Research on Automatic Bleeding Detection in Arthroscopic Videos Based on Composite Color and Statistical Features

Arthroscopic surgery is a primary technique for treating joint-related diseases, widely embraced in clinical practice for its minimally invasive and precise nature. However, intraoperative bleeding often generates blood mist, significantly impairing the surgeon’s field of view and necessitating prompt high-flow drainage for clearance. Therefore, accurate bleeding detection and localization is a prerequisite for blood mist removal. This paper introduces a pixel-based feature extraction scheme aimed at detecting bleeding frames in arthroscopic videos. In contrast to previous bleeding detection methods, this approach utilizes statistical features based on composite color to analyze arthroscopic images and extract features. Then, a feature selection strategy is proposed to select the best features from the extracted features.Subsequently, the selected features are fused and then classified using an improved KNN classifier to differentiate between bleeding and non-bleeding images. In addition to this, a post-processing scheme is introduced to enhance bleed frame detection performance by exploiting temporal variations across consecutive frames in arthroscopic videos. Lastly, a region-based detection algorithm is proposed for identifying bleeding zones within images depicting bleeding. By conducting extensive experimental analysis on the arthroscopic image and video dataset. The proposed method achieves accuracies of 95.8%, 97.3%, and 95.3% for bleed frame detection in terms of accuracy, sensitivity, and specificity respectively. The results demonstrate that the proposed algorithm effectively detects bleeding frames and bleeding zones in arthroscopic videos.