Osteoradionecrosis Region Estimation Using Machine Learning

Abstract

Osteoradionecrosis is a disease caused by a bone resorption inhibitor or the radiation therapy to the head and neck cancer. Conservative therapy using antibacterial drug or surgery to remove the necrosis bone have been done in the ORN treatment. Currently, the surgical operation often takes longer time than the pre-operative surgical plan, because it is difficult to recognize the bone necrosis area in the 3D CT image. Therefore, a system to accurately estimates the osteoradionecrosis area in the pre-operative 3D CT image is needed. This paper proposes a method to estimate the osteoradionecrosis area using image texture features and machine learning. Experiments using two osteonecrosis patients CT images showed that the necrosis area was successfully extracted by the F-measure score of 0.729, and we confirmed the necrosis area estimation result through the visual inspection.