An actively updated collection of studies on large language models for medical time series analysis.
2
5 commits
updated Sep 6, 2026
README
Large Language Models in Medical Time Series Analysis
An actively updated companion repository for our review on Large Language Models for Medical Time Series Analysis (MedTSLLMs).
Last updated: September 2026
📌 Overview
This repository accompanies the review:
Large Language Models in Medical Time Series Analysis
Yu Han, Cigdem Beyan, Xiang Zhang, Xiaofeng Liu, Nan Liu, Jimeng Sun, Shenda Hong, Cheng Ding, and Vittorio Murino
Medical time series (MedTS), including electrocardiograms (ECG), electroencephalograms (EEG), photoplethysmography (PPG), vital-sign recordings, and wearable sensor data, are central to clinical diagnosis and health monitoring. This repository provides a curated and continuously updated collection of studies that connect MedTS with large language models (LLMs), including work on representation learning, signal-language alignment, reasoning, generation, retrieval, and agent-based analysis.
The current manuscript reviews 62 studies and organizes them into six clinically oriented application scenarios. Because the field is evolving rapidly, this repository serves as a living companion resource for tracking studies that appear after the literature collection used in the manuscript.
Scope. We use MedTSLLM as a broad term for LLM-based approaches to MedTS analysis rather than for a single model architecture. MedTSLLMs may combine language models with signal encoders, multimodal alignment modules, retrieval systems, or other task-specific components. The scope therefore includes different forms of LLM involvement in MedTS representation learning, signal-language alignment, reasoning, generation, and agent-based analysis.
🗂️ Table of Contents
📢 News
- 2026-09 — Repository released with the literature collection from the current review manuscript.
- 2026-XX — Paper link and final citation will be added after publication.
📊 Review Snapshot
The 62 studies summarized in the current manuscript are organized according to their primary clinical application in Table 4:
| Application | # Studies |
|---|---|
| Medical disease diagnosis | 16 |
| Clinical report generation | 8 |
| Medical question answering | 13 |
| Neuro-signal translation | 12 |
| Health assessment support | 9 |
| Physiological signal synthesis | 4 |
The reviewed literature spans ECG, EEG, PPG, vital signs, wearable sensors, and multimodal physiological data. Modality counts are not mutually exclusive because some studies use more than one signal type.
📚 Paper Collection
The organization below follows the clinical application taxonomy used in Table 4 of the review. Some studies naturally span multiple applications; for consistency, each study is listed under its primary application in the manuscript.
Metadata note. Year and venue information follows the bibliography used for the current manuscript. As preprints are formally published, the repository can be updated to link to the final versions while preserving the original review corpus.
Legend: — indicates that a public code repository was not listed in the manuscript or verified at the time of this update.
🩺 Medical Disease Diagnosis
16 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| HeartBEiT A foundational vision transformer improves diagnostic performance for electrocardiograms | 2023 | ECG | Backbone | Cardiac disease diagnosis (LVEF, HCM, STEMI) | npj Digital Medicine | code |
| sEHR-ECG-Text Ecg representation learning with multi-modal ehr data | 2023 | ECG + EHR | Text encoder | Multimodal ECG representation learning and diagnosis | TMLR | — |
| ECGBERT Ecgbert: Understanding hidden language of ecgs with self-supervised representation learning | 2023 | ECG | Backbone | Arrhythmia diagnosis and sleep-apnea detection | arXiv | — |
| METS Frozen language model helps ecg zero-shot learning | 2024 | ECG + clinical text | Text encoder | Zero-shot ECG diagnosis | MIDL | — |
