pith. sign in

arxiv: 2605.16452 · v1 · pith:YUA4ROFLnew · submitted 2026-05-15 · 💻 cs.LG · cs.AI

Peak-Detector: Explainable Peak Detection via Instruction-Tuned Large Language Models in Physiological Sign

Jiahui Li , Yida Zhang , Zixuan Zeng , Jiayu Chen , Yingjian Song , Yin Xiao , Nishan Dong , Junjie Lu
show 5 more authors
Younghoon Kwon Xiang Zhang Jin Lu Wenzhan Song Fei Dou
This is my paper

Pith reviewed 2026-05-20 19:45 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords peak detectionlarge language modelsphysiological signalsexplainable AIECGPPGcross-modal detectioninstruction tuning
0
0 comments X

The pith

Instruction-tuned LLMs with a peak-representation technique detect peaks in ECG, PPG, BCG, and BSG signals at top accuracy while generating explanations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents Peak-Detector, a framework that applies instruction-tuned large language models to peak detection in multiple types of cardiac physiological signals. It introduces a peak-representation method that condenses the time-series input to focus the model on key events instead of raw noisy data. Training proceeds in two stages with supervised fine-tuning followed by reinforcement learning using a multi-objective reward, plus additional fine-tuning on a custom dataset of peak explanations. The approach is tested across seven datasets covering four signal modalities and achieves the best or tied-best results under clinical timing tolerances. The generated rationales allow users to inspect why detections succeed or fail.

Core claim

Peak-Detector is a framework that leverages instruction-tuned Large Language Models for robust, cross-modal, and explainable peak detection. A core innovation is a peak-representation technique that transforms time-series data into a condensed format, preserving critical event information while significantly reducing signal length. This representation provides a crucial inductive bias, guiding the LLM to reason over physiologically meaningful events rather than raw, noisy data. The model is optimized through a two-stage process: supervised fine-tuning followed by reinforcement learning with a multi-objective reward function. The model's self-explanation capabilities are cultivated by fine-tn

What carries the argument

The peak-representation technique, which condenses time-series signals into a shorter format that keeps essential event markers and supplies an inductive bias for the LLM to reason about physiologically meaningful events.

If this is right

  • Peak detection becomes possible across modalities without writing separate expert-tuned algorithms for each signal type.
  • Generated rationales make it possible for clinicians to verify detections and diagnose specific failure cases.
  • The same trained model can be applied to both public benchmarks and real-world cohorts with comparable performance.
  • Error analysis is supported directly by the model's own explanations rather than post-hoc methods.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same condensation approach could be tested on other physiological time-series tasks such as arrhythmia classification or sleep-stage detection.
  • Combining the LLM outputs with existing rule-based detectors might create hybrid systems that are both accurate and auditable.
  • If the representation length can be further reduced without accuracy loss, the method could run on lower-power devices for continuous monitoring.

Load-bearing premise

The peak-representation technique successfully preserves all information needed for accurate peak detection while guiding the LLM away from raw noise.

What would settle it

Run the trained model on a fresh high-noise dataset from an unseen signal modality and measure whether detection F1-score falls below the best conventional single-modality algorithm under the same temporal tolerance.

Figures

Figures reproduced from arXiv: 2605.16452 by Fei Dou, Jiahui Li, Jiayu Chen, Jin Lu, Junjie Lu, Nishan Dong, Wenzhan Song, Xiang Zhang, Yida Zhang, Yingjian Song, Yin Xiao, Younghoon Kwon, Zixuan Zeng.

