Semi-supervised method for determining the maxillary and mandibular boundaries on panoramic radiographs

Dental radiology techniques (X-ray, tomography) play an important role in the detection, verification of symptoms and diagnosis of many dental and anatomic anomalies in the maxillofacial area that are hard to detect at first sight such as dental disorders, dental caries and inflammation, lesions, missing or impacted teeth, bone and tissue pathologies. The process of diagnosing anomalies in maxillofacial relies on examining of X-ray images by one or more clinicians. The fact that variation in dental anomalies and anatomic structures is higher and the fact that standardization of X-ray images is quite difficult constitutes some drawbacks against diagnosing and curing the anomalies successfully. In addition, in some cases where certain complications and findings are similar to other disorders' complications and findings, the diagnosis and treatment process can take a lot of time and the diagnosis and treatment method can vary from dentist to dentist. In order to minimize these difficulties and prevent such problems, as in other radiology fields, imaging techniques and computer-aided diagnostic systems of these images can be useful in the field of dental radiology, in order to facilitate the diagnosis process and to reduce the error rate. These studies focus on increasing the quality, resolution and sensitivity of radiography images, segmentation of areas that may cause risk of disorders on various structures (teeth, jaw, etc.) and classification of anomalies in these structures. One of these studies is the semi-automatic segmentation of maxillary and mandible structures in dental X-ray images. In this study where the method developed for the segmentation of maxillary and mandibular structures is analyzed, the image-specific mask identification method is applied to the segmentation of panoramic radiography images. This mask is defined as the initial input to the segmentation algorithm, and it is ensured that these structures are segmented with an expected active contour based algorithm that is required to give the boundaries closest to the structures to be determined. Once the desired areas are segmented, a smoothing method is applied to the segmentation boundaries and the results of the segmentation are aimed to obtain results similar to the actual boundaries of the structures, which have softer corner passes rather than sharp transitions. In order to determine the similarities the results of the proposed method with the estimated results where the areas are predicted by dentists, Sorensen-Dice measure is used.