Random errors in localization of landmarks in postero-anterior cephalograms.

The aim of the present study was to evaluate the random error in localization of the most common landmarks in postero-anterior cephalograms (PAC). The study took place at the Department of Orthodontics of Aarhus University during the period 1993-1995. The material consisted of 30 standardized PAC taken in natural head position. Five examiners had to identify 34 landmarks on each cephalogram. Subsequently, all examiners had to identify again the same 34 landmarks on one randomly selected cephalogram five times with a time interval of at least 24 hours. All landmarks were digitized, related to an X-Y co-ordinate system, and an arithmetical mean was calculated. The accuracy of digitizing was evaluated by digitizing one randomly selected cephalogram 10 times. The main findings of this study are: (1) The digitizing error is negligible compared to the errors introduced by landmark identification. (2) Each landmark has its own characteristic pattern of variance, which is very similar on both sides. (3) Significant differences in accuracy exist between the various postero-anterior landmarks. The six most accurate landmarks are mastoid left (l) and right (r), latero-orbitale (l) and (r), and antegonion (l) and (r). The six least accurate landmarks are coronoid (l) and (r), condylar (l) and (r), and mandibular foramen (l) and (r). (4) A significant difference in the accuracy of landmark identification between the five examiners was only seen for seven of the 34 landmarks. (5) No evidence was found that one examiner was consistently better/worse than the others. (6) No improvement in the accuracy was found after repeated identification, thus there seems to be no short-term 'learning process'. Refereed Paper