| MERL-CKEPE Zero-shot ecg classification with multimodal learning and test-time clinical knowledge enhancement | 2024 | ECG + clinical notes | Prompt generator | Zero-shot ECG classification / abnormality diagnosis | arXiv | code |
| Zero-shot RAG Zero-shot ECG diagnosis with large language models and retrieval-augmented generation | 2023 | ECG | Text decoder | Zero-shot arrhythmia and sleep-apnea diagnosis | ML4H | — |
| ECG-GPT Automated diagnostic reports from images of electrocardiograms at the point-of-care | 2024 | ECG image | Text decoder | Free-text ECG diagnostic interpretation | medRxiv | — |
| CardioGPT Cardiogpt: An ecg interpretation generation model | 2024 | ECG | Classifier | ECG interpretation and abnormality diagnosis | IEEE Access | — |
| CQA-ESI Ecg semantic integrator (esi): A foundation ecg model pretrained with llm-enhanced cardiological text | 2024 | ECG + clinical text | Text encoder + decoder | Knowledge-enhanced ECG diagnosis | arXiv | code |
| EEG-GPT EEG-GPT: exploring capabilities of large language models for EEG classification and interpretation | 2024 | EEG | Classifier | EEG classification and interpretation | arXiv | — |
| ETP Etp: Learning transferable ecg representations via ecg-text pre-training | 2024 | ECG + clinical reports | Text encoder | Transferable ECG representation / zero-shot diagnosis | ICASSP 2024 | — |
| GPT-PPG GPT-PPG: a GPT-based foundation model for photoplethysmography signals | 2025 | PPG | Backbone | PPG foundation modeling and downstream diagnosis | Physiological Measurement | — |
| GPT-4 ECG interpretation Beyond text: the impact of clinical context on GPT-4’s 12-lead electrocardiogram interpretation accuracy | 2025 | ECG + clinical context | Text decoder | Clinical-context-aware ECG interpretation | Canadian Journal of Cardiology | — |
| GEM Gem: Empowering mllm for grounded ecg understanding with time series and images | 2025 | ECG time series + image + text | Text decoder + image/time-series encoder | Grounded ECG understanding | arXiv | code |
| MedualTime MedualTime: A dual-adapter language model for medical time series-text multimodal learning | 2024 | ECG + EEG + clinical text | Backbone | Medical time-series/text multimodal classification | arXiv | code |
| ZETA Interpretable multimodal zero shot ECG diagnosis via structured clinical knowledge alignment | 2026 | ECG | Text encoder + observation generator | Interpretable zero-shot ECG diagnosis | npj Cardiovascular Health | code |
📝 Clinical Report Generation
8 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| JoLT JoLT: jointly learned representations of language and time-series for clinical time-series interpretation (student abstract) | 2024 | ECG + text | Text decoder | Clinical time-series interpretation / report generation | AAAI 2024 | — |
| SignalGPT BioSignal Copilot: Leveraging the power of LLMs in drafting reports for biomedical signals | 2023 | Biomedical signals + text | Agent | Biomedical signal report drafting | medRxiv | — |
| MEIT MEIT: Multimodal Electrocardiogram Instruction Tuning on Large Language Models for Report Generation | 2025 | ECG | Prompt generator + text decoder | ECG report generation | Findings of ACL 2025 | code |
| PhysioLLM Physiollm: Supporting personalized health insights with wearables and large language models | 2024 | Wearable / Fitbit | Agent | Personalized wearable-data interpretation | IEEE BHI 2024 | — |
| ECG-Chat Ecg-chat: A large ecg-language model for cardiac disease diagnosis | 2025 | ECG | Backbone + report generator | ECG report generation | IEEE ICME 2025 | code |
| ECG-ReGen Electrocardiogram Report Generation and Question Answering via Retrieval-Augmented Self-Supervised Modeling | 2025 | ECG + retrieved reports | Text encoder + decoder | Retrieval-augmented ECG report generation | ICASSP 2025 | — |
| ECG-Bench Retrieval-Augmented Generation for Electrocardiogram-Language Models | 2025 | ECG + retrieved reports | Backbone | Retrieval-augmented ECG-language generation | arXiv | code |
