EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

PhysioGen-KAN: A Transparent and Uncertainty-Aware Probabilistic Framework for Adaptive Physiological Signal Processing

Yajnaseni Dash, Vinayak Gupta, Sudhir C. Sarangi, B. K. Panigrahi, Ajith Abraham, Pradeep Kumar

IEEE Open Journal of the Computer Society · 2026 · 7 · 1119-1131

DOI: 10.1109/ojcs.2026.3692803 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Open Journal of the Computer Society, 2026. Verfasst von Yajnaseni Dash und 5 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Manipal University.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

As the adoption of black-box artificial intelligence (AI) in healthcare increases, there is a growing need for models that are not only accurate but also interpretable, computationally efficient, and reliable for clinical decision-making. In this work, we propose PhysioGen-KAN, a probabilistic and uncertainty-aware framework based on Kolmogorov-Arnold Networks (KANs), designed for modeling multivariate physiological time-series data. The proposed architecture extends the static KAN formulation into a temporal setting through modality-specific encoders that capture the distinct dynamics of physiological signals such as glucose, heart rate, and physical activity. In addition, a probabilistic aggregation head is introduced to model the predictive distribution of insulin dosage, enabling uncertainty-aware inference. Extensive experiments on insulin dosage prediction demonstrate that the proposed framework achieves strong predictive performance, with a mean absolute error of 4.12 mg/dL, while maintaining high computational efficiency (0.63 GFLOPs) and interpretability (score of 0.91). Compared to established baselines including Long Short-Term Memory (LSTM), Temporal Convolutional Networks (TCNs), and Transformers, the model consistently achieves improved or competitive performance while remaining significantly more lightweight. Clinical relevance is further validated using Clarke Error Grid Analysis, where 95.0% of predictions fall within clinically acceptable regions, indicating a favourable safety profile. These results highlight the potential of PhysioGen-KAN as a transparent, efficient, and uncertainty-aware framework for physiological signal modeling, with promising implications for deployment in resource-constrained healthcare settings.

Förderung

  • Manipal University

Zitieren

APA

Dash, Y., Gupta, V., Sarangi, S. C., Panigrahi, B. K., Abraham, A., & Kumar, P. (2026). PhysioGen-KAN: A Transparent and Uncertainty-Aware Probabilistic Framework for Adaptive Physiological Signal Processing. IEEE Open Journal of the Computer Society, 7, 1119-1131. https://doi.org/10.1109/ojcs.2026.3692803

Vancouver

Dash Y, Gupta V, Sarangi SC, Panigrahi BK, Abraham A, Kumar P. PhysioGen-KAN: A Transparent and Uncertainty-Aware Probabilistic Framework for Adaptive Physiological Signal Processing. IEEE Open Journal of the Computer Society. 2026;7:1119-1131. doi:10.1109/ojcs.2026.3692803

BibTeX
@article{Dash2026,
  author = {Dash, Yajnaseni and Gupta, Vinayak and Sarangi, Sudhir C. and Panigrahi, B. K. and Abraham, Ajith and Kumar, Pradeep},
  title = {PhysioGen-KAN: A Transparent and Uncertainty-Aware Probabilistic Framework for Adaptive Physiological Signal Processing},
  journal = {IEEE Open Journal of the Computer Society},
  year = {2026},
  volume = {7},
  pages = {1119--1131},
  doi = {10.1109/ojcs.2026.3692803}
}
Kostenlos starten →