Defect Detection For Surgical Instrument Quality Assurance

<p dir="ltr">Ensuring the safety of surgical instruments requires reliable detection of visual defects. In current practice, instruments are manually inspected by trained technicians, a process that is labor-intensive, subjective, and prone to human error. Automated defect detection offers a promising solution to reduce mistakes and improve efficiency. However, existing methods are typically trained on natural or industrial images and do not transfer well to the surgical domain. Direct application or fine-tuning of these methods often results in false positives from textured backgrounds, poor sensitivity to subtle defects, and inadequate capture of instrument-specific features caused by domain shift. To address these challenges, this work introduces a method that adapts unsupervised defect detection specifically designed for surgical instruments. The proposed method combines three key components: background masking to suppress irrelevant artifacts, patch-based analysis to enhance sensitivity to fine-grained defects, and efficient domain adaptation via Low-Rank Adaptation (LoRA) to better capture instrument-specific features. Our approach is evaluated on a custom dataset of surgical instruments that includes both normal and defective examples, with defects simulated to resemble realistic contamination and surface damage. Experiments demonstrate that the proposed method substantially outperforms baseline approaches, achieving high pixel-level AUROC scores and producing accurate visual localization of defects. These results highlight the potential of this method to support safer and more efficient clinical quality assurance through automated surgical instrument inspection.</p>