To assess the accuracy of detecting robot-simulated head movements using video observation (VO) and 3-dimensional head tracking (HT) in a cone beam computed tomography examination setup.Study Design.
A mannequin head was mounted on a robot that was programmed to simulate patient head movements. Six types of movement (tremor, nodding, lateral rotation, lifting, swallowing, and anteroposterior translation), 3 distances (0.25, 1, and 5 mm), and 2 speeds (1 and 5 mm/s) were tested in triplicate (108 examinations). An additional 18 videos and HT of no-motion served as controls for a total of 126 examinations. Three blinded observers viewed video recordings of the examinations, scoring whether the head moved and the movement type. HT provided quantitative measures of movement distance. Accuracy, sensitivity, and specificity for movement detection by VO and HT were calculated, related to true type, distance, and speed of movement. Differences between the true and the measured movement distances were assessed for HT.Results.
VO movement detection presented accuracy of 0.86, sensitivity of 0.85, and specificity of 0.94. Anteroposterior translation (33.3%) and 0.25 mm movements (41.7%) were often not detected by VO. HT correctly detected all cases (accuracy = 1). HT presented small differences between the true and the measured movement distances (average 20-54 μm).Conclusions.
VO missed 41.7% of the 0.25 mm movements. HT correctly detected all movements and quantified movements with an average error <55 μm.