Figure 1
Figure 1. Figure 1: Physiology signals with labelled peaks Large Language Models (LLMs) are increasingly applied to phys￾iological time-series analysis, though research currently focuses on high-level semantic tasks like sleep stage captioning in OpenTSLM [34] or arrhythmia diagnosis in ECG-LM [75]. Despite their power, these models often struggle with "numeric ground￾ing"—the precise localization of events like a specific pe… view at source ↗
Figure 2
Figure 2. Figure 2: Peak-Detector Framework undergoes supervised fine-tuning on the Peak-Explanation Dataset to internalize the specific signal syntax and explanatory format. This process yields the SFT Model, which functions as the reference policy for the subsequent optimization stage. • Block 3: Reinforcement Learning (Stage 2). Following the SFT phase, the framework transitions to the optimization block (Section 3.4.2). H… view at source ↗
Figure 3
Figure 3. Figure 3: Signal approximation by interpolating peaks. A fixed reference time 𝑇0 = 2020-01-01 00:00:00 is used, and absolute indices are computed across segments to prevent window￾local ambiguities. The index 𝑡𝑖 is transformed into elapsed seconds by dividing by the sampling frequency 𝐹𝑠 (i.e., calendar-second = 𝑡𝑖/𝐹𝑠 ), then formatted into the “HH:MM:SS” string. For example, at 𝐹𝑠 = 1 Hz, a peak at 𝑡𝑖 = 97 correspo… view at source ↗
Figure 4
Figure 4. Figure 4: A representative sample from the Peak-Explanation Dataset (BCG Arrhythmia subset). The figure illustrates the [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Teacher LLM-supervised data construction and structured response template. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Training strategies comparison 3.4.2 Stage 2: Reinforcement Learning Optimization. Following Supervised Fine-Tuning (SFT), the model under￾goes a second stage of optimization using Reinforcement Learning (RL) with Group Relative Policy Optimization (GRPO) [56]. As illustrated in the RL schematic (Fig. 6b), this process refines the model’s policy at the sequence level through a structured, block-by-block pr… view at source ↗
Figure 7
Figure 7. Figure 7: Qualitative visualization of peak detection performance across challenging segments of (a) ECG with T-wave interfer [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Radar Plot of Explanation Evalu￾ation The evaluation framework assesses the Peak-Detector’s interpretabil￾ity across five key dimensions, demonstrating high quality and clinical readiness. Faithfulness and Robustness quantify the alignment between qualitative explanations and quantitative signal features, specifically verify￾ing assertions against input data and testing stability under perturbations like c… view at source ↗
Figure 9
Figure 9. Figure 9: Demonstration of Peak-Detector’s emergent analytical capabilities on zero-shot tasks. (a) The model correctly classifies [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Validation of the Peak Representation. (a) Data Retention Ratio, showing the percentage of data points remaining [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Qualitative comparison of original signals (blue) and their reconstructions (orange) from the sparse Peak Representa [PITH_FULL_IMAGE:figures/full_fig_p018_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: F1-score comparison between Peak-Detector and traditional machine learning classifiers trained on features from the [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
Figure 14
Figure 14. Figure 14: The results indicate that performance consistently improves as the average number of detected peaks [PITH_FULL_IMAGE:figures/full_fig_p019_14.png] view at source ↗
Figure 13
Figure 13. Figure 13: Sensitivity analysis of Peak Representation metrics as a function of the minimum horizontal distance parameter in [PITH_FULL_IMAGE:figures/full_fig_p020_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Impact of the distance parameter on detection error. The plot demonstrates the trade-off between the average [PITH_FULL_IMAGE:figures/full_fig_p020_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: The impact of model scaling on Peak-Detector performance. As model size increases from 0.5B to 7B parameters, [PITH_FULL_IMAGE:figures/full_fig_p021_15.png] view at source ↗
Figure 16
Figure 16. Figure 16: Visualization of signal quality under different additive white Gaussian noise levels. Increasing noise progressively [PITH_FULL_IMAGE:figures/full_fig_p033_16.png] view at source ↗
Figure 17
Figure 17. Figure 17: Analysis of failure cases in the MIT-BIH dataset: (a) morphological confusion between S-peaks and R-peaks, and (b) [PITH_FULL_IMAGE:figures/full_fig_p034_17.png] view at source ↗
Figure 18
Figure 18. Figure 18: Impact of model scale on performance. We analyze the trade-off between the number of trainable parameters and [PITH_FULL_IMAGE:figures/full_fig_p037_18.png] view at source ↗
Figure 19
Figure 19. Figure 19: Interactive verification interface. (a) Visualization of valid Target Peaks with morphological details. (b) Visualization [PITH_FULL_IMAGE:figures/full_fig_p039_19.png] view at source ↗
Figure 20
Figure 20. Figure 20: MIT-BIH Arrhythmia Recall [PITH_FULL_IMAGE:figures/full_fig_p044_20.png] view at source ↗
read the original abstract

