Incorporating spatial information and line feature on adaptive classifier for trabecular bone segmentation

Dental Panoramic Radiograph (DPR) is a two-dimensional (2-D) image of teeth that captures the entire structure of the mouth. DPR contains much informations such as trabecular bone structure that can be used to identify many diseases. However, it is hard to determine the area of trabecular bone in DPR because of low contrast, uneven lighting and high amount of noise in the image. In this research, we propose incorporation of spatial information and line feature as a feature data for adaptive classifier. Features of Region of Interest (ROI) from DPR will be extracted using Gabor filter. Gabor filter's orientation will be adjusted with dominant orientation of objects inside ROI. Two spatial informations, mean value of neighboring pixels intensity and Y-axis coordinate of the pixels are extracted as well. The extracted feature will be clustered by K-means Clustering into two classes: area of trabecular bone and area of non-trabecular bone. Some pixels that have ambiguous membership in its cluster because its feature data differ too much with the cluster's centroid feature data, will be classified further using RGDT to prevent false classification. Testing is concluded on 30 ROI images from DPR. The result shows that our proposed method gives an accurate trabecular bone area segmentation result with average accuracy, sensitivity and specificity of 92.52%, 91.67%, and 90.90% respectively.