| DiagECG DiagECG: An LLM-Driven Framework for Diagnostic Reasoning via Discretized ECG Tokenization | 2025 | ECG | Backbone + report generator | Diagnostic reasoning and ECG report generation | arXiv | — |
💬 Medical Question Answering
13 studies
🧠 Neuro-Signal Translation
12 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| EEG-ETB Integrating llm, eeg, and eye-tracking biomarker analysis for word-level neural state classification in semantic inference reading comprehension | 2023 | EEG + eye tracking | Classifier | Word-level neural-state classification / reading comprehension | arXiv | — |
| DeWave Dewave: Discrete eeg waves encoding for brain dynamics to text translation | 2023 | EEG | Text decoder | EEG-to-text translation | arXiv | code |
| MTAM Can brain signals reveal inner alignment with human languages? | 2023 | EEG + language context | Text encoder | EEG-language alignment / semantic classification | Findings of EMNLP 2023 | code |
| CFEHC Contextual feature extraction hierarchies converge in large language models and the brain | 2024 | iEEG + language context | Backbone | Brain–LLM representational alignment | Nature Machine Intelligence | — |
| iEEG-GPT Enhancing neural decoding with large language models: A GPT-based approach | 2024 | iEEG + spectral/topographic features | Text decoder | Neural decoding and interpretation | IEEE BCI 2024 | — |
| WERE From word embedding to reading embedding using large language model, eeg and eye-tracking | 2024 | EEG + eye tracking | Text encoder + classifier | Reading embedding / neural-state classification | IEEE EMBC 2024 | code |
| Neuro-GPT Neuro-gpt: Towards a foundation model for eeg | 2024 | EEG | Text decoder | EEG foundation modeling and interpretation | IEEE ISBI 2024 | code |
| BELT BELT: bootstrapped EEG-to-language training by natural language supervision | 2024 | EEG | Text decoder | EEG-to-language decoding | IEEE TNSRE | — |
| BELT-2 Belt-2: Bootstrapping eeg-to-language representation alignment for multi-task brain decoding | 2024 | EEG | Text decoder | Multi-task EEG-to-language decoding | arXiv | — |
| CET-MAE Enhancing eeg-to-text decoding through transferable representations from pre-trained contrastive eeg-text masked autoencoder | 2024 | EEG | Text decoder | EEG-to-text decoding | arXiv | — |
| BSLA LLMs Help Alleviate the Cross-Subject Variability in Brain Signal and Language Alignment | 2025 | EEG | Backbone + text decoder | Cross-subject brain-signal/language alignment | arXiv | — |
| Thought2Text Thought2Text: text generation from EEG signal using large language models (LLMs) | 2025 | EEG + image stimulus | Caption generator + evaluator | EEG-to-text generation | Findings of NAACL 2025 | code |
⌚ Health Assessment Support
9 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| ALPHA Alpha: Anomalous physiological health assessment using large language models | 2023 | PPG + HR + SpO₂ | Classifier | Physiological health assessment | arXiv | code |
| Health-Learner Large language models are few-shot health learners | 2023 | Wearable biosignals | Predictor | Few-shot health prediction | arXiv | — |
| AdaCT Large transformers are better eeg learners | 2023 | EEG + activity sensors | Backbone + classifier | Seizure, sleep-stage and activity classification | arXiv | code |
| Health-LLM Health-llm: Large language models for health prediction via wearable sensor data | 2024 | Wearable / Fitbit | Classifier | Health prediction from wearable sensor data | arXiv | code |
| CBPM-LLaMA Large language models for cuffless blood pressure measurement from wearable biosignals | 2024 | ECG + PPG | Predictor | Cuffless blood-pressure estimation | ACM BCB 2024 | — |
| WDAI-LLM Large language models for wearable data analysis and interpretation | 2024 | Wearable / Fitbit | Predictor | Wearable-data analysis and health prediction | Tiny Papers @ ICLR 2024 | — |