Accurate peak detection across diverse cardiac physiological signals, including the Electrocardiogram (ECG), Photoplethysmogram (PPG), Ballistocardiogram (BCG), and Bodyseismography (BSG), is fundamental for cardiovascular monitoring but is often hindered by artifacts and signal variability. Conventional algorithms are typically engineered with expert knowledge for a single signal modality, limiting their generalizability. Conversely, deep learning-based methods often lack interpretability, limiting transparency for expert verification and hindering expert-computer interaction. To address these limitations, we introduce Peak-Detector, a novel framework that leverages instruction-tuned Large Language Models (LLMs) for robust, cross-modal, and explainable peak detection. A core innovation of our framework is a "peak-representation" technique that transforms time-series data into a condensed format, preserving critical event information while significantly reducing signal length. This representation provides a crucial inductive bias, guiding the LLM to reason over physiologically meaningful events rather than raw, noisy data. The model is optimized through a two-stage process: supervised fine-tuning (SFT) followed by reinforcement learning (RL) with a multi-objective reward function. The model's self-explanation capabilities are cultivated by fine-tuning on a custom-built Peak-Explanation dataset. Across four modalities-ECG, PPG, BCG, and BSG-spanning seven datasets (six public benchmarks plus one real-world cohort), Peak-Detector demonstrates strong cross-modal performance, achieving best or tied-best detection under clinically relevant temporal tolerance. Beyond accuracy, the generated rationales surface failure modes and support verification and error analysis.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The manuscript introduces Peak-Detector, a framework leveraging instruction-tuned LLMs for robust, cross-modal, and explainable peak detection in cardiac physiological signals (ECG, PPG, BCG, BSG). A core component is the peak-representation technique that condenses time-series data while preserving critical event information to provide an inductive bias for the LLM. Training proceeds in two stages—supervised fine-tuning followed by reinforcement learning with a multi-objective reward function—and the model is further tuned on a custom Peak-Explanation dataset to generate self-explanations. Across seven datasets spanning the four modalities (six public benchmarks plus one real-world cohort), the paper claims best or tied-best detection performance under clinically relevant temporal tolerance, with generated rationales aiding failure-mode analysis and verification.

Significance. If the central performance claims hold and the LLM component is shown to contribute meaningfully beyond any preprocessing, the work could provide a valuable generalizable and interpretable alternative to modality-specific conventional algorithms or black-box deep-learning detectors. The cross-modal scope and emphasis on explainable rationales would be particularly useful for clinical cardiovascular monitoring applications where transparency supports expert verification.

major comments (1)
  1. [Abstract and Methods (peak-representation description)] Abstract and Methods (peak-representation description): The peak-representation is presented as transforming time-series into a condensed format that preserves critical event information and supplies an inductive bias guiding the LLM to reason over physiologically meaningful events. However, if this representation is constructed by first applying a conventional or heuristic peak finder to locate events before condensing (as raised in the stress-test note), then the reported accuracy would largely be inherited from the preprocessor rather than emerging from instruction tuning or RL. This is load-bearing for the claims of cross-modal generalizability and LLM-driven detection; explicit details on the exact construction steps, including any initial peak-location preprocessing, must be provided and ablated to substantiate the central claims.
minor comments (2)
  1. [Abstract] Abstract: Performance results are stated without accompanying quantitative tables, error bars, per-dataset metrics, or ablation studies, which hinders immediate assessment of the 'best or tied-best' outcomes even though the full manuscript presumably contains these details.
  2. [Methods] The multi-objective reward function used in the RL stage is referenced but not detailed in the abstract; ensure its formulation is fully specified in the methods to allow reproduction.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful review and for identifying a point that is indeed central to the validity of our claims. We address the concern about the peak-representation construction below and outline the revisions we will make to strengthen the manuscript.

read point-by-point responses

  1. Referee: Abstract and Methods (peak-representation description): The peak-representation is presented as transforming time-series into a condensed format that preserves critical event information and supplies an inductive bias guiding the LLM to reason over physiologically meaningful events. However, if this representation is constructed by first applying a conventional or heuristic peak finder to locate events before condensing (as raised in the stress-test note), then the reported accuracy would largely be inherited from the preprocessor rather than emerging from instruction tuning or RL. This is load-bearing for the claims of cross-modal generalizability and LLM-driven detection; explicit details on the exact construction steps, including any initial peak-location preprocessing, must be provided and ablated to substantiate the central claims.

    Authors: We thank the referee for highlighting this load-bearing issue. The peak-representation does not rely on any conventional or heuristic peak finder. As described in Section 3.2, the construction begins with z-score normalization of the raw time-series, followed by a fixed sliding-window encoding that computes local statistics (first and second derivatives, amplitude range, and zero-crossing rate) within each window and maps these to a compact sequence of discrete tokens. No peak localization step occurs; the representation is generated directly from the signal without identifying candidate events. This design supplies the inductive bias by emphasizing regions of rapid change rather than presupposing peak locations. To address the request for explicit details and ablation, we will expand Section 3.2 with pseudocode of the exact tokenization procedure and add a new ablation (Table X) that compares Peak-Detector against an otherwise identical LLM trained on raw (non-condensed) signals. These additions will be included in the revised manuscript. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical LLM framework is self-contained

full rationale

The paper describes an empirical pipeline: peak-representation to condense signals, followed by SFT then RL with multi-objective reward, plus fine-tuning on a custom Peak-Explanation dataset. Performance is reported via cross-dataset evaluation under temporal tolerance, not via any closed-form derivation or parameter fit that is renamed as a prediction. No equations, uniqueness theorems, or self-citations are invoked to force the central cross-modal accuracy claim. The inductive bias supplied by the representation is presented as an input design choice whose effectiveness is measured externally on held-out data, satisfying the criteria for non-circular empirical work.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the unverified assumption that the peak-representation step retains all physiologically relevant information and that the LLM can reliably reason from this compressed form. No free parameters or invented entities are explicitly quantified in the abstract.

