In pursuit of our objective, we conducted a comprehensive retrospective analysis. Our dataset comprised 2,262 chest radiographs, which included images featuring nasogastric tubes. These radiographs were sourced from two prominent medical institutions: Gangneung Asan Hospital and Chuncheon Sacred Heart Hospital. To ensure the accuracy and reliability of our dataset, each of the 2,262 chest radiographs was meticulously annotated by two experienced pulmonologists.

The cornerstone of our methodology was the division of 2,062 of these chest X-rays into eight distinct queries, as part of an active learning process. This approach allowed for the iterative improvement of our model's accuracy and efficiency. We employed the nnUNET model, a neural network specifically designed for medical image segmentation. The model's performance was rigorously evaluated using a test set of 200 chest X-rays. Additionally, for the purpose of classification, we analyzed the data derived from the segmentation model, utilizing the generated masks to further refine our results.