Artificial Neural Networks as a Support and Learning Tool in Medical Practice

Abstract

This study compared the performance of an artificial neural network (ANN), implemented as a convolutional neural network (CNN), with biological neural networks (BNNs) represented by medical students, residents, and specialists. The task consisted of classifying magnetic resonance imaging (MRI) scans in both binary (physiological vs. pathological) and multiclass settings (physiological, Chiari malformation, cortical degeneration, and brainstem glioma). The CNN, trained in MATLAB, achieved 100% accuracy (acc) (AUC = 0.99) in binary classification and 72.5% acc (AUC = 0.78) in multiclass classification, consistently outperforming all human groups, whose maximum acc reached 50%. Additional metrics (precision, recall, and F1-score) confirmed the network’s robustness, while statistical analyses (chi-square test, 95% CI) revealed no significant correlation between participants’ expertise and diagnostic performance. These findings demonstrate the superior diagnostic capacity of CNNs and emphasize their potential as complementary tools in medical practice. Moreover, they highlight the educational relevance of CNNs, suggesting their role in supporting the development of anatomical and diagnostic skills and in bridging knowledge gaps during medical training.