Research map / Detection and segmentation / Open-set and few-shot detection
Tuning Medical Foundation Models for Inner Ear Temporal CT Analysis with Plug-and-play Domain Knowledge Aggregator
Weixun Wan, Xinyang Jiang, Zilong Wang, Bei Li, Cairong Zhao
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 DKA with domain knowledge tokens to the foundation model instead of full fine-tuning | Merlin · CIED outcome prediction | F1 (%): 77.17 -> 92.49 | Improved |
| add DKA with domain knowledge tokens instead of full fine-tuning | CT-FM · CIED six-class structural abnormality (micro) | F1 (%): 22.88 -> 48.33 | Improved |
| add DKA instead of full fine-tuning | Merlin · CIED six-class structural abnormality (micro) | F1 (%): 80.14 -> 80.79 | No clear effect |
| add DKA instead of full fine-tuning | SAM-Med3D · CIED binary structural abnormality | F1 (%): 50.97 -> 76.71 | Improved |
| use LoRA on SAM-Med3D instead of DKGT | SAM-Med3D · CIED outcome prediction | F1 (%): 58.16±30.37 vs DKGT 80.08±9.09 | Hurt |
| use adapter (655.23M parameters) on SAM-Med3D instead of DKGT | SAM-Med3D · CIED binary structural abnormality | F1 (%): 35.57 vs DKGT 76.71 | Hurt |
| fuse domain knowledge features with 3DResNet by late fusion | 3DResNet · CIED binary structural abnormality | F1 (%): 50.46 -> 52.49 | Improved |
| add DKA module to SAM-Med3D (element-wise addition design) (ablation) | SAM-Med3D · CIED outcome prediction | F1 (%): 70.25 -> 80.08 | Improved |
| let image tokens query domain knowledge tokens instead of reversing the query direction (ablation) | DKA on SAM-Med3D · CIED outcome prediction | F1 (%): 76.04 -> 80.08 | Improved |
| combine learnable context tokens and domain tokens by element-wise addition instead of concatenation (ablation) | DKA on SAM-Med3D · CIED outcome prediction | F1 (%): 78.28 -> 80.08 | Improved |
| add domain knowledge from all three structures (IAM, cochlea, CNC) instead of none (ablation) | DKGT on CT-FM · CIED outcome prediction | F1 (%): 78.24 -> 82.55 | Improved |
| add domain knowledge from all three structures instead of none (ablation) | DKGT on Merlin · CIED outcome prediction | F1 (%): 91.35 -> 92.49 | Improved |
Closest papers by meaning
- Large-scale and Fine-grained Vision-language Pre-training for Enhanced CT Image Understanding (ICLR 2025 · similarity 0.84)
- AOR: Anatomical Ontology-Guided Reasoning for Medical Large Multimodal Model in Chest X-Ray Interpretation (NeurIPS 2025 · similarity 0.82)
- Toward a Vision-Language Foundation Model for Medical Data: Multimodal Dataset and Benchmarks for Vietnamese PET/CT Report Generation (NeurIPS 2025 · similarity 0.82)
- FETAL-GAUGE: A BENCHMARK FOR ASSESSING VISION-LANGUAGE MODELS IN FETAL ULTRASOUND (ICLR 2026 · similarity 0.81)
- Bootstrapping Chest CT Image Understanding by Distilling Knowledge from X-ray Expert Models (CVPR 2024 · similarity 0.81)
- A Textbook Remedy for Domain Shifts: Knowledge Priors for Medical Image Analysis (NeurIPS 2024 · similarity 0.81)
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.
