Pustaka Ilmiah Universitas Padjadjaran

Abstrak

Characterize Image Segmentation at Dental Periapical Radiograph using Receiver Operating Characteritistic Analysis

Objective: Normally, dental periapical radiography could not be directly used to help a dentist diagnosing the stages of the pulp such as the normal pulp, reversible and irreversible pulpitis and necrotic and judging to be in need of treatment. Image segmentation could be applied to identify its region of interest (ROI) in order to find its original feature extractions with an acceptable accuracy. Our research aims to characterize image segmentation at dental periapical radiography using receiver operating characteritistic (ROC) analysis when various decision attitudes, from interventionist to non-interventionist are held. Material and Methods: Twenty of periapical radiographs were shown to a dentist, who was asked to specify, for the stages of the pulp such as the normal pulp, the reversible and irreversible pulpitis and the necrotic could be extracted. Feature extraction analysis is treated by applying bilateral filter of an image sampled to find the best contrast of the filtered image. It was needed to tabulate data using a descriptive statistical method. These secondary data could be obtained by processing through classification as well as the operational criteria.The original image segmented with an acceptable accuracy could be constructed by plotting the sensitivity (or true positive prediction) of decisions made against the false positive prediction (equivalent to 1-specificity). Results: Image segmentation validated by the ROC curve had been compared with its original image segmentation where it was found the range of accuracy from 98% to 99%, sensitivity from 90% to 99%, and specificity between 98 % and 99%. Conclusion: Image segmentation using ROC analysis has been acceptable to the sensitivity and the specifivity but the sensitivity tends to be improved cause the lower intensity if only if the periapical radiography is characterized at the frequential domain.