CardioFusion-AI: Robust ECG--PPG Fusion for Multimodal Physiological Monitoring Under Signal Degradation

Navaneetha Krishnan Kamalakannan; Janakiraman Kamalakannan

eess.SPAug 26, 2026

CardioFusion-AI: Robust ECG--PPG Fusion for Multimodal Physiological Monitoring Under Signal Degradation

Authors: Navaneetha Krishnan Kamalakannan, Janakiraman Kamalakannan

Organizations: Department of Electronics and Communication Engineering, Saveetha School of Engineering · Government Kilpauk Medical College and Hospital

Abstract

Wearable electrocardiogram (ECG) and photoplethysmogram (PPG) sensors are complementary but individually fragile: motion artifact, poor contact, and sensor dropout can degrade one or both signals. Fusion strategies that assume both modalities are equally trustworthy can become less reliable than a single clean modality under degradation. We present CardioFusion-AI, a framework whose signal-processing front end, including R-peak and systolic-peak detection, an Orphanidou-type signal-quality index, and beat-by-beat pulse transit time estimation, is validated on 53 real intensive-care recordings (848 windows; heart-rate mean absolute error 1.61 bpm for ECG and 2.78 bpm for PPG) and a real annotated fetal ECG database (R-peak F1 0.89-0.98). We then conduct a controlled synthetic degradation study comparing eight ECG-PPG fusion strategies across six degradation regimes spanning graded corruption and complete modality loss, using five independent training seeds. Attention fusion achieved the lowest descriptive overall error (1.66+/-0.43 bpm). Both adaptive gates reallocated weight toward the healthy modality under complete modality loss, but showed near-zero correlation between gate weight and signal quality under graded degradation (r = 0.10-0.24). Signal-quality conditioning produced a specific improvement under missing-PPG conditions (1.56+/-0.59 bpm), approaching the 1.48 bpm unimodal ceiling. With only five training seeds, no pairwise comparison survives Holm-corrected significance testing; effect sizes and confidence intervals are therefore reported. These results indicate that modality availability and modality quality are functionally distinct problems for adaptive fusion.

Figures & tables

Regime	ECG level	PPG level	# grid cells
Both clean	0	0	1
ECG degraded only	1–3	0	3
PPG degraded only	0	1–3	3
Both degraded	1–3	1–3	9
ECG missing	4	0	1
PPG missing	0	4	1

TABLE I: Controlled degradation conditions. “Level” spans clean (0) through severe (3); “missing” (4) denotes complete sensor dropout and is evaluated only against a clean partner modality.

Fig. 1: Shared CardioFusion-AI dual-stream architecture. Each modality is independently preprocessed, quality-assessed, and encoded; the fusion mechanism (the only component that differs across the eight compared strategies) combines the two embeddings into a physiological estimate. The signal-quality index is an explicit gate input only in the SQI-conditioned adaptive variant.

Model	Overall	Both clean	ECG degraded	PPG degraded	Both degraded	ECG missing	PPG missing
ECG-only	2.18 $\pm$ 0.45	1.86 $\pm$ 0.76	1.52 $\pm$ 0.41	1.41 $\pm$ 0.56	1.43 $\pm$ 0.41	14.34 $\pm$ 0.38	1.48 $\pm$ 0.67
PPG-only	2.61 $\pm$ 0.30	1.49 $\pm$ 0.32	1.38 $\pm$ 0.30	1.92 $\pm$ 0.22	2.13 $\pm$ 0.35	1.47 $\pm$ 0.34	15.02 $\pm$ 1.51
Fixed-average fusion	2.63 $\pm$ 0.64	2.39 $\pm$ 0.71	2.11 $\pm$ 0.89	2.14 $\pm$ 0.73	2.17 $\pm$ 0.76	7.01 $\pm$ 0.82	5.77 $\pm$ 1.33
Feature-level fusion	2.87 $\pm$ 0.53	2.53 $\pm$ 0.57	2.17 $\pm$ 0.31	2.25 $\pm$ 0.53	2.40 $\pm$ 0.67	6.95 $\pm$ 0.76	7.28 $\pm$ 1.22
Attention fusion	1.66 $\pm$ 0.43	1.24 $\pm$ 0.26	1.65 $\pm$ 0.39	1.18 $\pm$ 0.18	1.55 $\pm$ 0.31	3.17 $\pm$ 1.91	2.97 $\pm$ 1.73
Global-weighted late fusion	2.67 $\pm$ 0.30	1.98 $\pm$ 0.44	1.83 $\pm$ 0.62	2.05 $\pm$ 0.35	2.26 $\pm$ 0.40	6.86 $\pm$ 0.54	7.24 $\pm$ 0.80

TABLE II: Test-set MAE (bpm), mean

\pm

SD across 5 independent training seeds, by degradation regime.

