Trabecular Bone Segmentation Based On Segment Profile Characteristics Using Extreme Learning Machine On Dental Panoramic Radiographs

Rizqi Okta Ekoputris, Agus Zainal Arifin, Arya Yudhi Wijaya, Dini Adni Navastara


Dental panoramic radiograph contains a lot of Information which one of them can be identified from trabecular bone structure. This research proposes segmentation of trabecular bone area on dental panoramic radiograph based on segment profile characteristics using Extreme Learning Machine as classification method. The input of this method is dental panoramic radiograph. The selection of region of interest (ROI) is performed on the lower jawbone of the trabecular bone area in which there are teeth and cortical bone. The ROI is subdivided into two where the upper ROI contains the teeth and the lower ROI contains cortical bone. After that, the result of the ROI deduction is done by preprocessing using mean and median filters for upper ROI and motion blur filter for lower ROI. The separate images are extracted each pixel into four features consisting of image intensity, 2D Gaussian filter with two different sigma, and Log Gabor filter for upper ROI. For lower ROI, five feature extractions are image intensity, Gaussian 2D filter with two different sigma, phase congruency, and Laplacian of Gaussian. Then used some sample pixels as training data to create Extreme Learning Machine model. The output of this classifier is the segmentation area of trabecular bone. On the upper ROI, the average of sensitivity, specificity, and accuracy were 82.31%, 93.67%, and 90.33%, respectively. While on the lower ROI obtained the average of sensitivity, specificity, and accuracy of 95.01%, 96.50%, and 95.29%, respectively.


dental panoramic radiograph; Extreme Learning Machine; trabecular bone; segmentation

Full Text:



  • There are currently no refbacks.