LVQE: Laparoscopic Video Quality Enhancement Using GAN-Based Smoke Elimination Guided by an Embedded Dark Channel
Akib Jayed Islam, Sultanus Salehin, Sayem Ul Alam, M. Kamrul Islam, Shreya Paul, Prithy Paul · 2024
Inthis study, a technique for effectively removing smoke from laparoscopic video recordings-which are often negatively impacted during surgeries-using an image-to-image generative adversarial network (GAN) with a built-in guide mask derived from the dark channel is presented. Smoke from tissue dissection tools can obstruct the surgeon's view and introduce errors into computer vision-based systems employed for clinical navigation. The GAN addresses this issue by eliminating noise while maintaining perceptually satisfactory image quality. Using the Laparoscopic Video Quality (LVQ) evaluation dataset, the method is trained and assessed. The section on experimental results provides qualitative examples demonstrating the effectiveness of the proposed approach. With a larger, more diverse dataset for training and validation, further improvements are anticipated. The proposed technique generated smoke-free images and videos, surpassing previous methods.