Deep learning-based vein visualization and its mobile implementation☆

doi:10.1016/j.bspc.2024.107272

Deep learning-based vein visualization and its mobile implementation☆

2025-04-01

0

PRE

AI

DOI：10.1016/j.bspc.2024.107272

Original

Original request for help

Save

Abstract

En 中文

Locating veins is the prerequisite for intravenous cannulation, which is frequently used in medical treatment. At present, vein localization is still performed manually. However, in some special cases, the senses of touch and vision of medical staff will be greatly restricted. For example, in fighting pandemics, medical staff must wear goggles and protective gloves to prevent infection, which will affect the success rate of intravenous cannulation. In this paper, a deep learning-based method is proposed to solve this problem. A lightweight convolutional neural network called VV-Net is proposed to visualize veins from RGB skin images. Feature loss is included in the loss function to emphasize the relationships inside a neighborhood of the output image. A fusion strategy including structure optimization, parameter pruning and post training quantization is presented to further compress the network. Then the model is deployed to a smartphone. Experiments were conducted to evaluate the proposed method and its mobile terminal performance. Subjective observation and objective indices show that the proposed method can achieve good visualization results. The generalization performance as well as the test on skin images with vein disease are also satisfactory. It shows that the proposed method has prospective applications in the future medical treatment.

Keywords:

Intravenous cannulation

Vein visualization

Deep learning

Network model compression

Mobile deployment

Deep learning-based vein visualization and its mobile implementation☆

Abstract

Journal

Organization