Development of a Head Radiography Simulator System Using Pose Estimation Techniques

Abstract

This study proposes a system that can simulate head radiography by combining a technique for estimating human posture from moving images (hereafter referred to as "pose estimation technique") and use of two cameras capable of acquiring RGB images to determine body position during positioning. The angles of the median sagittal plane (MS), axial plane (AX), and orbitomeatal baseline (OM) were obtained using the pose estimation technique from frontal and lateral images captured after positioning. The resulting radiographs were displayed according to the results. The head tilt during positioning could be determined based on the coordinate data of feature points acquired using the pose estimation technique. In an imaging experiment using a simulated human patient, errors increased as head tilt increased; however, the mean error values in each axis were 0.9° for MS, 0.8° for AX, and 1.5°for OM, when the patient was correctly positioned. The pose estimation technique can assist in evaluating positioning accuracy in radiography and is expected to be used as a potential simulator system.