pith. sign in

arxiv: 2605.16441 · v1 · pith:RMUP5W36new · submitted 2026-05-15 · 💻 cs.LG · cs.AI

DeepArrhythmia: Segment-Contextualized ECG Arrhythmia Classification via Selective Evidence Acquisition

Jiahui Li , Ruili Fang , Zishuai Liu , WenZhan Song , Jin Lu , Fei Dou This is my paper

Pith reviewed 2026-05-20 21:13 UTC · model grok-4.3

classification 💻 cs.LG cs.AI
keywords ECG arrhythmia classificationbeat-level detectionmultimodal frameworkselective evidence acquisitionR peak localizationrhythm contextsegment confidence
0
0 comments X

The pith

DeepArrhythmia classifies each ECG beat by combining raw signals with waveform images and selectively using richer evidence based on segment confidence.

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

The paper introduces DeepArrhythmia to address the limitation of treating ECG beats in isolation by incorporating multi-beat rhythm context. It processes a segment of the ECG by merging the raw signal data with a visual rendering of the waveform. R peaks are located to mark individual beats, enabling structured predictions that account for timing and morphological consistency across beats. The system then uses an overall confidence score for the segment to choose between using basic or more detailed physiological information for the final classification. This selective approach recognizes that extra evidence does not always lead to better results.

Core claim

DeepArrhythmia is a tool-grounded multimodal framework for segment-contextualized beat-level ECG arrhythmia classification. Given a multi-beat ECG segment, it combines the raw ECG signal and a rendered waveform image, localizes R peaks to identify beat instances, and produces structured beat-level predictions. The framework decouples physiological measurement from evidence integration using specialized tools for beat localization, numerical rhythm-morphology extraction, and morphology-focused textual analysis, and uses segment-level confidence to route between minimal and rich evidence states.

What carries the argument

Segment-level confidence routing mechanism that decides whether to operate in a minimal evidence state or acquire richer physiological details for classifying beats within the segment.

If this is right

  • Beat-level predictions gain accuracy from rhythm context without processing full details for every segment.
  • Classification performance remains stable or improves when evidence acquisition is gated by confidence rather than applied uniformly.
  • Decoupling of measurement tools from integration allows modular updates to specific analysis components.
  • Structured outputs at the beat level support downstream tasks like rhythm pattern identification.

Where Pith is reading between the lines

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

  • This selective evidence strategy could apply to other biosignal classifications where over-processing noisy segments wastes resources.
  • Future extensions might include adapting the routing thresholds dynamically based on patient history or device type.
  • Comparing performance across datasets with different arrhythmia complexities would test the robustness of the confidence estimator.

Load-bearing premise

Richer physiological evidence is not uniformly useful across all ECG segments, and segment-level confidence can reliably determine when to switch to more detailed analysis without reducing classification accuracy.

What would settle it

An experiment on a standard ECG dataset such as MIT-BIH showing that the confidence-routed version achieves equal or higher accuracy than always using rich evidence, or that disabling the router causes a measurable drop in beat-level F1 score on high-variance segments.

Figures

Figures reproduced from arXiv: 2605.16441 by Fei Dou, Jiahui Li, Jin Lu, Ruili Fang, Wenzhan Song, Zishuai Liu.

