Research map / Detection and segmentation / Open-set and few-shot detection
Dataset Collections: Challenges of Large-Scale Data Aggregation in 3D Medical Image Datasets
Yannick Kirchhoff, Saikat Roy, Elisa Stegmeier, Hamideh Haghiri, Constantin Ulrich, Maximilian R. Rokuss, Tassilo Wald, Karol Gotkowski, Benjamin Hamm, Raphael Stock, Nico Disch, Michael Baumgartner and 21 more
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Examines the challenges of aggregating disparate public 3D medical imaging datasets into large-scale dataset collections, where combining datasets obscures data quality and provenance issues.
Problem
Large-scale dataset collections built by aggregating public 3D medical image datasets can obscure data quality and heterogeneity issues, while expert radiologist annotation limits curated dataset size.
Approach
Studies the pitfalls of aggregating disparate public 3D medical imaging datasets into collections, rather than proposing a new model; the abstract is truncated so the specific proposed analysis is not stated.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| apply intensity clipping to scans before hashing comparison (ablation) | perceptual hash duplicate detection · 883 CT scans | TPR at threshold 100: 0.962 TPR retained | No clear effect |
| apply 2-voxel cropping to scans (ablation) | perceptual hash duplicate detection · 883 CT scans | TPR at threshold 100: 0.962 -> 0.087 | Hurt |
| apply 2-voxel padding to scans (ablation) | perceptual hash duplicate detection · 883 CT scans | TPR at threshold 100: 0.962 -> 0.078 | Hurt |
| raise Hamming threshold above 100 (ablation) | perceptual hash duplicate detection · manual validation 5,000 pairs | NPV: 0.76 -> 0.86 at 200, 0.92 at 300 | Improved |
| add dataset-specific deduplication (e.g. clipping) and manual tracing to perceptual hashing (ablation) | duplicate detection · 565k-volume collection | duplicate pairs found: 56,051 -> 118,487 | Improved |
| deduplicate the aggregated collection | dataset collection · 565k-volume collection | effective dataset size: 564,885 -> 524,250 volumes (7.2% reduction) | Mixed |
Closest papers by meaning
- Copycats: the many lives of a publicly available medical imaging dataset (NeurIPS 2024 · similarity 0.87)
- Knowledge Transfer Scaling Laws for 3D Medical Imaging (NeurIPS 2026 · similarity 0.84)
- Better Tokens for Better 3D: Advancing Vision-Language Modeling in 3D Medical Imaging (NeurIPS 2025 · similarity 0.83)
- TumorChain: Interleaved Multimodal Chain-of-Thought Reasoning for Traceable Clinical Tumor Analysis (ICLR 2026 · similarity 0.83)
- 3D-RAD: A Comprehensive 3D Radiology Med-VQA Dataset with Multi-Temporal Analysis and Diverse Diagnostic Tasks (NeurIPS 2025 · similarity 0.83)
- Training Like a Medical Resident: Context-Prior Learning Toward Universal Medical Image Segmentation (CVPR 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