axioms (1)
  • domain assumption The peak-representation technique preserves critical event information while reducing signal length.
    Invoked in the abstract as the core inductive bias that allows the LLM to focus on meaningful events rather than raw data.

pith-pipeline@v0.9.0 · 5859 in / 1300 out tokens · 29456 ms · 2026-05-20T19:45:55.839200+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

89 extracted references · 89 canonical work pages · 4 internal anchors

  1. [1]

    Alaleh Alivar, Charles Carlson, Ahmad Suliman, Steve Warren, Punit Prakash, David E Thompson, and Balasubramaniam Natarajan

    work page

  2. [2]

    Motion artifact detection and reduction in bed-based ballistocardiogram.Ieee Access7 (2019), 13693–13703

    work page 2019

  3. [3]

    Mikhail L Arbuzov, Alexey A Shvets, and Sisong Beir. 2025. Beyond Exponential Decay: Rethinking Error Accumulation in Large Language Models.arXiv preprint arXiv:2505.24187(2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  4. [4]

    S Azhaginiyan, M Anish, Menon K Shivaranjan, and M Ganesan. 2019. Denoising of BCG signal using multi resolution analysis. In2019 5th International conference on advanced computing & communication systems (ICACCS). IEEE, 1005–1008

    work page 2019

  5. [5]

    David Baehrens, Timon Schroeter, Stefan Harmeling, Motoaki Kawanabe, Katja Hansen, and Klaus-Robert Müller. 2010. How to explain individual classification decisions.The Journal of Machine Learning Research11 (2010), 1803–1831

    work page 2010

  6. [6]

    S Banerjee, R Gupta, and M Mitra. 2012. Delineation of ECG characteristic features using multiresolution wavelet analysis method. Measurement45, 3 (2012), 474–487

    work page 2012

  7. [7]

    Xiao Bi, Deli Chen, Guanting Chen, Shanhuang Chen, Damai Dai, Chengqi Deng, Honghui Ding, Kai Dong, Qiushi Du, Zhe Fu, et al

    work page

  8. [8]

    Deepseek llm: Scaling open-source language models with longtermism.arXiv preprint arXiv:2401.02954(2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  9. [9]

    Steven M Bishop and Ari Ercole. 2018. Multi-scale peak and trough detection optimised for periodic and quasi-periodic neuroscience data. InIntracranial Pressure & Neuromonitoring XVI. Springer, 189–195

    work page 2018

  10. [10]

    Dwaipayan Biswas, Luke Everson, Muqing Liu, Madhuri Panwar, Bram-Ernst Verhoef, Shrishail Patki, Chris H Kim, Amit Acharyya, Chris Van Hoof, Mario Konijnenburg, et al. 2019. CorNET: Deep learning framework for PPG-based heart rate estimation and biometric identification in ambulant environment.IEEE transactions on biomedical circuits and systems13, 2 (201...

    work page 2019

  11. [11]

    Marc Blackstein. 2025. RTX 5090: Designed for the Age of Neural Rendering. In2025 IEEE Hot Chips 37 Symposium (HCS). IEEE Computer Society, 1–20

    work page 2025

  12. [12]

    C Brüser, Stefan Winter, and Steffen Leonhardt. 2013. Robust inter-beat interval estimation in cardiac vibration signals.Physiological measurement34, 2 (2013), 123

    work page 2013

  13. [13]

    Charles Carlson, Vanessa-Rose Turpin, Ahmad Suliman, Carl Ade, Steve Warren, and David E Thompson. 2020. Bed-based ballistocar- diography: Dataset and ability to track cardiovascular parameters.Sensors21, 1 (2020), 156

    work page 2020

  14. [14]

    Denisse Castaneda, Aibhlin Esparza, Mohammad Ghamari, Cinna Soltanpur, and Homer Nazeran. 2018. A review on wearable photoplethysmography sensors and their potential future applications in health care.International journal of biosensors & bioelectronics 4, 4 (2018), 195

    work page 2018

  15. [15]

