Optimal Integration of Improved ECA Module in cGAN Architecture for Hand Vein Segmentation

Segmentation of hand vein images is crucial for various applications, including precise biometric identification and facilitating medical intravenous procedures. This study introduces a method for hand vein image segmentation using deep learning, specifically a conditional generative adversarial network (cGAN). The cGAN is trained adversarially and enhanced with a modified Efficient Channel Attention (ECA) mechanism. The effectiveness of this approach is evaluated using two hand vein datasets: one sourced internally and the other from SUAS. Comparative analysis demonstrates that our method achieves superior sensitivity, accuracy, and dice coefficient on the self-acquired dataset, as well as improved sensitivity and accuracy on the SUAS dataset. Experimental results highlight the significant capability of our segmentation technique in enhancing hand vein patterns and improving the accuracy of dorsal hand vein detection.