Research map / Label-efficient and robust learning / Recognition and long-tail
A case for reframing automated medical image classification as segmentation
Sarah Hooper, Mayee F Chen, Khaled Kamal Saab, Kush Bhatia, Curtis Langlotz, Christopher Re
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use rule-based segmentation-for-classification instead of classification network, 10% labels | segmentation-for-classification · CANDID | mean AUROC: 10.0% | Improved |
| use rule-based segmentation-for-classification instead of classification network, 10% labels | segmentation-for-classification · ISIC | mean AUROC: 6.9% | Improved |
| use rule-based segmentation-for-classification instead of classification network, 10% labels | segmentation-for-classification · SPINE | mean AUROC: 2.8% | Improved |
| use rule-based segmentation-for-classification instead of classification network, all labels | segmentation-for-classification · SPINE | mean AUROC: 6.1% | Improved |
| use rule-based segmentation-for-classification instead of classification network, all labels | segmentation-for-classification · ISIC | mean AUROC: 0.6% | Improved |
| segmentation instead of classification for small melanoma lesions | segmentation-for-classification · ISIC | recall: 0.18 -> 0.47 | Improved |
| segmentation instead of classification when chest tube spurious correlation absent | segmentation-for-classification · CANDID | AUROC: 0.58 -> 0.84 | Improved |
| segmentation instead of classification across lung regions (worst-case region) | segmentation-for-classification · CANDID | recall: 0.71 -> 0.87 | Improved |
| use threshold, fully connected or classifier summarizing functions on segmentation output vs on embeddings (ablation) | segmentation-for-classification · CANDID, ISIC, SPINE | mean AUROC | Improved |
| use multitask network (Y-Net) instead of segmentation-for-classification (ablation) | multitask network · CANDID, ISIC, SPINE | mean AUROC | Hurt |
| add semi-supervised segmentation with unlabeled data (ablation) | segmentation-for-classification · CANDID, ISIC, SPINE | mean AUROC | Improved |
| add boosted semi-supervision using classification labels (ablation) | segmentation-for-classification · CANDID, ISIC, SPINE | mean AUROC | Improved |
Closest papers by meaning
- Deep Interactive Segmentation of Medical Images: A Systematic Review and Taxonomy (TPAMI 2024 · similarity 0.90)
- Training Like a Medical Resident: Context-Prior Learning Toward Universal Medical Image Segmentation (CVPR 2024 · similarity 0.88)
- Generative Medical Segmentation (AAAI 2025 · similarity 0.87)
- Tyche: Stochastic In-Context Learning for Medical Image Segmentation (CVPR 2024 · similarity 0.87)
- Enhancing SAM with Efficient Prompting and Preference Optimization for Semi-supervised Medical Image Segmentation (CVPR 2025 · similarity 0.86)
- Interactive Medical Image Segmentation: A Benchmark Dataset and Baseline (CVPR 2025 · similarity 0.86)
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.
