Semi-automatic segmentation of apical lesions in cone beam computed tomography images

Abstract

The accurate detection of the border surrounding the apical lesions observed in cone-beam Computed Tomography (CT) images is very important for the accurate calculation of the features of these lesions. This study describes a semi-automatic hybrid segmentation method that will be used to determine the limits of apical lesions observed in three-dimensional (3D) KIBT images. The apical cyst and tumor lesions detected in 3D images obtained from 42 different patients who came to Karadeniz Technical University Oral Diagnosis and Radiology Clinic for routine controls became the dataset of this work. Lesions in the dataset were manually segmented by Oral Diagnosis and Radiology experts via software developed for this study. The 3D regions of interest are once again segmented by the hybrid segmentation method obtained from the algorithms mentioned in this work. After the experiments, the volumetric regions obtained by the manual and semi-automatic methods were compared. Two different similarity indices (Jaccard index, Dice coefficient) were used to verify segmentation results. As a result of the comparisons an average accuracy of 81% Jaccard index and 89% Dice coefficient values was achieved for the semi-automatic method. In conclusion, semi-automated hybrid segmentation is very successful in detecting apical lesions.