Research map / Label-efficient and robust learning / Recognition and long-tail
Diagnose Like a Radiologist: Hybrid Neuro-Probabilistic Reasoning for Attribute-Based Medical Image Diagnosis
Gangming Zhao, Quanlong Feng, Chaoqi Chen, Zhen Zhou, Yizhou Yu
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 cross-network attention and residual fusion (CNA-RES) (ablation) | Hybrid BN+GCN (EfficientNet-B4) · LIDC-IDRI | accuracy (%): 93.64 -> 95.31 | Improved |
| add first BN module (BN-1) and its coupling with GCN (ablation) | Hybrid (EfficientNet-B4) · LIDC-IDRI | accuracy (%): 92.41 -> 95.31 | Improved |
| add second BN module (BN-2) (ablation) | Hybrid (EfficientNet-B4) · LIDC-IDRI | accuracy (%): 94.74 -> 95.31 | Improved |
| add SE channel attention in GCN (ablation) | Hybrid (EfficientNet-B4) · LIDC-IDRI | accuracy (%): 94.01 -> 95.31 | Improved |
| add gradient backpropagation through BN modules (ablation) | Hybrid (EfficientNet-B4) · LIDC-IDRI | accuracy (%): 93.81 -> 95.31 | Improved |
| add alternating training strategy (ablation) | Hybrid (EfficientNet-B4) · LIDC-IDRI | accuracy (%): 94.31 -> 95.31 | Improved |
| replace relational network after GCN with two BN modules (ablation) | Relation network variant · LIDC-IDRI | accuracy (%): 92.15 -> 95.31 | Improved |
| add standard data augmentation | Our-Efficient-B4-FPN · LIDC-IDRI | accuracy (%): 95.31 -> 95.36 | Improved |
| add FPN to EfficientNet-B4 backbone | Efficient-B4 · LIDC-IDRI | accuracy (%): 89.21 -> 90.91 | Improved |
| add hybrid BN+GCN reasoning on top of ResNet-50-FPN | ResNet-50-FPN · LIDC-IDRI | accuracy (%): 90.01 -> 93.74 | Improved |
| add hybrid BN+GCN reasoning on top of EfficientNet-B4-FPN | Efficient-B4-FPN · TB-Xatt | accuracy (%): 94.19 -> 97.13 | Improved |
| add cross-network attention and residual fusion (CNA-RES) (ablation) | Hybrid (EfficientNet-B4) · TB-Xatt | accuracy (%): 96.21 -> 97.13 | Improved |
Closest papers by meaning
- Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding (ICLR 2025 · similarity 0.85)
- Visual and Domain Knowledge for Professional-level Graph-of-Thought Medical Reasoning (ICML 2025 · similarity 0.84)
- Bootstrapping Chest CT Image Understanding by Distilling Knowledge from X-ray Expert Models (CVPR 2024 · similarity 0.84)
- Novel Class Discovery in Chest X-rays via Paired Images and Text (AAAI 2024 · similarity 0.84)
- M3CoTBench: Benchmark Chain-of-Thought of MLLMs in Medical Image Understanding (ICLR 2026 · similarity 0.84)
- Learning Low-Rank Feature for Thorax Disease Classification (NeurIPS 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.