Figure 1
Figure 1. Figure 1: VEB/SVEB discrimination requires morphology in context. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the DeepArrhythmia framework. (1) Multimodal ECG inputs with three tool￾grounded evidence sources. (2) Two-stage mask-SFT yields Minimal, Rich, and Routed specialists. (3) At inference, confidence Cseg vs. threshold τ routes between minimal and rich evidence. Beat-anchor interface. The Peak Detector instantiates the abstract beat anchors Ax by localizing candidate R peaks from xts. These anchor… view at source ↗
Figure 3
Figure 3. Figure 3: Illustration of the shift-based augmentation strategy used to alleviate class imbalance. Rare [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Confusion-matrix panels for tool use across four datasets. [PITH_FULL_IMAGE:figures/full_fig_p017_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Comparison of MIT-BIH evaluation outcomes under DS1/DS2 and random split protocols. [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Relationship between prediction confidence and segment-level Micro-F1 on the MIT-BIH [PITH_FULL_IMAGE:figures/full_fig_p020_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Confusion-matrix panels for DeepArrhythmia across four datasets. [PITH_FULL_IMAGE:figures/full_fig_p026_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Sensitivity analyses for context length and peak-detector selection. [PITH_FULL_IMAGE:figures/full_fig_p027_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Effect of Qwen3.5 backbone scale on DeepArrhythmia beat classification on MIT-BIH Ar [PITH_FULL_IMAGE:figures/full_fig_p029_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Class-wise bigram statistics extracted from interpretation captions across beat classes N, [PITH_FULL_IMAGE:figures/full_fig_p030_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Interpretation example from MIT-BIH Record 100 (segment 27). Morphology analyzer [PITH_FULL_IMAGE:figures/full_fig_p031_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Second interpretation example from MIT-BIH Record 105 (segment 64). The model [PITH_FULL_IMAGE:figures/full_fig_p032_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Failure-case example from MIT-BIH Record 105 (segment 131). Misclassifications are [PITH_FULL_IMAGE:figures/full_fig_p032_13.png] view at source ↗
Figure 14
Figure 14. Figure 14: Illustrative interpretation produced by the morphology-analyzer student distilled from [PITH_FULL_IMAGE:figures/full_fig_p033_14.png] view at source ↗
read the original abstract

Beat-level Electrocardiography (ECG) arrhythmia detection aims to assign an arrhythmia class to each beat in a recording, yet many existing systems treat beats as isolated local instances. This is limiting because beat labels often depend on multi-beat rhythm context, including timing, compensatory pauses, and beat-to-beat morphological consistency. We present DeepArrhythmia, a tool-grounded multimodal framework for segment-contextualized beat-level ECG arrhythmia classification. Given a multi-beat ECG segment, DeepArrhythmia combines the raw ECG signal and a rendered waveform image, localizes R peaks to identify beat instances, and produces structured beat-level predictions. The framework decouples physiological measurement from evidence integration using specialized tools for beat localization, numerical rhythm--morphology extraction, and morphology-focused textual analysis. DeepArrhythmia uses segment-level confidence to route between minimal and rich evidence states, since richer physiological evidence is not uniformly useful. This agentic design integrates rhythm context, explicit physiological grounding, and selective evidence acquisition for decision making.

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

2 major / 0 minor

Summary. The paper introduces DeepArrhythmia, a multimodal framework for segment-contextualized beat-level ECG arrhythmia classification. It processes multi-beat ECG segments by combining raw signals with rendered waveform images, localizes R peaks to identify individual beats, decouples physiological measurement via specialized tools (beat localization, numerical rhythm-morphology extraction, textual morphology analysis), and routes between minimal and rich evidence states using segment-level confidence to incorporate rhythm context such as timing and compensatory pauses.

Significance. If the selective evidence routing proves reliable, the framework could advance beat-level ECG classification by avoiding uniform application of rich evidence and better handling context-dependent arrhythmias. The agentic design with explicit physiological grounding is a conceptual strength, though the absence of any reported validation limits assessment of practical impact.

major comments (2)
  1. Abstract: the central claim that segment-level confidence can route between minimal and rich evidence states without degrading beat-level predictions lacks any explicit formulation (e.g., entropy-based, learned gating, or threshold procedure), making the selective acquisition mechanism untestable from the provided description.
  2. Abstract: no ablation studies, baseline comparisons (fixed minimal vs. fixed rich vs. selective), or performance metrics are supplied on context-dependent cases such as compensatory pauses or beat-to-beat inconsistencies, so the premise that richer evidence is not uniformly useful remains unsupported.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments, which identify key opportunities to strengthen the clarity and empirical grounding of the selective evidence routing in DeepArrhythmia. We respond point-by-point to the major comments below.

read point-by-point responses

  1. Referee: Abstract: the central claim that segment-level confidence can route between minimal and rich evidence states without degrading beat-level predictions lacks any explicit formulation (e.g., entropy-based, learned gating, or threshold procedure), making the selective acquisition mechanism untestable from the provided description.

