Research map / Vision-language and multimodal / CLIP and zero-shot
Temporal Inversion for Learning Interval Change in Chest X-Rays
Hanbin Ko, Kyeongmin Jeon, Doowoong Choi, Chang Min Park
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 Change-aware Sigmoid pretraining loss on inverted pairs (ablation) | BioViL-T SigLIP · MS-CXR-T supervised | Standard avg macro-accuracy: 61.1 -> 64.6 | Improved |
| add Change-aware Sigmoid pretraining loss (ablation) | BioViL-T SigLIP · MS-CXR-T supervised | Consistency avg: 39.5 -> 45.4 | Improved |
| add Bidirectional Cross-Entropy (BiCE) in fine-tuning (ablation) | BioViL-T + Change-aware SigLIP · MS-CXR-T supervised | Consistency avg: 45.4 -> 53.1 | Improved |
| add BiCE in fine-tuning (ablation) | BioViL-T + Change-aware SigLIP · MS-CXR-T supervised | Standard avg: 64.6 -> 65.2 | No clear effect |
| add Temporal Consistency Loss (TCL) on top of BiCE (ablation) | BioViL-T · MS-CXR-T supervised | Consistency avg: 53.1 -> 57.4 | Improved |
| add TCL on top of BiCE (ablation) | BioViL-T · MS-CXR-T supervised | Standard avg: 65.2 -> 64.1 | Hurt |
| replace CLIP loss with SigLIP loss when retraining (ablation) | BioViL-T · MS-CXR-T supervised | Standard avg: 60.1 -> 61.1 | No clear effect |
| use paired images instead of single image | BioViL-T · MIMIC retrieval | TEM: 3.5 -> 12.1 | Improved |
| add Change-aware Sigmoid loss | BioViL-T SigLIP · MIMIC retrieval | Recall@k / TEM: TEM 15.7 -> 17.1; I2T@5 35.1 -> 34.8 | Mixed |
| add TILA to BioViL-T SigLIP in zero-shot | BioViL-T SigLIP · MS-CXR-T zero-shot | Consistency avg: 24.0 -> 30.5 | Improved |
| add TILA to BioViL-T SigLIP in supervised training on private cohort | BioViL-T SigLIP · Private cohort | Standard avg: 55.3 -> 66.1 | Improved |
| use TILA representations for binary interval-change screening | BioViL-T · RexGradient | AUC: 0.565 -> 0.702 | Improved |
Closest papers by meaning
- Multimodal Disease Progression Modeling via Spatiotemporal Disentanglement and Multiscale Alignment (NeurIPS 2025 · similarity 0.87)
- BenchX: A Unified Benchmark Framework for Medical Vision-Language Pretraining on Chest X-Rays (NeurIPS 2024 · similarity 0.87)
- Enhanced Contrastive Learning with Multi-view Longitudinal Data for Chest X-ray Report Generation (CVPR 2025 · similarity 0.86)
- Unlocking the Power of Spatial and Temporal Information in Medical Multimodal Pre-training (ICML 2024 · similarity 0.86)
- Addressing Asynchronicity in Clinical Multimodal Fusion via Individualized Chest X-ray Generation (NeurIPS 2024 · similarity 0.85)
- HC-LLM: Historical-Constrained Large Language Models for Radiology Report Generation (AAAI 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.
