Research map / Detection and segmentation / CNN features
Multi-Stage Pathological Image Classification Using Semantic Segmentation
Shusuke Takahama, Yusuke Kurose, Yusuke Mukuta, Hiroyuki Abe, Masashi Fukayama, Akihiko Yoshizawa, Masanobu Kitagawa, Tatsuya Harada
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 |
|---|---|---|---|
| add a segmentation model over patch features (Separate Learning) | Classifier Only · Stomach biopsy | accuracy: 95.92 -> 98.53 | Improved |
| add a segmentation model over patch features (Separate Learning) | Classifier Only · Stomach biopsy | PR-AUC: 96.08 -> 99.30 | Improved |
| add end-to-end fine-tuning of the feature extractor with segmentation loss (ablation) | Separate Learning · Stomach biopsy | PR-AUC: 99.30 -> 99.34 | Improved |
| add end-to-end fine-tuning of the feature extractor with segmentation loss (ablation) | Separate Learning · Stomach biopsy | accuracy: 98.53 -> 98.45 | No clear effect |
| train the feature extractor on patches instead of using ImageNet features (ablation) | Segmentation on ImageNet features · Stomach biopsy | accuracy: 89.35 -> 98.53 | Improved |
| use full-resolution patch features instead of downsized 1/8 whole image (ablation) | Segmentation Only · Stomach biopsy | PR-AUC: 87.21 -> 99.30 | Improved |
| add the segmentation model (End-to-End) | Classifier Only · Camelyon16 | accuracy: 96.23 -> 98.14 | Improved |
| add the segmentation model (End-to-End) | Classifier Only · Camelyon16 | PR-AUC: 96.39 -> 99.24 | Improved |
| add the segmentation model (End-to-End) | Classifier Only · Camelyon17 | kappa score: 71.9 -> 76.4 | Improved |
| use gradient retention and partial training instead of segmenting at 1/8 resolution | Segmentation Only · Stomach biopsy | GPU memory: 7,000 MiB -> 10,000 MiB at full resolution | Improved |
| use End-to-End instead of Separate Learning (ablation) | Separate Learning · Stomach biopsy | training time: 38 hours -> 73 hours | Hurt |
Closest papers by meaning
- Rotation-Agnostic Image Representation Learning for Digital Pathology (CVPR 2024 · similarity 0.88)
- PathVQ: Reforming Computational Pathology Foundation Model for Whole Slide Image Analysis via Vector Quantization (NeurIPS 2025 · similarity 0.88)
- Dynamic Graph Representation with Knowledge-aware Attention for Histopathology Whole Slide Image Analysis (CVPR 2024 · similarity 0.87)
- Unsupervised Foundation Model-Agnostic Slide-Level Representation Learning (CVPR 2025 · similarity 0.87)
- Sequential Attention-based Sampling for Histopathological Analysis (NeurIPS 2025 · similarity 0.87)
- FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification (CVPR 2025 · similarity 0.87)
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.
