Development of a Near-Infrared (NIR) forearm subcutaneous vein extraction using deep residual U-Net

Abstract

Impotence to locate the forearm subcutaneous vein leads to multiple intravenous (IV) attempts causing pain and injuries to patients such as bruise or vein damages. Various technologies and techniques were proposed and developed to overcome the multiple IV access problems. The standard techniques used in research and hospitals are Transillumination, Ultrasound Imaging, and Near-Infrared (NIR). Among those techniques, NIR is the most optimal way of locating the subcutaneous vein because of its non-invasive properties, low-cost implementation. The device can be assembled in a small size product. Nevertheless, the NIR forearm images contain noises that cause difficulties in extracting the vein features. Hence, the performance of NIR vein extraction is having the bottleneck of detecting the vein pixel accurately. Many research studies have been conducted to work on the NIR forearm subcutaneous vein detection due to such a limitation. Artificial intelligence is one of the powerful technologies that would benefit this study. However, a limited number of articles were found on the patentability search, and thus we propose an automatic vein extraction algorithm using Deep Residual U-Net architecture. Our algorithm shows 75 percent of the accuracy in extracting the NIR vein from the experiments that tested. These results show the evidence that the Deep Residual U-Net can be applied to extract the NIR vein.