    Davide Castelvecchi. 2016. Can we open the black box of AI?Nature News538, 7623 (2016), 20

    work page 2016

  16. [16]

    Abhishek Chakraborty, Deboleena Sadhukhan, and Madhuchhanda Mitra. 2020. A robust PPG onset and systolic peak detection algorithm based on Hilbert transform. In2020 IEEE Calcutta Conference (CALCON). IEEE, 176–180

    work page 2020

  17. [17]

    Tianqi Chen and Carlos Guestrin. 2016. Xgboost: A scalable tree boosting system. InProceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining. 785–794

    work page 2016

  18. [18]

    Yuqi Chen, Kan Ren, Yansen Wang, Yuchen Fang, Weiwei Sun, and Dongsheng Li. 2023. Contiformer: Continuous-time transformer for irregular time series modeling.Advances in Neural Information Processing Systems36 (2023), 47143–47175

    work page 2023

  19. [19]

    Zhenqin Chen, Mengying Wang, Meiyu Zhang, Wei Huang, Hanjie Gu, and Jinshan Xu. 2023. Post-processing refined ECG delineation based on 1D-UNet.Biomedical Signal Processing and Control79 (2023), 104106

    work page 2023

  20. [20]

    Zhenqin Chen, Kaixiao Zheng, Junhua Shen, Yiwei Lin, Yan Feng, and Jinshan Xu. 2023. Sample point classification of abdominal ECG through CNN-Transformer model enables efficient fetal heart rate detection.IEEE Transactions on Instrumentation and Measurement73 (2023), 1–12

    work page 2023

  21. [21]

    Byung Hun Choi, Gih Sung Chung, Jin-Seong Lee, Do-Un Jeong, and Kwang Suk Park. 2009. Slow-wave sleep estimation on a load-cell-installed bed: a non-constrained method.Physiological measurement30, 11 (2009), 1163

    work page 2009

  22. [22]

    Haotian Dong, Jingyan Jiang, Rongwei Lu, Jiajun Luo, Jiajun Song, Bowen Li, Ying Shen, and Zhi Wang. 2025. Beyond A Single AI Cluster: A Survey of Decentralized LLM Training.arXiv preprint arXiv:2503.11023(2025)

    work page arXiv 2025

  23. [23]

    Mohamed Elgendi. 2012. On the analysis of fingertip photoplethysmogram signals.Current cardiology reviews8, 1 (2012), 14–25

    work page 2012

  24. [24]

    Mohamed Elgendi, Ian Norton, Matt Brearley, Derek Abbott, and Dale Schuurmans. 2013. Systolic peak detection in acceleration photoplethysmograms measured from emergency responders in tropical conditions.PloS one8, 10 (2013), e76585

    work page 2013

  25. [25]

    MAZ Fariha, Ryojun Ikeura, Soichiro Hayakawa, and Shigeyoshi Tsutsumi. 2020. Analysis of Pan-Tompkins algorithm performance with noisy ECG signals. InJournal of Physics: Conference Series, Vol. 1532. IOP Publishing, 012022. , Vol. 1, No. 1, Article . Publication date: May 2026. 26•Li et al

    work page 2020

  26. [26]

    Elizabeth Fons, Rachneet Kaur, Soham Palande, Zhen Zeng, Tucker Balch, Manuela Veloso, and Svitlana Vyetrenko. 2024. Evaluating large language models on time series feature understanding: A comprehensive taxonomy and benchmark.arXiv preprint arXiv:2404.16563 (2024)

    work page arXiv 2024

  27. [27]

    Parry Fung, Guy Dumont, Craig Ries, Chris Mott, and Mark Ansermino. 2004. Continuous noninvasive blood pressure measurement by pulse transit time. InThe 26th annual international conference of the IEEE engineering in medicine and biology society, Vol. 1. IEEE, 738–741

    work page 2004

  28. [28]

    Evangelos Georganas, Dhiraj Kalamkar, and Alexander Heinecke. 2025. Pushing the Envelope of LLM Inference on AI-PC.arXiv preprint arXiv:2508.06753(2025)

    work page arXiv 2025

  29. [29]

    Jeremy W Gordon. 1877. Certain molar movements of the human body produced by the circulation of the blood.Journal of anatomy and physiology11, Pt 3 (1877), 533

    work page

  30. [30]

    Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2025. A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions.ACM Transactions on Information Systems43, 2 (Jan. 2025), 1–55. doi:10.1145/3703155

    work page doi:10.1145/3703155 2025

  31. [31]

    Walter Karlen, Srinivas Raman, J Mark Ansermino, and Guy A Dumont. 2013. Multiparameter respiratory rate estimation from the photoplethysmogram.IEEE Transactions on biomedical engineering60, 7 (2013), 1946–1953

    work page 2013

  32. [32]