    Authors: We agree that the abstract provides only a high-level description of routing via segment-level confidence and does not specify the exact procedure. The full manuscript motivates the approach through the agentic, tool-grounded design but does not include a formal definition of the gating function. We will revise the abstract to state the mechanism explicitly (e.g., threshold on segment-level entropy or confidence score) and add a concise methods paragraph detailing the implementation so that the selective acquisition is fully specified and reproducible. revision: yes

  2. Referee: Abstract: no ablation studies, baseline comparisons (fixed minimal vs. fixed rich vs. selective), or performance metrics are supplied on context-dependent cases such as compensatory pauses or beat-to-beat inconsistencies, so the premise that richer evidence is not uniformly useful remains unsupported.

    Authors: We acknowledge that the current version does not report ablation studies or targeted metrics on context-dependent arrhythmias. The premise is supported conceptually by the framework's decoupling of physiological tools and selective routing, yet we recognize that direct empirical comparisons would provide stronger validation. We will add ablation experiments comparing fixed-minimal, fixed-rich, and selective routing, with performance breakdowns on cases involving compensatory pauses and beat-to-beat morphological inconsistencies. revision: yes

Circularity Check

0 steps flagged

No significant circularity in architectural framework

full rationale

The paper presents DeepArrhythmia as a multimodal architectural framework that combines raw ECG signals with rendered images, performs R-peak localization for beat instances, and routes between minimal and rich evidence states using segment-level confidence. No equations, derivations, fitted parameters, or first-principles predictions are described that could reduce to inputs by construction. The selective routing is motivated by the stated premise that richer physiological evidence is not uniformly useful, but this remains an explicit design assumption rather than a self-referential definition or statistically forced output. No self-citations, uniqueness theorems, or ansatzes from prior author work are invoked as load-bearing support for the core claims. The framework is therefore self-contained as a proposed system architecture without internal circular dependencies visible in the derivation chain.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Based on abstract only; the central assumption is that beat labels depend on multi-beat context and that selective evidence use improves decisions.

axioms (1)
  • domain assumption Beat labels often depend on multi-beat rhythm context, including timing, compensatory pauses, and beat-to-beat morphological consistency.
    Explicitly stated as the limitation of existing isolated-beat systems.
invented entities (1)
  • DeepArrhythmia framework no independent evidence
    purpose: Segment-contextualized beat-level ECG classification via selective evidence acquisition
    Newly introduced system whose performance is not demonstrated in the abstract.

pith-pipeline@v0.9.0 · 5724 in / 1310 out tokens · 56327 ms · 2026-05-20T21:13:28.574238+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.

  • IndisputableMonolith/Cost/FunctionalEquation.lean Jcost_pos_of_ne_one echoes
    ?
    echoes

    ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

    DeepArrhythmia uses segment-level confidence to route between minimal and rich evidence states, since richer physiological evidence is not uniformly useful.

  • IndisputableMonolith/Foundation/BranchSelection.lean branch_selection echoes
    ?
    echoes

    ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

    If Cseg(x) ≥ τd, the model returns the initial minimal-evidence prediction; if Cseg(x) < τd, the model invokes the optional evidence-producing tools.

