Adaptive Multi-Resolution Temporal Convolutional Graph Network with Explainable AI for Real-Time Cardiovascular Event
Computer and Decision Making: An International Journal · 04.11.2026 · 3 (6S) · 1420-1441
DOI: 10.1922/comdem.v3i6s.18921 ↗Kurz gefasst
Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 04.11.2026. Verfasst von Dr. Jyothirmai Digumarthi und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Continuous vital-sign monitoring could alert to cardiovascular events before they happen, but a usable warning model needs to be accurate, small enough for edge hardware, and open to inspection. In this paper, we propose the Adaptive Multi-resolution Temporal Convolutional Graph Network (AMR-TCGN). It combines three dilated causal convolution branches with different temporal scales, an input-conditioned gate to combine these branches, recency-aware attention pooling and a learnable graph over physiological channels. Without additional training, explanations are revealed in the form of gate weights, attention profiles, the channel graph, and input-gradient saliency. Since no ethically cleared clinical stream was available, all models were evaluated on a disclosed synthetic benchmark comprising 180 simulated patients with patient-disjoint splits. AMR-TCGN achieved a test AUROC of 0.961 and F1 of 0.857, but did not outperform an LSTM (AUROC 0.988, F1 0.891) or a single-resolution TCN (0.979, 0.868). The removal of the graph stage or the multi-resolution branches did not reduce accuracy and the 95% intervals on recall overlap for all temporal models. An edge-lite variant with 1,988 parameters preserved an AUROC of 0.955 while being 12.1 times smaller and twice as fast. Our claim is not that AMR-TCGN is more accurate, but that it is an interpretable and efficient alternative.
Zitieren
APA
Digumarthi, D. J., G.Vasavi, D., N.Vimala, D., & B.Giridhar (2026). Adaptive Multi-Resolution Temporal Convolutional Graph Network with Explainable AI for Real-Time Cardiovascular Event. Computer and Decision Making: An International Journal, 3(6S), 1420-1441. https://doi.org/10.1922/comdem.v3i6s.18921
Vancouver
Digumarthi DJ, G.Vasavi D, N.Vimala D, B.Giridhar. Adaptive Multi-Resolution Temporal Convolutional Graph Network with Explainable AI for Real-Time Cardiovascular Event. Computer and Decision Making: An International Journal. 2026;3(6S):1420-1441. doi:10.1922/comdem.v3i6s.18921
BibTeX
@article{Digumarthi2026,
author = {Digumarthi, Dr. Jyothirmai and G.Vasavi, Dr. and N.Vimala, Dr and B.Giridhar},
title = {Adaptive Multi-Resolution Temporal Convolutional Graph Network with Explainable AI for Real-Time Cardiovascular Event},
journal = {Computer and Decision Making: An International Journal},
year = {2026},
volume = {3},
number = {6S},
pages = {1420--1441},
doi = {10.1922/comdem.v3i6s.18921}
}
EUVIMED