| PSRT The Prediction of Stress in Radiation Therapy: Integrating Artificial Intelligence with Biological Signals | 2024 | ECG + PPG + EEG + wearable | Classifier | Stress prediction | Cancers | — |
| PH-LLM Towards a personal health large language model | 2024 | Wearable / Fitbit | Backbone | Personalized sleep and fitness assessment | arXiv | — |
| SensorLM SensorLM: Learning the Language of Wearable Sensors | 2025 | Wearable sensors | Text decoder | Sensor-to-text / health assessment | arXiv | — |
🫀 Physiological Signal Synthesis
4 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| Auto-TTE Text-to-ecg: 12-lead electrocardiogram synthesis conditioned on clinical text reports | 2023 | ECG + clinical text | Text decoder | Text-conditioned 12-lead ECG synthesis | ICASSP 2023 | code |
| ECG-LLM ECG-LLM: Leveraging Large Language Models for Low-Quality ECG Signal Restoration | 2024 | ECG | Backbone | Low-quality ECG restoration | IEEE BIBM 2024 | code |
| BCG2ECG Adapting LLMs for Ballistocardiographic Signals: A Multi-Task Learning Framework for BCG to ECG Reconstruction | 2024 | BCG + ECG + PPG | Backbone | BCG-to-ECG reconstruction | RICAI 2024 | — |
| DiffuSETS DiffuSETS: 12-Lead ECG generation conditioned on clinical text reports and patient-specific information | 2025 | ECG + clinical text + patient context | Backbone | Text-conditioned 12-lead ECG generation | Patterns | code |
🆕 Recent Additions
This section is reserved for relevant MedTSLLM studies that appeared after the literature collection used in the current manuscript. Newly added studies will be kept separate from the original 62-study review corpus so that the provenance of the published review remains clear.
| Study | Year | Modality | Application | Paper | Code | Added |
|---|---|---|---|---|---|---|
| New studies will be added here. | — | — | — | — | — | — |
🤝 Contributing
We welcome contributions from the community to keep this resource up to date. If you would like to add a relevant MedTSLLM study, please open an Issue or submit a Pull Request.
Please provide the following information when suggesting a paper:
- Paper title
- Authors
- Venue and year
- Paper URL
- Code/model URL, if available
- MedTS modality (e.g., ECG, EEG, PPG, wearable sensors, vital signs)
- Primary clinical application
- Primary role of the LLM in the pipeline
For consistency, please use one of the six primary application categories adopted in the review:
- Medical Disease Diagnosis
- Clinical Report Generation
- Medical Question Answering
- Neuro-Signal Translation
- Health Assessment Support
- Physiological Signal Synthesis
Suggested entry format:
| **Model/Study Name**<br>[Paper title](PAPER_URL) | 2026 | ECG | Answer generator | ECG question answering | Venue | [code](CODE_URL) |
Inclusion note
This repository focuses on studies in which an LLM or language-model-style architecture plays a substantive role in medical time-series modeling, interpretation, reasoning, generation, or multimodal alignment. General medical LLM studies without a MedTS component and conventional time-series models without meaningful LLM involvement are outside the primary scope.
📑 Citation
If you find this review or repository useful, please consider citing our work. The final bibliographic information will be updated after publication.
@misc{han2026medtsllm,
title = {Large Language Models in Medical Time Series Analysis},
author = {Han, Yu and Beyan, Cigdem and Zhang, Xiang and Liu, Xiaofeng and Liu, Nan and Sun, Jimeng and Hong, Shenda and Ding, Cheng and Murino, Vittorio},
year = {2026},
note = {Review manuscript}
}
⭐ Stay Updated
We will continue to update this repository as new MedTSLLM models, datasets, benchmarks, and clinical applications emerge. If you find the resource useful, please consider giving the repository a ⭐ and submitting an Issue or Pull Request when relevant work is missing.
Maintained by the authors of the review.
Contributors
5 commits
Similar repos
An actively updated collection of studies on large language models for medical time series analysis.
