Deep Learning Based Method for the Estimation of Patient’s Angles from Lateral Skull Radiographs

Radiography is used for initial diagnosis and postoperative follow-up. If a radiograph deemed unsuitable diagnosis, it rejected. Retaking disadvantageous the patient because prolongs examination time increases radiation dose. Skull radiography position in which retaking occurs most frequently. In skull radiography, patient’s rotational direction estimated from minute changes inner ear’s structure lateral radiograph. When retaking, amount of correction positioning generally errors rejected image through empirical evidence. Therefore, considerable expertise needed to correct appropriately, inexperienced radiologic technologists take estimate this error. This study aimed angle radiographs compensate technologists’ lack experience reduce burden on patients. The simulated 2-D line-integral projection 3-D CT image, we developed an estimation method using deep learning with supervised training. network based re-scaled ResNet. was superior-inferior directions. We evaluated accuracy projected images 256 cases. were 0.48 ± 0.41° 0.55 0.50° angles, respectively. These findings suggest that can be accurately learning, compensating reducing time.