This paper investigates the classification of Interstitial Lung Disease (ILD) using a hybrid model architecture incorporating a convolutional neural network (CNN) and a Visual Transformer (ViT). Several configurations were evaluated, including variants with different numbers of filters, transformer layers, and dense units. The results demonstrate that the CNN+ViT combination is capable of achieving high accuracy and stability, some models reaching F1-scores above 0.97 and AUC values around 0.99. Models with larger CNN filter sizes achieved better generalization, while simpler architectures with fewer transformer layers outperformed deeper and more complex ones. The findings confirm that the proposed CNN+ViT architecture provides an effective and robust solution for ILD image classification, establishing a solid foundation for future developments with larger CNN backbones and extended training durations.