Research map / Vision-language and multimodal / Multimodal LLMs
Do Multimodal LLMs Understand Intraoral Dental Data? Dataset, Platform, and Baselines
Luca Lumetti, Federico Rizzo, Francesca Cremonini, Ettore Candeloro, Lombardo Luca, Costantino Grana, Federico Bolelli
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 field-by-field (multi-turn) instead of one-shot prompting (ablation) | IOS-Qwen (LoRA) · Occlusal report generation (IOS-only) | Accuracy: 0.512 -> 0.632 | Improved |
| use field-by-field decoding instead of one-shot (ablation) | IOS-Qwen (LoRA) · Occlusal report generation (IOS-only) | ROUGE-L F1: 0.687 -> 0.763 | Improved |
| fine-tune the LLM with LoRA instead of freezing it (ablation) | IOS-Qwen · Occlusal report generation (IOS-only) | Accuracy: 0.411 -> 0.512 | Improved |
| pretrain PointTransformer with per-field classification heads (frozen LLM) (ablation) | IOS-Qwen · Occlusal report generation (IOS-only) | Accuracy: 0.372 -> 0.411 | Improved |
| pretrain the encoder with per-field heads (ablation) | IOS-Qwen · Occlusal report generation (IOS-only) | RadFact F1: 0.282 -> 0.308 | Improved |
| fine-tune LLM with LoRA instead of pretrained encoder with frozen LLM (ablation) | IOS-Qwen · Occlusal report generation (IOS-only) | RadFact F1: 0.308 -> 0.233 | Hurt |
| use per-field decoding costs about 9 times more inferences (ablation) | IOS-Qwen · Occlusal report generation (IOS-only) | inference cost: ~9x | Hurt |
| use IOS geometry model instead of best zero-shot IOP MLLM | IOS-Qwen · Occlusal report generation | Accuracy: 0.580 -> 0.632 | Improved |
| use medical-specific MedGemma-27B instead of MedGemma-4B | MedGemma · Occlusal report generation (IOP-only) | Accuracy: 0.514 -> 0.544 | Mixed |
| use the annotation platform with automatic pipelines | clinical workflow · Report time per patient | seconds: 300 -> 60 | Improved |
Closest papers by meaning
- OralGPT-Omni: A Versatile Dental Multimodal Large Language Model (CVPR 2026 · similarity 0.90)
- Towards Better Dental AI: A Multimodal Benchmark and Instruction Dataset for Panoramic X-ray Analysis (NeurIPS 2025 · similarity 0.89)
- Segmenting Maxillofacial Structures in CBCT Volumes (CVPR 2025 · similarity 0.86)
- OralGPT-Plus: Learning to Use Visual Tools via Reinforcement Learning for Panoramic X-ray Analysis (CVPR 2026 · similarity 0.85)
- Teeth-SEG: An Efficient Instance Segmentation Framework for Orthodontic Treatment based on Multi-Scale Aggregation and Anthropic Prior Knowledge (CVPR 2024 · similarity 0.85)
- OralXrays-9: Towards Hospital-Scale Panoramic X-ray Anomaly Detection via Personalized Multi-Object Query-Aware Mining (CVPR 2025 · similarity 0.84)
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.
