Publication: Improved hand vein imaging system based in deep learning techniques
Date
2023-10-01
Authors
Marlina Binti Yakno
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Difficulty in achieving peripheral intravenous (IV) access in some patients is a clinical problem, which may lead to some negative impacts such as fainting, hematoma, and pain associated with multiple punctures. The conventional image processing techniques have been used to aid in locating veins for IV procedures but these have not produced satisfactorily clear vein patterns. Deep learning has become increasingly popular and is able to successfully solve various imaging applications. Therefore, this research focuses on the development of an improved hand vein imaging system based on deep learning techniques. Basically, the system consists of a hand vein image-acquisition and an image processing system. The image acquisition system comprises sets of three near-infrared (NIR) LED arrays with a wavelength of 850nm. The image processing system involves three stages. In the first stage, captured hand images based on different hand poses and locations are fed into Faster R-CNN for vein region of interest (ROI) extraction. In the second stage, the detected ROI vein extracted images are enhanced by an improved weighted average fusion between fuzzy adaptive gamma (FAG) and contrast limited adaptive histogram equalization (CLAHE) technique. Finally, the enhanced vein images are segmented using a conditional generative adversarial network (cGAN) integrated with a modified efficient channel attention (ECA) module as a vein refiner. Simulation investigations show that the proposed techniques have achieved the best performance with average sensitivity, average accuracy, and average dice coefficient of vein patterns of 0.8878, 0.9639, and 0.7904 for the self-acquisition, SUAS, WILCHES, and BOSPHORUS datasets, respectively. These results demonstrate that the proposed deep-learning intelligent systems are able to enhance hand vein image patterns.