KACQ-DCNN: Uncertainty-Aware Interpretable Kolmogorov-Arnold Classical-Quantum Dual-Channel Neural Network for Heart Disease Detection

Heart failure remains a critical global health issue, contributing significantly to cardiovascular disease and accounting for 17.8 millions of annual deaths. The need for innovative diagnostic strategies is pressing, as classical machine learning models face challenges such as handling complex, high-dimensional data, class imbalances, poor categorical feature representations, limited performance on small datasets, and the absence of uncertainty quantification. Moreover, the interpretability of these models is often hindered by their'black box' nature, complicating clinical trust and decision-making. While quantum machine learning shows potential, existing hybrid models have yet to fully capitalize on quantum advantages. To address these gaps, we propose Kolmogorov-Arnold Classical-Quantum Dual-Channel Neural Network (KACQ-DCNN), a novel hybrid dual-channel neural network that integrates Kolmogorov-Arnold Networks (KANs) in place of traditional multilayer perceptions, enabling univariate learnable activation functions on edges. As an early adopter of KAN components, we observed that the approach significantly improved the ability of the network to approximate continuous functions with reduced complexity and improved generalizability. Our comprehensive evaluation demonstrates that the KACQ-DCNN 4-qubit 1-layered model outperforms 37 benchmark models, including 16 classical machine learning models, 12 quantum neural networks, six hybrid models, and three variants of the KACQ-DCNN architecture. It achieved an accuracy of 92.03%, along with a macro-average precision, recall, and F1 score of 92.00%, representing significant improvements across all metrics. Moreover, KACQ-DCNN achieved a ROC-AUC score of 94.77%, supported by two-tailed paired t-tests against nine top-performing models, with a significance level (α) of 5% and a Bonferroni correction applied (α