2
5 commits
updated Sep 6, 2026
README
Large Language Models in Medical Time Series Analysis
An actively updated companion repository for our review on Large Language Models for Medical Time Series Analysis (MedTSLLMs).
Last updated: September 2026
📌 Overview
This repository accompanies the review:
Large Language Models in Medical Time Series Analysis
Yu Han, Cigdem Beyan, Xiang Zhang, Xiaofeng Liu, Nan Liu, Jimeng Sun, Shenda Hong, Cheng Ding, and Vittorio Murino
Medical time series (MedTS), including electrocardiograms (ECG), electroencephalograms (EEG), photoplethysmography (PPG), vital-sign recordings, and wearable sensor data, are central to clinical diagnosis and health monitoring. This repository provides a curated and continuously updated collection of studies that connect MedTS with large language models (LLMs), including work on representation learning, signal-language alignment, reasoning, generation, retrieval, and agent-based analysis.
The current manuscript reviews 62 studies and organizes them into six clinically oriented application scenarios. Because the field is evolving rapidly, this repository serves as a living companion resource for tracking studies that appear after the literature collection used in the manuscript.
Scope. We use MedTSLLM as a broad term for LLM-based approaches to MedTS analysis rather than for a single model architecture. MedTSLLMs may combine language models with signal encoders, multimodal alignment modules, retrieval systems, or other task-specific components. The scope therefore includes different forms of LLM involvement in MedTS representation learning, signal-language alignment, reasoning, generation, and agent-based analysis.
🗂️ Table of Contents
📢 News
- 2026-09 — Repository released with the literature collection from the current review manuscript.
- 2026-XX — Paper link and final citation will be added after publication.
📊 Review Snapshot
The 62 studies summarized in the current manuscript are organized according to their primary clinical application in Table 4:
| Application | # Studies |
|---|---|
| Medical disease diagnosis | 16 |
| Clinical report generation | 8 |
| Medical question answering | 13 |
| Neuro-signal translation | 12 |
| Health assessment support | 9 |
| Physiological signal synthesis | 4 |
The reviewed literature spans ECG, EEG, PPG, vital signs, wearable sensors, and multimodal physiological data. Modality counts are not mutually exclusive because some studies use more than one signal type.
📚 Paper Collection
The organization below follows the clinical application taxonomy used in Table 4 of the review. Some studies naturally span multiple applications; for consistency, each study is listed under its primary application in the manuscript.
Metadata note. Year and venue information follows the bibliography used for the current manuscript. As preprints are formally published, the repository can be updated to link to the final versions while preserving the original review corpus.
Legend: — indicates that a public code repository was not listed in the manuscript or verified at the time of this update.
🩺 Medical Disease Diagnosis
16 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| HeartBEiT A foundational vision transformer improves diagnostic performance for electrocardiograms | 2023 | ECG | Backbone | Cardiac disease diagnosis (LVEF, HCM, STEMI) | npj Digital Medicine | code |
| sEHR-ECG-Text Ecg representation learning with multi-modal ehr data | 2023 | ECG + EHR | Text encoder | Multimodal ECG representation learning and diagnosis | TMLR | — |
| ECGBERT Ecgbert: Understanding hidden language of ecgs with self-supervised representation learning | 2023 | ECG | Backbone | Arrhythmia diagnosis and sleep-apnea detection | arXiv | — |
| METS Frozen language model helps ecg zero-shot learning | 2024 | ECG + clinical text | Text encoder | Zero-shot ECG diagnosis | MIDL | — |
