Automatic detection Using Deep Convolutional Neural Networks for 11 Abnormal Positioning of Tubes and Catheters in Chest X-ray Images

Abstract

Tubes and Catheters are very important devices for saving patients' lives. There is a variety of tubes and Catheters; those especially used during this study are: Endotracheal tube (ETT), Nasogastric (NG)], and Swan Ganz catheter. Errors in positioning these kinds of devices, if not detected early my caused crucial complications (even death). Airway tube malposition in adult patients intubated is seen in up to 25% of cases. Doctors and nurses use checklists to make sure the medical procedure goes smoothly, but these steps take a long time and more resources with the possibility of human errors during verification protocols especially when hospitals are at full capacity. In this article, we propose using transfer learning to train and compare several Keras applications on classification tube problems; the best-selected networks can help in the development of CAD (Computer Aided Detection). The main advantage of using a single Deep Convolutional Neural Network DCNN to detect abnormal positioning of several lines based on chest X-ray image processing is to avoid the complexity caused by using a DCNN (Deep Convolutional Neural Network) network for each type of line. Efficient DCNN can detect abnormal positioning in real-time and immediately notify doctors to adjust tube position. All tested networks during this work are improved after augmentations and parameters tuning, we get the best score for Resnet50V2 model AUC (80%).