Research map / Vision-language and multimodal / Cross-modal learning
TAMER: A Tri-Modal Contrastive Alignment and Multi-Scale Embedding Refinement Framework for Zero-Shot ECG Diagnosis
Xuewei Zhou, Yajie Meng, Pan Zeng, Xianfang Tang, Feifei Cui, Qiangguo Jin, Jialiang Yang, Junlin Xu
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 wave-level attentive interaction (WLAI) (ablation) | TAMER · zero-shot average | macro-AUC (%): 76.16 -> 81.19 | Improved |
| add report-guided wave-level refinement (RGWR) (ablation) | TAMER · zero-shot average | macro-AUC (%): 78.73 -> 81.19 | Improved |
| add rhythm-level contrastive alignment (RLCA) (ablation) | TAMER · zero-shot average | macro-AUC (%): 80.62 -> 81.19 | Improved |
| add WLAI, RGWR, RLCA respectively (ablation) | TAMER · domain shift average | macro-AUC (%): WLAI 78.94, RGWR 79.44, RLCA 81.49 -> 83.08 | Improved |
| add spectrogram as third modality (ablation) | TAMER · zero-shot average | macro-AUC (%): 77.53 -> 81.19 | Improved |
| add clinical report modality (ablation) | TAMER · domain shift average | macro-AUC (%): 63.33 -> 83.08 | Improved |
| use Med-CPT instead of PubMedBERT or Clinical ModernBERT as report encoder (ablation) | TAMER · zero-shot average | macro-AUC (%): 72.23 / 76.92 -> 81.19 | Improved |
| set temperature lambda to 0.04 rather than 0.03 or 0.05 (ablation) | TAMER · zero-shot average | macro-AUC (%): 80.42 / 79.91 -> 81.19 | Improved |
| TAMER zero-shot versus ST-MEM fine-tuned, PTBXL-Super to CSN transfer | TAMER · PTBXL-Super to CSN domain shift | macro-AUC (%): 84.50 -> 80.95 | Hurt |
| TAMER versus MERL on PTBXL-Super to CPSC2018 transfer | TAMER · PTBXL-Super to CPSC2018 domain shift | macro-AUC (%): 88.21 -> 84.71 | Hurt |
| add spectrogram branch to the ECG encoder (compute cost) | TAMER · model complexity vs MERL | parameters / FLOPs: 114.27M -> 122.05M params; 1.77 -> 14.40 GFLOPs | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MERL — MERL: Multimodal Event Representation Learning in Heterogeneous Embedding Spaces (AAAI 2021) · better on 4 of 5 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Boosting Masked ECG-Text Auto-Encoders as Discriminative Learners (ICML 2025 · similarity 0.89)
- From Token to Rhythm: A Multi-Scale Approach for ECG-Language Pretraining (ICML 2025 · similarity 0.88)
- GEM: Empowering MLLM for Grounded ECG Understanding with Time Series and Images (NeurIPS 2025 · similarity 0.87)
- Zero-Shot ECG Classification with Multimodal Learning and Test-time Clinical Knowledge Enhancement (ICML 2024 · similarity 0.86)
- Parallel-Learning of Invariant and Tempo-variant Attributes of Single-Lead Cardiac Signals: PLITA (AAAI 2025 · similarity 0.86)
- Medformer: A Multi-Granularity Patching Transformer for Medical Time-Series Classification (NeurIPS 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.
