ACCURACY AND RELIABILITY OF FULLY AUTOMATED ARTIFICIAL INTELLIGENCE DRIVEN WEB BASED SOFTWARE ON POSTEROANTERIOR CEPHALOGRAM- A RETROSPECTIVE STUDY

KARTHICK NATRAJAN,
KARTHIKRAJA SELVARAJAN,*
Ajith Geevee,
elankalai elango,
Hanumanth Sankar,
prabhakar krishnan

Abstract

BACKGROUND AND OBJECTIVES: Lateral cephalogram plays a vital role in orthodontic diagnosis and treatment planning, from the advent of cephalometric radiography, whereas limited attention was given to posteroanterior cephalogram which evaluates skeletal discrepancies in transverse dimension. Thus, our study focuses on comparing the accuracy and reliability of fully automated AI driven software ‘WebCeph’, a semi automated software ‘EasyCeph’ and manual tracing on posteroanterior cephalogram. MATERIALS AND METHODS: A sample of 30 pretreatment posteroanterior cephalometric radiographic digital images were collected and analyzed using manual tracing, WebCeph and EasyCeph with 20 landmarks (16 linear and 2 angular measurements). For manual tracing, hard copies of digital images were obtained, whereas for EasyCeph and WebCeph, direct digital images used for analysis. ANOVA Test was done to compare the measurements of each parameter among WebCeph, EasyCeph and manual tracing. Bonferroni post hoc test was performed for individual comparison among 3 groups. The intraoperator reliability were evaluated by correlation coefficient test after 1 month by retracing 5 randomly selected cephalograms. RESULTS: On comparing three methods, 6 out of 18 parameters showed statistically significant differences. On individual comparison, EasyCeph shows significant differences with its counterparts, whereas WebCeph and manual tracing values shows no statistically significant difference. The intraclass correlation coefficient shows strong correlation for manual tracing (0.84 to 0.95) and EasyCeph (0.78 to 0.89) CONCLUSION: The automated cephalometric measurements from WebCeph are reasonably consistent, accurate and reliable when compared with Manual tracing and EasyCeph.