    Kianoosh Kazemi, Juho Laitala, Iman Azimi, Pasi Liljeberg, and Amir M Rahmani. 2022. Robust ppg peak detection using dilated convolutional neural networks.Sensors22, 16 (2022), 6054

    work page 2022

  33. [33]

    Daniel J Kelley, Terrence R Oakes, Larry L Greischar, Moo K Chung, John M Ollinger, Andrew L Alexander, Steven E Shelton, Ned H Kalin, and Richard J Davidson. 2008. Automatic physiological waveform processing for FMRI noise correction and analysis.PloS one3, 3 (2008), e1751

    work page 2008

  34. [34]

    Chang-Sei Kim, Stephanie L Ober, M Sean McMurtry, Barry A Finegan, Omer T Inan, Ramakrishna Mukkamala, and Jin-Oh Hahn. 2016. Ballistocardiogram: Mechanism and potential for unobtrusive cardiovascular health monitoring.Scientific reports6, 1 (2016), 31297

    work page 2016

  35. [35]

    Srinivas Kuntamalla and L Ram Gopal Reddy. 2014. An efficient and automatic systolic peak detection algorithm for photoplethysmo- graphic signals.International Journal of Computer Applications97, 19 (2014)

    work page 2014

  36. [36]

    Patrick Langer, Thomas Kaar, Max Rosenblattl, Maxwell A Xu, Winnie Chow, Martin Maritsch, Aradhana Verma, Brian Han, Daniel Seung Kim, Henry Chubb, et al. 2025. OpenTSLM: Time-Series Language Models for Reasoning over Multivariate Medical Text-and Time-Series Data.arXiv preprint arXiv:2510.02410(2025)

    work page arXiv 2025

  37. [37]

    Michael P LaValley. 2008. Logistic regression.Circulation117, 18 (2008), 2395–2399

    work page 2008

  38. [38]

    Thararin Lerddararadsamee and Yuttapong Jiraraksopakun. 2012. Local maximum detection for fully automatic classification of EM algorithm. In2012 9th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology. IEEE, 1–4

    work page 2012

  39. [39]

    Bing Nan Li, Ming Chui Dong, and Mang I Vai. 2010. On an automatic delineator for arterial blood pressure waveforms.Biomedical Signal Processing and Control5, 1 (2010), 76–81

    work page 2010

  40. [40]

    Ethan Li, Anders Boesen Lindbo Larsen, Chen Zhang, Xiyou Zhou, Jun Qin, Dian Ang Yap, Narendran Raghavan, Xuankai Chang, Margit Bowler, Eray Yildiz, et al. 2025. Apple intelligence foundation language models: Tech report 2025.arXiv preprint arXiv:2507.13575 (2025)

    work page arXiv 2025

  41. [41]

    Yong-Xian Li, Jiong-Ling Huang, Xin-Yu Yao, Si-Qi Mu, Shou-Xin Zong, and Yan-Fei Shen. 2024. A ballistocardiogram dataset with reference sensor signals in long-term natural sleep environments.Scientific Data11, 1 (2024), 1091

    work page 2024

  42. [42]

    Scott M Lundberg and Su-In Lee. 2017. A unified approach to interpreting model predictions.Advances in neural information processing systems30 (2017)

    work page 2017

  43. [43]

    Yashasvi Makin and Rahul Maliakkal. 2025. Sustainable AI Training via Hardware–Software Co-Design on NVIDIA, AMD, and Emerging GPU Architectures. In2025 IEEE International Conference on Service-Oriented System Engineering (SOSE). IEEE, 204–215

    work page 2025

  44. [44]

    Dominique Makowski, Tam Pham, Zen J Lau, Jan C Brammer, François Lespinasse, Hung Pham, Christopher Schölzel, and SH Annabel Chen. 2021. NeuroKit2: A Python toolbox for neurophysiological signal processing.Behavior research methods53, 4 (2021), 1689–1696

    work page 2021

  45. [45]

    George B Moody and Roger G Mark. 1992. MIT-BIH arrhythmia database.(No Title)(1992)

    work page 1992

  46. [46]

    Mohsen Nabian, Yu Yin, Jolie Wormwood, Karen S Quigley, Lisa F Barrett, and Sarah Ostadabbas. 2018. An open-source feature extraction tool for the analysis of peripheral physiological data.IEEE journal of translational engineering in health and medicine6 (2018), 1–11

    work page 2018

  47. [47]

