Detection of pneumoperitoneum in the abdominal radiograph images using artificial neural networks

• DOI: 10.1016/j.ejro.2020.100316
• Publication Date: 2020-12-21T17:53:46Z
• Authors: Mimi Kim; Jong Soo Kim; Changhwan Lee; Bo-Kyeong Kang
• Keywords: Artificial neural network; 0301 basic medicine; Abdominal image; Medical physics. Medical radiology. Nuclear medicine; 03 medical and health sciences; Pneumoperitoneum; R895-920; Deep learning; Article

The purpose of this study was to assess the diagnostic performance artificial neural networks (ANNs) detect pneumoperitoneum in abdominal radiographs for first time. This approach applied a novel deep-learning algorithm, simple ANN training process without employing convolution network (CNN), and also used widely utilized method, ResNet-50, comparison. By applying ResNet-50 radiographs, we obtained an area under ROC curve (AUC) 0.916 accuracy 85.0 % with sensitivity 85.7 predictive value negative tests (NPV) 91.7 %. Compared most commonly methods such as CNN, our extremely small structures process. approach, 88.6 NPV 91.3 %, compared decently that ResNet-50. results showed ANN-based computer-assisted diagnostics can be accurately reduce time delay diagnosing urgent diseases pneumoperitoneum, increase effectiveness clinical practice patient care.