Hand blood vessels pattern recognition

Tan, Xuan Qing (2023) Hand blood vessels pattern recognition. Final Year Project, UTAR.

Preview

PDF Download (3038Kb) | Preview

Abstract

This study aimed to address the problem of difficult venous access by proposing a vein feature extraction algorithm for the forearm using transfer learning on a U-net model with EfficientNetB3 as the backbone. The limited availability of forearm Near Infrared (NIR) image datasets and the lack of vein feature extraction algorithms focused on the forearm part were the main research problems. To evaluate the proposed model, the NTUIFDB v1 dataset containing 250 NIR forearm images was used and the performance was measured using Dice Coefficient and Jaccard Index. The results showed that the proposed model achieved an 83.56% Dice Coefficient and 71.76% Jaccard Index, outperforming four handcrafted techniques tested and other pre-trained models. This research contributes to the field by being the first to implement transfer learning on the NTUIFDB v1 dataset and provides a baseline for future studies to improve the proposed model. The proposed algorithm could aid in improving the success rate of venipuncture in patients with difficult venous access, such as pediatrics, geriatrics, obesity, and dark skin tone patients.

Actions (login required)