Дослідження використання нейронних мереж для автоматизації діагностики на основі рентгенівських знімків

Abstract

The research focuses on improving automated pneumonia diagnosis using deep convolutional neural networks (CNNs). A modified DenseNet-121 archІТecture was proposed, incorporating adaptive hyperparameter tuning and enhanced preprocessing techniques. The Adam optimization algorІТhm was employed to refine weight updates, ensuring efficient training convergence. Data augmentation techniques, including affine transformations, were applied to enhance model generalization. Experimental results demonstrated a classification accuracy of 97.4%, outperforming baseline models.