| MERL-CKEPE Zero-shot ecg classification with multimodal learning and test-time clinical knowledge enhancement | 2024 | ECG + clinical notes | Prompt generator | Zero-shot ECG classification / abnormality diagnosis | arXiv | code |
| Zero-shot RAG Zero-shot ECG diagnosis with large language models and retrieval-augmented generation | 2023 | ECG | Text decoder | Zero-shot arrhythmia and sleep-apnea diagnosis | ML4H | — |
| ECG-GPT Automated diagnostic reports from images of electrocardiograms at the point-of-care | 2024 | ECG image | Text decoder | Free-text ECG diagnostic interpretation | medRxiv | — |
| CardioGPT Cardiogpt: An ecg interpretation generation model | 2024 | ECG | Classifier | ECG interpretation and abnormality diagnosis | IEEE Access | — |
| CQA-ESI Ecg semantic integrator (esi): A foundation ecg model pretrained with llm-enhanced cardiological text | 2024 | ECG + clinical text | Text encoder + decoder | Knowledge-enhanced ECG diagnosis | arXiv | code |
| EEG-GPT EEG-GPT: exploring capabilities of large language models for EEG classification and interpretation | 2024 | EEG | Classifier | EEG classification and interpretation | arXiv | — |
| ETP Etp: Learning transferable ecg representations via ecg-text pre-training | 2024 | ECG + clinical reports | Text encoder | Transferable ECG representation / zero-shot diagnosis | ICASSP 2024 | — |
| GPT-PPG GPT-PPG: a GPT-based foundation model for photoplethysmography signals | 2025 | PPG | Backbone | PPG foundation modeling and downstream diagnosis | Physiological Measurement | — |
| GPT-4 ECG interpretation Beyond text: the impact of clinical context on GPT-4’s 12-lead electrocardiogram interpretation accuracy | 2025 | ECG + clinical context | Text decoder | Clinical-context-aware ECG interpretation | Canadian Journal of Cardiology | — |
| GEM Gem: Empowering mllm for grounded ecg understanding with time series and images | 2025 | ECG time series + image + text | Text decoder + image/time-series encoder | Grounded ECG understanding | arXiv | code |
| MedualTime MedualTime: A dual-adapter language model for medical time series-text multimodal learning | 2024 | ECG + EEG + clinical text | Backbone | Medical time-series/text multimodal classification | arXiv | code |
| ZETA Interpretable multimodal zero shot ECG diagnosis via structured clinical knowledge alignment | 2026 | ECG | Text encoder + observation generator | Interpretable zero-shot ECG diagnosis | npj Cardiovascular Health | code |
📝 Clinical Report Generation
8 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| JoLT JoLT: jointly learned representations of language and time-series for clinical time-series interpretation (student abstract) | 2024 | ECG + text | Text decoder | Clinical time-series interpretation / report generation | AAAI 2024 | — |
| SignalGPT BioSignal Copilot: Leveraging the power of LLMs in drafting reports for biomedical signals | 2023 | Biomedical signals + text | Agent | Biomedical signal report drafting | medRxiv | — |
| MEIT MEIT: Multimodal Electrocardiogram Instruction Tuning on Large Language Models for Report Generation | 2025 | ECG | Prompt generator + text decoder | ECG report generation | Findings of ACL 2025 | code |
| PhysioLLM Physiollm: Supporting personalized health insights with wearables and large language models | 2024 | Wearable / Fitbit | Agent | Personalized wearable-data interpretation | IEEE BHI 2024 | — |
| ECG-Chat Ecg-chat: A large ecg-language model for cardiac disease diagnosis | 2025 | ECG | Backbone + report generator | ECG report generation | IEEE ICME 2025 | code |
| ECG-ReGen Electrocardiogram Report Generation and Question Answering via Retrieval-Augmented Self-Supervised Modeling | 2025 | ECG + retrieved reports | Text encoder + decoder | Retrieval-augmented ECG report generation | ICASSP 2025 | — |
| ECG-Bench Retrieval-Augmented Generation for Electrocardiogram-Language Models | 2025 | ECG + retrieved reports | Backbone | Retrieval-augmented ECG-language generation | arXiv | code |
