Multi-Feature Clustering Approach for Firearm Wound Identification on CT Images

Damage evaluation and trajectory analysis are critical to the emergency treatment firearm wound which is disturbed by complex wound shape and heterogeneous filling materials. Consequently, accurate identification firearm wound is essential to evaluate the firearm wound. In this study, a firearm wound identification algorithm based on multi-feature clustering was presented. This identification algorithm was divided into three stages: feature extraction, K-means clustering and Gaussian Mixture Model clustering. Six features were extracted from porcine CT volume data, and clustering results from k-means clustering method were used as the input Gaussian Mixture Model. The average accuracy, sensitivity, specificity and Dice similarity coefficient were 0.92, 0.95, 0.63 and 0.53, respectively. Our results showed that the hybrid method with six features was a potential method to identify complex firearm wound.