    Tuyen Nguyen et al. 2025. An Evaluation of LLMs Inference on Popular Single-board Computers.arXiv preprint arXiv:2511.07425(2025)

    work page arXiv 2025

  48. [48]

    Thomas Penzel, Jan W Kantelhardt, Ronny P Bartsch, Maik Riedl, Jan F Kraemer, Niels Wessel, Carmen Garcia, Martin Glos, Ingo Fietze, and Christoph Schöbel. 2016. Modulations of heart rate, ECG, and cardio-respiratory coupling observed in polysomnography.Frontiers in physiology7 (2016), 460

    work page 2016

  49. [49]

    Marco AF Pimentel, Alistair EW Johnson, Peter H Charlton, Drew Birrenkott, Peter J Watkinson, Lionel Tarassenko, and David A Clifton

    work page

  50. [50]

    Toward a robust estimation of respiratory rate from pulse oximeters.IEEE Transactions on Biomedical Engineering64, 8 (2016), 1914–1923. , Vol. 1, No. 1, Article . Publication date: May 2026. Peak-Detector: Explainable Peak Detection via Instruction–Tuned Large Language Models in Physiological Signal•27

    work page 2016

  51. [51]

    Esteban J Pino, Javier AP Chávez, and Pablo Aqueveque. 2017. BCG algorithm for unobtrusive heart rate monitoring. In2017 IEEE healthcare innovations and point of care technologies (HI-poct). IEEE, 180–183

    work page 2017

  52. [52]

    Zaid Farooq Pitafi, Yingjian Song, Zaipeng Xie, Benjamin Brainard, and Wenzhan Song. 2025. Contactless Vital Signs Monitoring for Animals.IEEE Internet of Things Journal(2025)

    work page 2025

  53. [53]

    Attila Reiss, Ina Indlekofer, Philip Schmidt, and Kristof Van Laerhoven. 2019. Deep PPG: Large-scale heart rate estimation with convolutional neural networks.Sensors19, 14 (2019), 3079

    work page 2019

  54. [54]

    Steven J Rigatti. 2017. Random forest.Journal of insurance medicine47, 1 (2017), 31–39

    work page 2017

  55. [55]

    Pritam Sarkar and Ali Etemad. 2021. Cardiogan: Attentive generative adversarial network with dual discriminators for synthesis of ecg from ppg. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 35. 488–496

    work page 2021

  56. [56]

    L Sathyapriya, L Murali, and T Manigandan. 2014. Analysis and detection R-peak detection using Modified Pan-Tompkins algorithm. In 2014 IEEE International Conference on Advanced Communications, Control and Computing Technologies. IEEE, 483–487

    work page 2014

  57. [57]

    Christoph Schranz, Christina Halmich, Sebastian Mayr, and Dominik PJ Heib. 2024. Surrogate modelling of heartbeat events for improved J-peak detection in BCG using deep learning.Frontiers in Network Physiology4 (2024), 1425871

    work page 2024

  58. [58]

    Sabar Setiawidayat and Aviv Yuniar Rahman. 2018. New method for obtaining Peak Value R and the duration of each cycle of Electrocardiogram. In2018 International Conference on Sustainable Information Engineering and Technology (SIET). IEEE, 77–81

    work page 2018

  59. [59]

    Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, et al

    work page

  60. [60]

    Deepseekmath: Pushing the limits of mathematical reasoning in open language models.arXiv preprint arXiv:2402.03300(2024)

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  61. [61]

    JH Shin, BH Choi, Yong Gyu Lim, DU Jeong, and KS Park. 2008. Automatic ballistocardiogram (BCG) beat detection using a template matching approach. In2008 30th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE, 1144–1146

    work page 2008

  62. [62]

    Yingjian Song, Bingnan Li, Dan Luo, Glenna S Brewster Glasgow, Bradley G Phillips, Yuan Ke, and Wenzhan Song. 2024. Real-Time Continuous Blood Pressure Estimation with Contact-Free Bedseismogram. InICC 2024-IEEE International Conference on Communications. IEEE, 214–219

    work page 2024

  63. [63]

    Yingjian Song, Bingnan Li, Dan Luo, Zaipeng Xie, Bradley G Phillips, Yuan Ke, and Wenzhan Song. 2023. Engagement-free and contactless bed occupancy and vital signs monitoring.IEEE internet of things journal11, 5 (2023), 7935–7947

    work page 2023

  64. [64]

    Yingjian Song, Haotian Xiang, Zixuan Zeng, Jiayu Chen, Yida Zhang, Zaid Farooq Pitafi, He Yang, Qin Lu, Xiang Zhang, Bradley G Phillips, et al. 2025. Multi-granularity Supervised Contrastive Learning with Online Adaptation for Contactless In-bed Posture Classification. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies9, 2...