| DiagECG DiagECG: An LLM-Driven Framework for Diagnostic Reasoning via Discretized ECG Tokenization | 2025 | ECG | Backbone + report generator | Diagnostic reasoning and ECG report generation | arXiv | — |
💬 Medical Question Answering
13 studies
🧠 Neuro-Signal Translation
12 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| EEG-ETB Integrating llm, eeg, and eye-tracking biomarker analysis for word-level neural state classification in semantic inference reading comprehension | 2023 | EEG + eye tracking | Classifier | Word-level neural-state classification / reading comprehension | arXiv | — |
| DeWave Dewave: Discrete eeg waves encoding for brain dynamics to text translation | 2023 | EEG | Text decoder | EEG-to-text translation | arXiv | code |
| MTAM Can brain signals reveal inner alignment with human languages? | 2023 | EEG + language context | Text encoder | EEG-language alignment / semantic classification | Findings of EMNLP 2023 | code |
| CFEHC Contextual feature extraction hierarchies converge in large language models and the brain | 2024 | iEEG + language context | Backbone | Brain–LLM representational alignment | Nature Machine Intelligence | — |
| iEEG-GPT Enhancing neural decoding with large language models: A GPT-based approach | 2024 | iEEG + spectral/topographic features | Text decoder | Neural decoding and interpretation | IEEE BCI 2024 | — |
| WERE From word embedding to reading embedding using large language model, eeg and eye-tracking | 2024 | EEG + eye tracking | Text encoder + classifier | Reading embedding / neural-state classification | IEEE EMBC 2024 | code |
| Neuro-GPT Neuro-gpt: Towards a foundation model for eeg | 2024 | EEG | Text decoder | EEG foundation modeling and interpretation | IEEE ISBI 2024 | code |
| BELT BELT: bootstrapped EEG-to-language training by natural language supervision | 2024 | EEG | Text decoder | EEG-to-language decoding | IEEE TNSRE | — |
| BELT-2 Belt-2: Bootstrapping eeg-to-language representation alignment for multi-task brain decoding | 2024 | EEG | Text decoder | Multi-task EEG-to-language decoding | arXiv | — |
| CET-MAE Enhancing eeg-to-text decoding through transferable representations from pre-trained contrastive eeg-text masked autoencoder | 2024 | EEG | Text decoder | EEG-to-text decoding | arXiv | — |
| BSLA LLMs Help Alleviate the Cross-Subject Variability in Brain Signal and Language Alignment | 2025 | EEG | Backbone + text decoder | Cross-subject brain-signal/language alignment | arXiv | — |
| Thought2Text Thought2Text: text generation from EEG signal using large language models (LLMs) | 2025 | EEG + image stimulus | Caption generator + evaluator | EEG-to-text generation | Findings of NAACL 2025 | code |
⌚ Health Assessment Support
9 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| ALPHA Alpha: Anomalous physiological health assessment using large language models | 2023 | PPG + HR + SpO₂ | Classifier | Physiological health assessment | arXiv | code |
| Health-Learner Large language models are few-shot health learners | 2023 | Wearable biosignals | Predictor | Few-shot health prediction | arXiv | — |
| AdaCT Large transformers are better eeg learners | 2023 | EEG + activity sensors | Backbone + classifier | Seizure, sleep-stage and activity classification | arXiv | code |
| Health-LLM Health-llm: Large language models for health prediction via wearable sensor data | 2024 | Wearable / Fitbit | Classifier | Health prediction from wearable sensor data | arXiv | code |
| CBPM-LLaMA Large language models for cuffless blood pressure measurement from wearable biosignals | 2024 | ECG + PPG | Predictor | Cuffless blood-pressure estimation | ACM BCB 2024 | — |
| WDAI-LLM Large language models for wearable data analysis and interpretation | 2024 | Wearable / Fitbit | Predictor | Wearable-data analysis and health prediction | Tiny Papers @ ICLR 2024 | — |
