Proposal of Sparse Feature Extraction Method on Classification of Mandibular Reconstruction

Abstract

In this paper, we propose sparse feature extraction methods using virtual planning data in mandibular reconstruction. Although research to automatically generate surgical plans has been done, the mechanism of estimation was empirically designed based on insights of researchers. If relationships between clinically designed surgical procedures and quantitatively extracted features are clarified, the findings will contribute to the surgery. The proposed method estimates the number of segments and extracts the features by sparse modeling using a past planned data. We conducted experiments to extract dominant features for binary classification on the number of fibular segments. The results showed that three of the front lower jaw angle, the distance between the right angle and the right cutting plane, and the distance between the left mental tubercle and the right cutting plane are particularly important for classification.