Table A1.

The role of NLP in medical studies.

Study	Main Contribution	Methodology	Significance
M-FLAG: medical vision–language pre-training with frozen language models and latent space geometry optimization [205].	Introduces M-FLAG, a model that combines frozen language models with vision–language pre-training for medical applications.	Utilizes frozen language models to maintain robust linguistic features. Optimizes latent space geometry to enhance alignment between visual and textual information.	Improves medical image understanding and diagnostic capabilities. Facilitates more accurate and interpretable medical AI applications.
Frozen language model helps ECG zero-shot learning [206].	Demonstrates the effectiveness of frozen language models in performing zero-shot learning on ECG data.	Applies a pre-trained language model to interpret ECG signals without additional training on ECG- specific data. Leverages the generalization ability of language models to understand medical terminology and concepts.	Enables rapid deployment of ECG analysis tools without the need for extensive domain-specific data. Enhances the ability of AI to generalize across different medical datasets and tasks.
Med-UniC: unifying cross-lingual medical vision–language pre-training by diminishing bias [207].	Proposes Med-UniC, a model that addresses cross-lingual and cross-modal biases in medical vision–language pre-training.	Implements techniques to diminish biases present in multilingual and multimodal medical datasets. Uses a unified framework to align vision and language representations across different languages and medical contexts.	Promotes more equitable AI tools that perform consistently across diverse patient populations. Enhances the usability of medical AI in multilingual and multicultural healthcare settings.