Do we skip Hilar Enlargement? Determining Hilar Enlargement on chest x-rays by artificial intelligence

Today, artificial intelligence has a significant potential for use in the medical field. Artificial intelligence can assist physicians in many areas such as diagnosis, treatment, and patient follow-up. The potential for AI to assist in the triage and interpretation of chest radiographs is particularly important because lung radiography is a widely used imaging method that can be accessed easily and quickly by physicians. Recently, significant progress has been made in the development of artificial intelligence algorithms for the analysis of chest radiographs, and the deep learning method of artificial intelligence has become the dominant approach for image analysis. In the presented study, it is aimed to provide deep learning to artificial intelligence to detect patients with hilar enlargement by using chest radiographs, to prepare an application for this purpose and to increase the success of detecting hilar enlargement by using this application in outpatient clinic practices. Patients who had PA chest radiography with certain diagnostic codes at Süleyman Demirel University Research and Application Hospital between 2013 and 2023 were identified retrospectively. After the radiographs were randomized in terms of age and gender, 879 radiographs were evaluated by two separate independent researchers, a mutual decision was made; 184 radiographs in which hilar enlargement was detected and 154 radiographs evaluated as normal were included in the study. In total, 255 of 338 posteroanterior chest radiographs were used for artificial intelligence training and 83 for testing purposes. In the study, 54.1% of the cases were female (n = 183), 45.9% were male (n = 155), and the average age was 60.80±9.91 (min: 19-max: 106) years. While 45.6% (n=154) of the cases were patients with normal PA chest radiographs, 54.4% (n=184) were patients with hilar enlargement on their radiographs. Radiograph results were distributed randomly in terms of age and gender. The sensitivity of the developed artificial intelligence model was found to be 88% and the specificity was 88%. The overall success (F1 score) of the developed artificial intelligence model was found to be 93%. The model developed in the presented study has the potential to assist physicians in distinguishing PA chest radiographs as normal or as hilar enlargement. In this way, it can contribute to eliminating confusion about the presence of hilar enlargement, not to miss hilar enlargement if it is present, determining the underlying cause and initiating appropriate treatment. Early recognition of pathology and early diagnosis of the underlying disease can prevent the progression of the disease and reduce the risk of complications.