Sexual dimorphism in aortic architecture and its segmental abdominal influence using neural networks

Introduction: Sex-related anatomical differences in the aorta, modeled using neural networks, can reveal distinct morphological and functional patterns Objective: To analyze, by sex, the influence of thoracic aortic segments on abdominal aortic diameter using a multilayer perceptron neural network. Methods: This observational, cross-sectional, and analytical study used secondary data from 801 patients who underwent non-contrast thoracic computed tomography. Aortic diameters were measured at eight thoracic segments and the abdominal aorta, segmented using deep-learning-based artificial intelligence algorithms. Separate multilayer perceptron neural network models were developed for men and women. Model performance was assessed using mean squared error (MSE) and relative importance analysis of each aortic segment, with k-fold cross-validation. Results: The neural network models included a single hidden layer with 5 neurons for men (n=559) and 8 neurons for women (n=242). Model performance was superior in women, with lower MSE in training (26.17 vs. 66.88) and testing (13.13 vs. 29.76), as well as lower relative error during training (0.301 vs. 0.348). In both sexes, the aortic segment at the diaphragm showed the highest normalized importance (100%). In men, the next most influential segments were the mid-descending aorta (59.1%) and the ascending aorta (28.2%). In women, the proximal aortic arch (27.6%) and the sinotubular junction (11.1%) exhibited the greatest secondary relevance. Conclusions: The segmental contribution to predicting abdominal aortic diameter differs by sex, reinforcing the presence of sex-specific functional and structural vascular dimorphism, as revealed by artificial neural networks.