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

20 extracted references · 20 canonical work pages · 7 internal anchors

  1. [1]

    Beyond Exponential Decay: Rethinking Error Accumulation in Large Language Models

    Mikhail L Arbuzov, Alexey A Shvets, and Sisong Beir. Beyond exponential decay: Rethinking error accumulation in large language models.arXiv preprint arXiv:2505.24187,

    work page internal anchor Pith review Pith/arXiv arXiv

  2. [2]

    Qwen Technical Report

    Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wen Ge, Yu Han, Fei Huang, et al. Qwen technical report.arXiv preprint arXiv:2309.16609,

    work page internal anchor Pith review Pith/arXiv arXiv

  3. [3]

    Ecg-agent: On-device tool-calling agent for ecg multi-turn dialogue.arXiv preprint arXiv:2601.20323,

    Hyunseung Chung, Jungwoo Oh, Daeun Kyung, Jiho Kim, Yeonsu Kwon, Min-Gyu Kim, and Edward Choi. Ecg-agent: On-device tool-calling agent for ecg multi-turn dialogue.arXiv preprint arXiv:2601.20323,

    work page arXiv

  4. [4]

    Evaluating large language models on time series feature understanding: A comprehensive taxonomy and benchmark

    Elizabeth Fons, Rachneet Kaur, Soham Palande, Zhen Zeng, Tucker Balch, Manuela Veloso, and Svitlana Vyetrenko. Evaluating large language models on time series feature understanding: A comprehensive taxonomy and benchmark. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 21598–21634,

    work page 2024

  5. [5]

    Distilling step-by-step! outperforming larger language models with less training data and smaller model sizes

    Cheng-Yu Hsieh, Chun-Liang Li, Chih-Kuan Yeh, Hootan Nakhost, Yasuhisa Fujii, Alex Ratner, Ranjay Krishna, Chen-Yu Lee, and Tomas Pfister. Distilling step-by-step! outperforming larger language models with less training data and smaller model sizes. InFindings of the Association for Computational Linguistics: ACL 2023, pages 8003–8017,

    work page 2023

  6. [6]

    Ecg-tcn: Wearable cardiac arrhythmia detection with a temporal convolutional network

    Thorir Mar Ingolfsson, Xiaying Wang, Michael Hersche, Alessio Burrello, Lukas Cavigelli, and Luca Benini. Ecg-tcn: Wearable cardiac arrhythmia detection with a temporal convolutional network. In 2021 IEEE 3rd International Conference on Artificial Intelligence Circuits and Systems (AICAS), pages 1–4. IEEE,

    work page 2021

  7. [7]

    Search-R1: Training LLMs to Reason and Leverage Search Engines with Reinforcement Learning

    Bowen Jin, Hansi Zeng, Zhenrui Yue, Jinsung Yoon, Sercan Arik, Dong Wang, Hamed Zamani, and Jiawei Han. Search-r1: Training llms to reason and leverage search engines with reinforcement learning.arXiv preprint arXiv:2503.09516,

    work page internal anchor Pith review Pith/arXiv arXiv

  8. [8]

    ECG-R1: Protocol-Guided and Modality-Agnostic MLLM for Reliable ECG Interpretation

    10 Jiarui Jin, Haoyu Wang, Xingliang Wu, Xiaocheng Fang, Xiang Lan, Zihan Wang, Deyun Zhang, Bo Liu, Yingying Zhang, Xian Wu, et al. Ecg-r1: Protocol-guided and modality-agnostic mllm for reliable ecg interpretation.arXiv preprint arXiv:2602.04279,

    work page internal anchor Pith review Pith/arXiv arXiv

  9. [9]

    Gem: Em- powering mllm for grounded ecg understanding with time series and images.arXiv preprint arXiv:2503.06073,

    Xiang Lan, Feng Wu, Kai He, Qinghao Zhao, Shenda Hong, and Mengling Feng. Gem: Em- powering mllm for grounded ecg understanding with time series and images.arXiv preprint arXiv:2503.06073,

    work page arXiv

  10. [10]

    Peak-r1: Instruction-tuned large language models for robust j-peak detection in cardiomechanical signals

    Jiahui Li, Yida Zhang, Zixuan Zeng, Jiayu Chen, Xiang Zhang, Jin Lu, WenZhan Song, and Fei Dou. Peak-r1: Instruction-tuned large language models for robust j-peak detection in cardiomechanical signals. InNeurIPS 2025 Workshop on Learning from Time Series for Health,

    work page 2025

  11. [11]

    Teach multimodal llms to comprehend electro- cardiographic images.arXiv preprint arXiv:2410.19008,

