Vein Visualization Based on Deep Learning with a Smartphone

Abstract

Intravenous infusion is a method frequently used in medical treatment. Locating veins is the prerequisite for intravenous infusion. At present, vein localization in many medical scenarios is still performed manually. However, when encountering infectious diseases, the difficulty of manually locating vein will increase sharply due to the wearing of corresponding protective measures. Inspired by the successful application of deep learning in the field of semantic segmentation, a lightweight neural network is proposed to realize vein visualization from color skin images. The feature loss is included in the training loss to reduce the noise of the visualization result and enhance the smoothness of the image. The network is deployed to a smartphone, which greatly improves the convenience of vein visualization. Experiments were conducted to evaluate the proposed method and its mobile terminal performance. The experimental results show that the proposed method can clearly locate veins in skin areas, and thus can be used to assist medical treatment.