Search

Delineation of alveolar bone aids the diagnosis and treatment of periodontal diseases. In current practice, conventional 2D radiography and 3D cone-beam computed tomography (CBCT) imaging are used as the non-invasive approaches to image and delineate alveolar bone structures. Recently, high-frequency ultrasound imaging is proposed as an alternative to conventional imaging methods to prevent the harmful effects of ionizing radiation. However, the manual delineation of alveolar bone from ultrasound imaging is time-consuming and subject to inter and intraobserver variability. This study proposes to use a convolutional neural network-based machine learning framework to automatically segment the alveolar bone from ultrasound images. The proposed method consists of a homomorphic filtering based noise reduction and a u-net machine learning framework for automated delineation. The proposed method was evaluated over 15 ultrasound images of tooth acquired from procine specimens. The comparisons against manual ground truth delineations performed by three experts in terms of mean Dice score and Hausdorff distance values demonstrate that the proposed method yielded an improved performance over a recent state of the art graph cuts based method.

Malocclusion, a common dental anomaly, can lead to oral function issues such as difficulty in jaw movement, chewing, speech, and high susceptibility to periodontal diseases. High resolution cone-beam computed tomography (CBCT) images provide sharp visualization of the alveolar bone while delineation of gingiva is inferior. Ultrasound (US) images can delineate the thickness and extension of gingiva while partially showing the thickness of alveolar bones. Accurate registration of CBCT and US images of the tooth-periodontium would allow oral clinicians to visualize the gingiva and alveolar bone for proper diagnosis and orthodontic treatment. A probability-based point set registration using coherent point drift algorithm is proposed to tackle the problem of aligning CBCT and US images. The proposed algorithm was evaluated with images from human volunteers. The results indicate that the method is reasonably robust, and yield mean errors below the clinically accepted tolerance level of 0.5 mm. The limits of agreement between CBCT and US images measured in terms of landmark-based evaluations were [−0.22 mm, 0.26 mm] for intra-rater and [−0.19 mm, 0.21 mm] for inter-rater, respectively. More data are currently acquired to validate the accuracy and efficacy of the method.

Alveolar bone is a part of the periodontium complex supporting the teeth. Conventional radiography and cone-beam computed tomography are currently used to image the alveolar bones. Recently, ionizing radiation-free ultrasound has shown promising potential to image dento-periodontium. However, the ability to visualize alveolar bones in ultrasound images is a challenge for the dentists who are novice to ultrasonography. This study proposes a semi-automated technique to segment alveolar bone by using a multi-label graph cuts optimization approach, where the K-means clustering of intensity values was used in constructing the initial graph. A homomorphic filter was employed as a preprocessing step to de-noise the ultrasound data. The approach was evaluated by over 15 ultrasound images acquired from fresh porcine specimens. Four quantitative evaluators, namely Dice coefficient, sensitivity, specificity, and Hausdorff distance were measured from the proposed method and the manual ground truth by an expert orthodontist. The inter-rater and intra-rater variabilities were also calculated using the delineations by three raters with different levels of experience. The study has demonstrated that the proposed segmentation method provides consistent, reliable, and accurate results among raters and thus has potential to be used as a tool to help dentists to delineate alveolar bones for further analysis.

Computer-aided detection (CAD) systems are being increasingly deployed for medical applications in recent years with the goal to speed up tedious tasks and improve precision. Among others, segmentation is an important component in CAD systems as a preprocessing step to help recognize patterns in medical images. In order to assess the accuracy of a CAD segmentation algorithm, comparison with ground truth data is necessary. To-date, ground truth delineation relies mainly on contours that are either manually defined by clinical experts or automatically generated by software. In this paper, we propose a systematic ground truth delineation method based on a Local Consistency Set Analysis approach, which can be used to establish an accurate ground truth representation, or if ground truth is available, to assess the accuracy of a CAD generated segmentation algorithm. We validate our computational model using medical data. Experimental results demonstrate the robustness of our approach. In contrast to current methods, our model also provides consistency information at distributed boundary pixel level, and thus is invariant to global compensation error.