    work page 2025

  65. [65]

    Jianwei Su, Xuezhou Zhu, Xiaodong Zhang, Jintian Tang, and Lei Liu. 2009. Ballistocardiogram measurement system using three load-cell sensors platform in chair. In2009 2nd International Conference on Biomedical Engineering and Informatics. IEEE, 1–4

    work page 2009

  66. [66]

    Shuyan Sun. 2011. Meta-analysis of Cohen’s kappa.Health Services and Outcomes Research Methodology11, 3 (2011), 145–163

    work page 2011

  67. [67]

    Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, et al. 2023. Gemini: a family of highly capable multimodal models.arXiv preprint arXiv:2312.11805(2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  68. [68]

    Andriy Temko. 2017. Accurate heart rate monitoring during physical exercises using PPG.IEEE Transactions on Biomedical Engineering 64, 9 (2017), 2016–2024

    work page 2017

  69. [69]

    Viktor Tihonenko, Alexander Khaustov, Sergey Ivanov, and Alexei Rivin. 2007. St.-Petersburg institute of cardiological technics 12-lead arrhythmia database.(No Title)(2007)

    work page 2007

  70. [70]

    Erico Tjoa and Cuntai Guan. 2020. A survey on explainable artificial intelligence (xai): Toward medical xai.IEEE transactions on neural networks and learning systems32, 11 (2020), 4793–4813

    work page 2020

  71. [71]

    Sricharan Vijayarangan, R Vignesh, Balamurali Murugesan, SP Preejith, Jayaraj Joseph, and Mohansankar Sivaprakasam. 2020. RPnet: A Deep Learning approach for robust R Peak detection in noisy ECG. In2020 42nd annual international conference of the IEEE engineering in medicine & biology society (EMBC). IEEE, 345–348

    work page 2020

  72. [72]

    Pauli Virtanen, Ralf Gommers, Travis E Oliphant, Matt Haberland, Tyler Reddy, David Cournapeau, Evgeni Burovski, Pearu Peterson, Warren Weckesser, Jonathan Bright, et al. 2020. SciPy 1.0: fundamental algorithms for scientific computing in Python.Nature methods 17, 3 (2020), 261–272

    work page 2020

  73. [73]

    Zhenhua Wang, Guang Xu, and Ming Ren. 2024. Llm-generated natural language meets scaling laws: New explorations and data augmentation methods.arXiv preprint arXiv:2407.00322(2024)

    work page arXiv 2024

  74. [74]

    Guy JJ Warmerdam, Rik Vullings, Lars Schmitt, Judith OEH Van Laar, and Jan WM Bergmans. 2018. Hierarchical probabilistic framework for fetal R-peak detection, using ECG waveform and heart rate information.IEEE Transactions on Signal Processing66, 16 (2018), 4388–4397

    work page 2018

  75. [75]

    Robert M West. 2021. Best practice in statistics: Use the Welch t-test when testing the difference between two groups.Annals of clinical biochemistry58, 4 (2021), 267–269

    work page 2021

  76. [76]

    Leigha J Winters, Dale A Till, Mary L Bing, and James F Holmes. 2022. Time required for electrocardiogram interpretation in the emergency department.Academic Emergency Medicine29, 5 (2022)

    work page 2022

  77. [77]

    Xueyu Wu, Zhonghua Wang, Bo Xu, and Xibo Ma. 2020. Optimized pan-tompkins based heartbeat detection algorithms. In2020 Chinese Control And Decision Conference (CCDC). IEEE, 892–897. , Vol. 1, No. 1, Article . Publication date: May 2026. 28•Li et al

    work page 2020

  78. [78]

    do no harm

    Hanhui Xu and Kyle Michael James Shuttleworth. 2024. Medical artificial intelligence and the black box problem: a view based on the ethical principle of “do no harm”.Intelligent Medicine4, 1 (2024), 52–57

    work page 2024

  79. [79]

    Kai Yang, Massimo Hong, Jiahuan Zhang, Yizhen Luo, Suyuan Zhao, Ou Zhang, Xiaomao Yu, Jiawen Zhou, Liuqing Yang, Ping Zhang, et al. 2025. ECG-LM: Understanding Electrocardiogram with a Large Language Model.Health Data Science5 (2025), 0221

    work page 2025

  80. [80]

    Donghwan Yun, Hyung-Chul Lee, Chul-Woo Jung, Soonil Kwon, So-Ryoung Lee, Kwangsoo Kim, Yon Su Kim, and Seung Seok Han

    work page

Showing first 80 references.