    Ruoqi Liu, Yuelin Bai, Xiang Yue, and Ping Zhang. Teach multimodal llms to comprehend electro- cardiographic images.arXiv preprint arXiv:2410.19008,

    work page arXiv

  12. [12]

    Benchecg and xecg: a benchmark and baseline for ecg foundation models

    Riccardo Lunelli, Angus Nicolson, Samuel Martin Pröll, Sebastian Johannes Reinstadler, Axel Bauer, and Clemens Dlaska. Benchecg and xecg: a benchmark and baseline for ecg foundation models. arXiv preprint arXiv:2509.10151,

    work page arXiv

  13. [13]

    Arrhythmia classification on ecg using deep learning

    A Rajkumar, M Ganesan, and R Lavanya. Arrhythmia classification on ecg using deep learning. In 2019 5th international conference on advanced computing & communication systems (ICACCS), pages 365–369. IEEE,

    work page 2019

  14. [14]

    OpenAI GPT-5 System Card

    Aaditya Singh, Adam Fry, Adam Perelman, Adam Tart, Adi Ganesh, Ahmed El-Kishky, Aidan McLaughlin, Aiden Low, AJ Ostrow, Akhila Ananthram, et al. Openai gpt-5 system card.arXiv preprint arXiv:2601.03267,

    work page internal anchor Pith review Pith/arXiv arXiv

  15. [15]

    Gemini: A Family of Highly Capable Multimodal Models

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

    work page internal anchor Pith review Pith/arXiv arXiv

  16. [16]

    Gemma: Open Models Based on Gemini Research and Technology

    Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupatiraju, Shreya Pathak, Laurent Sifre, Morgane Rivière, Mihir Sanjay Kale, Juliette Love, et al. Gemma: Open models based on gemini research and technology.arXiv preprint arXiv:2403.08295,

    work page internal anchor Pith review Pith/arXiv arXiv

  17. [17]

    Tengda Zhou, Shaoyang Men, Jingxian Liang, Baoxian Yu, Han Zhang, and Xiaomu Luo

    doi: 10.1109/ICME59968.2025.11209476. Tengda Zhou, Shaoyang Men, Jingxian Liang, Baoxian Yu, Han Zhang, and Xiaomu Luo. 1d u-net++: an effective method for ballistocardiogram j-peak detection.Journal of Mechanics in Medicine and Biology, 21(10):2140058,

    work page doi:10.1109/icme59968.2025.11209476 2025

  18. [18]

    We first compute class frequencies and define target relative abundances for non-normal classes with respect to the normal (N) class. For each non-N beat, we generate additional fixed-length (10 s) segments by re-anchoring the window so that the beat’s R-peak appears at predefined fractional offsets along the segment axis. Rarer classes are assigned more ...

    work page 2021

  19. [19]

    On MIT-BIH Arrhythmia, MIT-BIH superventricular, and Incart, the classifier exhibits a strong tendency 16 (a) Confusion matrix on MIT-BIH Arrhythmia

    Table 6: Micro-F1 confidence thresholds for tool-use decision Dataset MIT-BIH Arrhythmia MIT-BIH Supraventricular INCART VitalDB Threshold 0.990529 0.980933 0.993532 0.98519 Figure 4 presents the row-normalized confusion matrices for the tool-use decision across datasets. On MIT-BIH Arrhythmia, MIT-BIH superventricular, and Incart, the classifier exhibits...

    work page arXiv 2020

  20. [20]

    narrow QRS

    LLM: [77:N] [370:N] [663:N] [947:N] [1231:N] [1515:N] [1809:N] [2045:N] [2403:N] [2706:N] [2998:N] [3283:N] [3560:N] LLM: Call_tools[Confidence Calculator] Tool_output: The confidence is 0.989392. which is lower than Threshold 0.990529, Call Feature Extractor and Morphology Analyzer. LLM: Call_tools[Feature Extractor] Tool_output: Best ECG-classification ...

    work page 2045