Cephalometric Landmark Detection: Artificial Intelligence vs Human Examination

Abstract

Background: Cephalometric landmark detection is important for accurate diagnosis and treatment planning. The most common cause of random errors, in both computer-aided cephalometry and manual cephalometric analysis, is inconsistency in landmark detection. These methods are time-consuming. As a result, attempts have been made to automate cephalometric analysis, to improve the accuracy and precision of landmark detection whilst also minimizing errors caused by clinician subjectivity. Aim: This mini-thesis aimed to determine the precision of two cephalometric landmark identification methods, namely an artificial intelligence programme (BoneFinder®) and a computer-assisted examination software (Dolphin ImagingTM). Methods: This was a retrospective quantitative cross-sectional analytical study. The dataset comprised of 409 cephalograms obtained from a South African population. 19 landmarks were selected and detected using a computer-assisted approach and an automatic approach. The x,y coordinates for each landmark per system was recorded and the Euclidean distance was calculated. Precision was determined by calculating the standard deviation and standard error of the mean. Results: The primary researcher acted as the gold standard and was calibrated prior to data collection. The inter- and intra-reliability tests yielded acceptable results. There were variations present in several landmarks between Dolphin and BoneFinder; however, they were statistically insignificant. The computer-aided approach was very sensitive to several variables. Attempts were made to draw valid comparisons and conclusions.