Uncertainty-Aware Organ Classification for Surgical Data Science Applications in Laparoscopy

摘要: Objective: Surgical data science is evolving into a research field that aims to observe everything occurring within and around the treatment process provide situation-aware data-driven assistance. In context of endoscopic video analysis, accurate classification organs in view camera proffers technical challenge. Herein, we propose new approach anatomical structure image tagging features an intrinsic measure confidence estimate its own performance with high reliability which can be applied both RGB multispectral imaging (MI) data. Methods: Organ recognition performed using superpixel strategy based on textural reflectance information. Classification estimated by analyzing dispersion class probabilities. Assessment proposed technology through comprehensive vivo study seven pigs. Results: When tagging, mean accuracy our experiments increased from 65% (RGB) 80% 90% 96% measure. Conclusion: Results showed had significant influence accuracy, MI are better suited for labeling than Significance: This paper significantly enhances state art automatic videos introducing use metric, being first laparoscopic tissue classification. The will released as dataset upon publication this paper.