Fig. 2: Test MAE vs. graded degradation severity (0 = clean to 3 = severe), with the partner modality held clean. Error bars: SD across 5 seeds.

Fig. 3: Test MAE by degradation regime for five representative fusion strategies. Complete modality loss (right two groups) separates strategies far more sharply than graded degradation (left four groups).

Fig. 4: Mean learned gate weight on ECG vs. PPG by degradation regime, for the implicit and SQI-conditioned adaptive gates. Both gates reallocate weight sharply only under complete modality loss (rightmost two regimes).

Comparison	Regime	Mean diff.	$d_{z}$	$p$	$p_{\mathrm{Holm}}$
SQI vs. Attn.	both clean	+0.74	2.51	0.063	1.00
SQI vs. Attn.	PPG degraded	+0.63	1.50	0.063	1.00
SQI vs. Implicit	overall	$-$ 0.07	0.18	0.81	1.00
Implicit vs. Attn.	overall	+0.40	0.56	0.31	1.00

TABLE III: Seed-level paired comparisons (

n=5

), selected regimes. Full 21-test family in supplementary material.

Multimodal Sensor Fusion

Multimodal sensor fusion combines data from different sensors to improve perception, tracking, localization, and prediction.

See all 417 papers

PPGFlowECG: Latent Rectified Flow with Cross-Modal Encoding for PPG-Guided ECG Generation and Cardiovascular Disease Detection
Sep 24, 2025Xiaocheng Fang, Jiarui Jin, Haoyu Wang +8Electrocardiography Conditional Generation
National Institute of Health Data Science, Peking University, China · School of Intelligence Science and Technology, Peking University, China · University of Chinese Academy of Sciences, China +1
CardioState-JEPA: Delay-Aware Cross-Modal Learning of a Shared Cardiac Representation
Aug 13, 2026Hamza Shafiq, Hung Manh Pham, Bin Zhu +3Cross-Modal Representation Learning Joint-Embedding Predictive Architecture
Eindhoven University of Technology, Netherlands · Singapore Management University, Singapore
Biosignal Fingerprinting: A Cross-Modal PPG-ECG Foundation Model
May 10, 2026Zhangdaihong Liu, Chang Liu, Fenglin Liu +4Cross-Modal Representation Learning Masked Autoencoders
Department of Engineering Science, University of Oxford, Oxford OX3 7DQ, UK · Oxford Suzhou Centre for Advanced Research, Suzhou, 215123, China · School of Public Health, Shanghai Jiao Tong University, Shanghai, China

Sep 24, 2025J/K move · Enter open · S save

PPGFlowECG: Latent Rectified Flow with Cross-Modal Encoding for PPG-Guided ECG Generation and Cardiovascular Disease Detection

Xiaocheng Fang, Jiarui Jin, Haoyu Wang, Che Liu+7

National Institute of Health Data Science, Peking University, China · School of Intelligence Science and Technology, Peking University, China · University of Chinese Academy of Sciences, China+1

Electrocardiography Conditional Generation Cross-Modal Representation Learning Photoplethysmography

Open paper AskPDF arXiv

Electrocardiography (ECG) is the clinical gold standard for cardiovascular disease (CVD) assessment, yet continuous monitoring is constrained by the need for dedicated hardware and trained personnel. Photoplethysmography (PPG) is ubiquitous in wearable devices and readily scalable, but it lacks electrophysiological specificity, limiting diagnostic reliability. While generative methods aim to translate PPG into clinically useful ECG signals, existing approaches are limited by the misalignment of physiological semantics in generative models and the complexity of modeling in high-dimensional signals. To address these limitations, we propose PPGFlowECG, a two-stage framework that aligns PPG and ECG in a shared latent space using the CardioAlign Encoder and then synthesizes ECGs with latent rectified flow. We further provide a formal analysis of this coupling, showing that the CardioAlign Encoder is necessary to guarantee stable and semantically consistent ECG synthesis under our formulation. Extensive experiments on four datasets demonstrate improved synthesis fidelity and downstream diagnostic utility. These results indicate that PPGFlowECG supports scalable, wearable-first CVD screening when standard ECG acquisition is unavailable.

Abstract

Multimodal Sensor Fusion

Explore similar work

PPGFlowECG: Latent Rectified Flow with Cross-Modal Encoding for PPG-Guided ECG Generation and Cardiovascular Disease Detection

CardioState-JEPA: Delay-Aware Cross-Modal Learning of a Shared Cardiac Representation

Biosignal Fingerprinting: A Cross-Modal PPG-ECG Foundation Model