[17a-A31-1~5] 4.3 Laser sources and Laser applications

10:45 AM - 11:00 AM JST(1:45 AM - 2:00 AM UTC)English Presentation

[17a-A31-1] 3D Reconstruction of Veins Using NIR by Efficientnet Model

Phuong Anh Dam¹, 〇(M1)Hoang Nhut Huynh¹, Tan Loc Huynh¹, Kien Vinh Vuong¹, Ngoc An Dang Nguyen¹, Anh Tu Tran¹, Trung Nghia Tran¹ (1.Ho Chi Minh City University of Technology (HCMUT), VNUHCM)

Keywords:

near-infrared imaging,transillumination,vascular structure

Near-infrared (NIR) transillumination imaging offers a promising method for visualizing light-absorbing structures within biological tissues. Despite its potential, NIR-TI faces challenges owing to severe blurring caused by light scattering and limitations in in-depth perception from 2D images. This study addresses these issues through depth estimation using a convolutional neural network (CNN) model and pixel-to-pixel scanning. This study introduces a novel approach employing an EfficientNet CNN model to reconstruct 3D structures from 2D NIR images. Through a training process and parameter adjustment, the model demonstrated proficiency in estimating depths within biological tissues, particularly in mapping vascular structures on the back of the hand. Simulation experiments and evaluation of real biological tissues corroborate the efficacy of the proposed method. Compared with previous techniques, the EfficientNet model exhibits improved sharpness and reduced pixel loss, especially at the object edges. By integrating depth estimation with scatter deblurring, this study achieved notable advancements in reconstructing 3D vascular maps from 2D NIR images.