Abstract
Extravasation is a condition of extravasated drug was leaked and potentially damaged the site of infusion. In serious injury cases can cause permanent loss of tissue organ. As a clinical workflow in Thailand, nurses have to define the grading and management of extravasation after intravenous (IV) infiltration. But this procedure also has a limitation by the traditional nurses’ shift schedule, which is make the grading and management of extravasation not continuously by the same person. So, this research project proposes an automated screening of extravasation for surveillance, assessment and monitoring during IV therapy. These included 4 steps: data collection, image labeling, convolution neural network modeling, and application development. Total of 1,084 extravasation images were collected from 5 hospitals: Rajavithi Hospital, Nopparatrajathanee Hospital, Rayong Hospital, Songkhla Hospital, and Thammasat Hospital using our own data collecting application. All images were labeled by three dermatologists by locating the skin lesion and rating the severity of extravasation. If no majority vote of any images, the fourth dermatologist will decide the severity of extravasation. After image labeling, all of image data was used to train a densely connected convolutional neural network (DenseNet-121) model, using five-fold cross validation and transfer learning with data augmentation technique. The results from hold-out testing set have a sensitivity and specificity of 96.30% and 93.90% in normal control images, 80.00% and 88.41 in mild extravasation images, 83.33% and 74.12% in moderate extravasation images, 90.91% and 86.74% in severe extravasation images, and 71.43% and 78.43% in others images, respectively. The results from the model were combined to the final application for surveillance, assessment and monitoring extravasation during IV therapy for providing an ease of extravasation management and increasing the efficiency of clinical workflow to reduce the economic cost spent on patient with extravasation in Thailand.
