Multimodal Biomedical Dataset: 206 Aligned BF and SHG Tissue Microarray Images

The Multimodal Biomedical Dataset for Evaluating Registration Methods consists of 206 aligned Bright-Field and Second-Harmonic Generation images of tissue microarray cores, each 2048x204 pixels. It was produced by Kevin W. Eliceiri of the University of Wisconsin–Madison for the paper 'Non-disruptive collagen characterization in clinical histopathology using cross-modality image synthesis'. The dataset is split into training (40 pairs), validation (32 pairs across two sets), and test (134 pairs) subsets for method development and evaluation.

Use Cases

• Benchmarking image registration algorithms based on the 206 aligned Bright-Field and Second-Harmonic Generation image pairs.
• Training multimodal image synthesis models based on the paired Bright-Field and Second-Harmonic Generation data.
• Developing computer vision methods for histopathology analysis based on tissue microarray core images.
• Evaluating registration method robustness using the dedicated validation and test sets described.

Strengths

• Contains 206 high-resolution image pairs, each 2048x2048 pixels, providing a substantial base for analysis.
• Includes a structured split of 40 training, 32 validation, and 134 test image pairs for method development and evaluation.
• Images are precisely aligned, which is a critical feature for registration and multimodal learning tasks.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• The dataset's scope is limited to a specific modality pair (BF and SHG) from tissue microarrays, which may not generalize to other imaging contexts.

Provenance

Source

University of Wisconsin–Madison (Kevin W. Eliceiri), associated with a Nature Communications paper.

Collection Method

Originally produced for research on non-disruptive collagen characterization using cross-modality image synthesis.