Research map / Vision-language and multimodal / Human and embodied modeling
Interleaved Text/Image Deep Mining on a Large-Scale Radiology Database for Automated Image Interpretation
Hoo-Chang Shin, Le Lu, Lauren Kim, Ari Seff, Jianhua Yao, Ronald M. Summers
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 VGG-19 instead of AlexNet (ablation) | CNN image-to-topic classifier · document-level topics | top-1 accuracy: 0.607 -> 0.658 | Improved |
| use VGG-16 instead of AlexNet (ablation) | CNN image-to-topic classifier · document-level sub-topics | top-1 accuracy: 0.33 -> 0.54 | Improved |
| use VGG-16 instead of AlexNet (ablation) | CNN image-to-topic classifier · sentence-level topics | top-1 accuracy: 0.48 -> 0.50 | Improved |
| add three conv layers (VGG-19 vs VGG-16) (ablation) | VGG-16 · document-level topics | top-5 accuracy: 0.93 -> 0.95 | Improved |
| add three conv layers (VGG-19 vs VGG-16) (ablation) | VGG-16 · document-level sub-topics | top-1 accuracy: 0.54 -> 0.54 | No clear effect |
| use deeper VGG-19 instead of AlexNet (training cost) (ablation) | AlexNet · training time to 70,000 iterations | training time: 4 hours 35 mins -> 4 days 40 mins | Hurt |
| initialize from ImageNet pre-trained weights instead of random (ablation) | VGG CNN · image-to-topic classification | test accuracy | Improved |
| fine-tune from related higher-level topic CNN instead of ImageNet model (ablation) | CNN for sub-topics · document-level sub-topic | test accuracy | Improved |
| deeper VGG instead of AlexNet only when pre-trained on ImageNet (ablation) | VGG-Net · image-to-topic classification | test accuracy | Mixed |
| add OpenI biomedical articles to word2vec training corpus (ablation) | word2vec on radiology reports · word vector quality | cosine similarity of neighbors: ~1 billion -> ~1.2 billion words | Improved |
| restrict labels to specific disease terms (UMLS T047 with negation) instead of loose topic labels | image labeling · dataset coverage | fraction of images with labels: about 90% of images lost | Hurt |
| restrict LDA to reports with associated images (ablation) | document-level LDA · topic mining | number of meaningful topics: 20 topics | No clear effect |
Closest papers by meaning
- Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding (ICLR 2025 · similarity 0.88)
- When Radiology Report Generation Meets Knowledge Graph (AAAI 2020 · similarity 0.88)
- Bootstrapping Chest CT Image Understanding by Distilling Knowledge from X-ray Expert Models (CVPR 2024 · similarity 0.87)
- AHIVE: Anatomy-aware Hierarchical Vision Encoding for Interactive Radiology Report Retrieval (CVPR 2024 · similarity 0.87)
- Bootstrapping Large Language Models for Radiology Report Generation (AAAI 2024 · similarity 0.85)
- PairAug: What Can Augmented Image-Text Pairs Do for Radiology? (CVPR 2024 · similarity 0.85)
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.
