Automatic Laparoscopic Lens Contamination Detection Based on ResNet18 and Corresponding Cleaning Device Prototype

Introduction: During minimally invasive surgery, laparoscopic lenses often get contaminated by fog and smoke, reducing video quality and affecting surgeon visibility. Various solutions have been proposed to detect and eliminate contamination in real time without interrupting the procedure.

Materials and Methods: This study utilizes the ResNet 18 architecture, modifying the output layer to contain five units corresponding to the classes in the Laparoscopic Video Quality (LVQ) database: noise (NO), smoke (SM), uneven illumination (UI), defocus blur (DB), and motion blur (MB). A review of existing patents revealed two significant systems for automatically detecting laparoscopic lens contamination. The first patent, by Ding et al., involves a system that activates a cleaning mechanism when image clarity falls below a threshold using pressurized liquid, air, and suction. However, it lacks details on the detection threshold and specific features used. The second patent, by Coffeen et al., describes a fluid-based cleaning system activated upon detecting lens deposits, but it does not specify the detection criteria. Both patents have limitations in the cleaning process. Inspired by Coffeen et al.'s patent, we propose an improved in vitro laparoscopic lens-cleaning device. Our device integrates a ResNet18-based detection system, automatically triggers cleaning, and optimizes fluid flow with angled nozzles. It uses warm saline and carbon dioxide, is safe for the human body, and fits through a standard-size trocar.

Results: Our detection system showed high performance, with 99.50% accuracy in training and 99.15% in validation. Performance metrics on 20 distorted LVQ videos revealed 100% accuracy for smoke and motion blur, 90% for noise and defocus blur, and 65% for uneven illumination.

Conclusion: The model demonstrates robust accuracy, particularly in detecting smoke, motion, and defocus blur, facilitating automated cleaning without surgeon intervention. Future research will test the lens-cleaning device prototype in real-world conditions and compare it with other cleaning devices.