Kolmogorov–Arnold Networks for predicting carotid intima-media thickness in cardiovascular risk assessment

• Published in: Scientific reports Vol. 15; no. 1; pp. 32108 - 20
• Main Authors: Al Bataineh, Ali, Vamsi, Bandi, El-Abd, Mohammed, Doppala, Bhanu Prakash
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 01.09.2025; Nature Publishing Group; Nature Portfolio
• Subjects: 639/705/1042; 639/705/117; Accuracy; Adult; Aged; Algorithms; Arteriosclerosis; Artificial intelligence; Atherosclerosis; Cardiovascular disease; Cardiovascular diseases; Cardiovascular Diseases - diagnosis; Cardiovascular risk prediction; Carotid Intima-Media Thickness; Carotid Intima-Media thickness (CIMT); Datasets; Female; Health care; Health risks; Humanities and Social Sciences; Humans; Kolmogorov–Arnold network (KAN); Machine learning; Machine learning models; Male; Medical imaging; Middle Aged; multidisciplinary; Neural networks; Neural Networks, Computer; Patients; Risk assessment; Risk Assessment - methods; ROC Curve; Science; Science (multidisciplinary); Statistical methods; Ultrasonic imaging; Variance analysis; Carotid Intima-Media thickness (CIMT); Machine learning models; Cardiovascular risk prediction; Kolmogorov–Arnold network (KAN)
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-025-14869-1

Carotid Intima-Media Thickness (CIMT) is defined as a non-invasive and well-validated sign of asymptomatic atherosclerosis and an early predictor of cardiovascular disease (CVD). We assembled a carefully curated dataset of 100 adult patients, encompassing 13 clinical, biochemical and demographic variables routinely collected in outpatient practice. After a five-stage pre-processing pipeline median/mode imputation, categorical encoding, Min–Max scaling, inter-quartile-range outlier removal and SMOTE-NC balancing we trained a Kolmogorov–Arnold Network (KAN) to assign each patient to one of four CIMT-defined risk tiers mentioned as “No”, “Low”, “Medium”, “High”. Feature-selection tests (Spearman, Pearson, ANOVA and χ²) removed redundant predictors and improved interpretability. The KAN, implemented with ELU-activated hidden layers and a Softmax output was benchmarked against six conventional algorithms like Support Vector Machine, Decision Tree, Logistic Regression, Stochastic Gradient Descent, Deep Neural Network, Random Forest and Multi-Layer Perceptron. On stratification of five-fold cross-validation the proposed model achieved 93% accuracy, 93% precision, 93% recall, 91% F1-score and a ROC-AUC of 0.97, outperforming all baseline models by 8–19%. These results demonstrate that KAN’s capacity in capturing arbitrary connections and handling multi-class tasks demonstrating its potential as a low-cost and promising tool for early cardiovascular risk hierarchy.