A Descriptive Analysis of Periodontal Bone Loss Patterns Using Ai Assisted Radiographic Interpretation
DOI:
https://doi.org/10.64149/J.Carcinog.24.3s.411-416Keywords:
Periodontitis, periodontal bone loss, artificial intelligence, deep learning, radiographic interpretation, descriptive epidemiology.Abstract
Background: Radiographic assessment of periodontal bone loss is central to diagnosing and staging periodontitis. However, conventional interpretation is vulnerable to inter and intra examiner variability. Artificial intelligence (AI) promises consistent, scalable, and objective quantification.
Objective: To describe the prevalence, distribution, and patterns of periodontal bone loss using AI assisted interpretation of periapical and bitewing radiographs and to compare our findings with previously published literature.
Methods: A descriptive cross sectional analysis of 520 high resolution digital radiographs (4,321 teeth) from adults aged 20–65 years was undertaken. A validated convolutional neural network (CNN) segmented cemento enamel junction (CEJ), alveolar crest, and root apex landmarks to compute bone loss percentage (BL%). Bone loss was categorized as mild (<20%), moderate (20–40%), or severe (>40%). Patterns (horizontal vs. vertical/angular), quadrants, arches, and tooth types were summarized. A 10% subset was manually annotated by two calibrated periodontists to assess agreement (intraclass correlation coefficient, ICC) and model performance (mean absolute error, MAE; Bland–Altman analysis).
Results: The overall prevalence of periodontal bone loss was 78.9% of examined teeth. Mild, moderate, and severe bone loss were observed in 41.6%, 33.2%, and 24.1% of teeth, respectively. Horizontal bone loss predominated in the anterior sextants of both the maxilla (71.8%) and mandible (68.4%), whereas vertical defects were more frequently detected in mandibular molars (25.9%) and maxillary first molars (19.6%). Severe bone loss was significantly clustered in posterior sextants, with an odds ratio of 1.84 compared to anterior regions. AI-assisted measurements demonstrated excellent agreement with expert assessments (ICC = 0.93), with a mean absolute error of 4.2 percentage points.
Conclusions: AI assisted radiographic interpretation reliably characterizes periodontal bone loss patterns. Findings mirror prior reports that vertical/angular defects are posterior predominant, while horizontal loss typifies anteriors. Routine AI integration may reduce variability, enable surveillance at scale, and support risk stratified care pathways.
