical, learning-based, and deep learning registration methods under identical conditions, replacing fragmented private-data approaches.
• →External validation on the independent SDM-MCs dataset demonstrates generalization beyond the primary benchmark, increasing confidence in comparative findings.
• →Public release of code and datasets reduces barriers to entry for academic and commercial teams developing mammography analysis tools.
• →The initiative establishes new standards for transparency and reproducibility that could influence benchmarking practices across medical imaging subfields.