Quantitative Measurement of Pneumothorax Using Artificial Intelligence Management Model and Clinical Application

• DOI: 10.3390/diagnostics12081823
• Publication Date: 2022-07-29T05:41:16Z
• Authors: Dohun Kim; Jae-Hyeok Lee; Si-Wook Kim; Jong-Myeon Hong; Sung-Jin Kim; Minji Song; Jong-Mun Choi; Sun-Yeop Lee; Hongjun Yoon; Jin-Young Yoo
• Keywords: Gold standard (test); Chest radiograph

Artificial intelligence (AI) techniques can be a solution for delayed or misdiagnosed pneumothorax. This study developed, deep-learning-based AI model to estimate the pneumothorax amount on chest radiograph and applied it treatment algorithm developed by experienced thoracic surgeons. U-net performed semantic segmentation classification of non-pneumothorax areas. The was measured using computed tomography (volume ratio, gold standard) radiographs (area true label) calculated predicted label). Each value compared analyzed based clinical outcomes. included 96 patients, which 67 comprised training set others test set. showed an accuracy 97.8%, sensitivity 69.2%, negative predictive 99.1%, dice similarity coefficient 61.8%. In set, average 15%, 16%, 13% in standard, predicted, labels, respectively. label not significantly different from standard (p = 0.11) but inferior (difference MAE: 3.03%). thoracostomy patients 21.6% cases 18.5% cases.