Desmoking and Perceptual Quality Restoration of Laparoscopic Videos Using CycleGAN
Sultanus Salehin, Akib Jayed Islam, Sayem Ul Alam, Shreya Paul, M. Kamrul Islam, Prithy Paul · 2024
Laparoscopic surgeries, valued for their minimally invasive nature, often suffer from visual obstructions caused by smoke from electrocauterization, impacting surgical precision and safety. This paper introduces a CycleGAN-based method for desmoking and restoring the perceptual quality of laparoscopic videos. Unlike traditional dehazing techniques, our approach uses inter-channel discrepancy and dark channel prior losses to effectively minimize residual smoke. Our key contributions include developing a novel GAN-based framework tailored for surgical video enhancement and introducing smoke-specific loss functions that improve visual clarity. Evaluated on the LVQ database, our model outperforms existing methods in image quality metrics and visual assessments. A user study with 25 surgeons confirmed its practical utility and potential for clinical integration, enhancing intraoperative imaging and surgical outcomes.