Rigid scopes are essential medical devices, yet their lenses incur damage over time due to dirt accumulation and external impacts. This damage, manifesting as cloudy or chipped image screens, can disrupt medical procedures. Traditionally, damage assessment relies on subjective judgment, lacking a quantitative index. To address this, our study developed a quantitative evaluation method for lens damage. A dedicated rigid-scope imaging system was built to ensure a consistent imaging environment. Four images per scope were captured under varying conditions with and without focus adjustment and at different viewing angles and the distance between the objective lens and the object was standardized using white paper. The images were processed in Python with edge enhancement and windowing techniques to emphasize damaged areas. Teacher images highlighting only the damage were created and paired with original images to form a dataset. A supervised machine learning model was developed using 156 training sets, 20 validation sets, and 44 test sets. The model’s performance was evaluated using the Dice coefficient, which ranged from 0.13 to 0.85. Notably, lens cracks and fogging achieved higher Dice scores than dust. These results indicate the model’s utility, and future work will focus on improving dust detection using U-net and larger datasets optimally.
Furudate et al. (Thu,) studied this question.