| PSRT The Prediction of Stress in Radiation Therapy: Integrating Artificial Intelligence with Biological Signals | 2024 | ECG + PPG + EEG + wearable | Classifier | Stress prediction | Cancers | — |
| PH-LLM Towards a personal health large language model | 2024 | Wearable / Fitbit | Backbone | Personalized sleep and fitness assessment | arXiv | — |
| SensorLM SensorLM: Learning the Language of Wearable Sensors | 2025 | Wearable sensors | Text decoder | Sensor-to-text / health assessment | arXiv | — |
🫀 Physiological Signal Synthesis
4 studies
| Study | Year | Modality | Primary LLM role | Main task | Venue | Code |
|---|---|---|---|---|---|---|
| Auto-TTE Text-to-ecg: 12-lead electrocardiogram synthesis conditioned on clinical text reports | 2023 | ECG + clinical text | Text decoder | Text-conditioned 12-lead ECG synthesis | ICASSP 2023 | code |
| ECG-LLM ECG-LLM: Leveraging Large Language Models for Low-Quality ECG Signal Restoration | 2024 | ECG | Backbone | Low-quality ECG restoration | IEEE BIBM 2024 | code |
| BCG2ECG Adapting LLMs for Ballistocardiographic Signals: A Multi-Task Learning Framework for BCG to ECG Reconstruction | 2024 | BCG + ECG + PPG | Backbone | BCG-to-ECG reconstruction | RICAI 2024 | — |
| DiffuSETS DiffuSETS: 12-Lead ECG generation conditioned on clinical text reports and patient-specific information | 2025 | ECG + clinical text + patient context | Backbone | Text-conditioned 12-lead ECG generation | Patterns | code |
🆕 Recent Additions
This section is reserved for relevant MedTSLLM studies that appeared after the literature collection used in the current manuscript. Newly added studies will be kept separate from the original 62-study review corpus so that the provenance of the published review remains clear.
| Study | Year | Modality | Application | Paper | Code | Added |
|---|---|---|---|---|---|---|
| New studies will be added here. | — | — | — | — | — | — |
🤝 Contributing
We welcome contributions from the community to keep this resource up to date. If you would like to add a relevant MedTSLLM study, please open an Issue or submit a Pull Request.
Please provide the following information when suggesting a paper:
- Paper title
- Authors
- Venue and year
- Paper URL
- Code/model URL, if available
- MedTS modality (e.g., ECG, EEG, PPG, wearable sensors, vital signs)
- Primary clinical application
- Primary role of the LLM in the pipeline
For consistency, please use one of the six primary application categories adopted in the review:
- Medical Disease Diagnosis
- Clinical Report Generation
- Medical Question Answering
- Neuro-Signal Translation
- Health Assessment Support
- Physiological Signal Synthesis
Suggested entry format:
| **Model/Study Name**<br>[Paper title](PAPER_URL) | 2026 | ECG | Answer generator | ECG question answering | Venue | [code](CODE_URL) |
Inclusion note
This repository focuses on studies in which an LLM or language-model-style architecture plays a substantive role in medical time-series modeling, interpretation, reasoning, generation, or multimodal alignment. General medical LLM studies without a MedTS component and conventional time-series models without meaningful LLM involvement are outside the primary scope.
📑 Citation
If you find this review or repository useful, please consider citing our work. The final bibliographic information will be updated after publication.
@misc{han2026medtsllm,
title = {Large Language Models in Medical Time Series Analysis},
author = {Han, Yu and Beyan, Cigdem and Zhang, Xiang and Liu, Xiaofeng and Liu, Nan and Sun, Jimeng and Hong, Shenda and Ding, Cheng and Murino, Vittorio},
year = {2026},
note = {Review manuscript}
}
⭐ Stay Updated
We will continue to update this repository as new MedTSLLM models, datasets, benchmarks, and clinical applications emerge. If you find the resource useful, please consider giving the repository a ⭐ and submitting an Issue or Pull Request when relevant work is missing.
Maintained by the authors of the review.
Similar repos
Contributors
5 commits