U-DAVIS-Deep Learning Based Arm Venous Image Segmentation Technique for Venipuncture

Arm Venous Segmentation plays a crucial role in smart venipuncture. The difficulties faced in locating veins for intravenous procedures can be diminished using computer vision for vein imaging. To facilitate this, a high-resolution dataset consisting of arm images was curated and has been presented in this study. Leveraging the ability of Near Infrared Imaging to easily detect veins, ambient lighting conditions were created inside a small enclosure to capture the images. The acquired images were annotated to create the corresponding masks for the dataset. To extend the scope and assert the usability of the dataset, the images, and corresponding masks were used to train an image segmentation model. In addition to using basic preprocessing and image augmentation based techniques, a U-Net based algorithmic architecture has been used to facilitate the task of segmentation. Subsequently, the results of performing image segmentation after applying the preprocessing methods have been compared using various evaluation metrics and have been visualised in the study. Furthermore, the possible applications of the presented dataset have been investigated in the study.