Cone beam computed tomography (CBCT) is a widely-used imaging modality in dental healthcare. It is an important task to segment each 3D CBCT image, which involves labeling lesions, bones, teeth, and restorative materials on a voxel-by-voxel basis, as it aids in lesion detection, diagnosis, and treatment planning. The current clinical practice relies on manual segmentation, which is labor-intensive and demands considerable expertise. Leveraging Artificial Intelligence (AI) to fully automate the segmentation process could tremendously improve the quality and efficiency of dental healthcare. The main hurdle in this advancement is reducing AI’s reliance on a large quantity of manually labeled images to train robust, accurate, and generalizable algorithms. To tackle this challenge, we propose a novel Oral-Anatomical Knowledge-informed Semi-Supervised Learning (OAK-SSL) model for 3D CBCT image segmentation and lesion detection. The uniqueness of OAK-SSL is its capability of integrating qualitative oral-anatomical knowledge of plausible lesion locations into the deep learning design. Specifically, the unique design of OAK-SSL includes three key elements, including transformation of qualitative knowledge into quantitative representation, knowledge-informed dual-task learning architecture, and knowledge-informed semi-supervised loss function. We apply OAK-SSL to a real-world dataset, focusing on segmenting CBCT images that contain small lesions. This task is inherently challenging yet holds significant clinical value as treating lesions at their early stages lead to excellent prognosis. OAK-SSL demonstrated significantly better performance than a range of existing methods. Note to Practitioners—This study tackles the challenges arising from a limited amount of labeled data due to the time-consuming manual segmentation of 3D dental cone beam computed tomography (CBCT) images. The scarcity of labeled data often impedes AI models from accurately segmenting periapical lesions. To overcome this, we introduce a novel semi-supervised learning algorithm that integrates the oral-anatomical knowledge about lesion location for 3D CBCT image segmentation. Our method effectively segments periapical lesions, including even small-sized periapical lesions, without solely relying on labeled data. The proposed method offers two significant benefits to clinicians. First, it reduces the necessity for large amounts of labeled data, particularly easing the burden of manually segmenting early-stage periapical lesions. Second, it helps reduce intra- and inter-observer disagreements and human errors by providing consistent and automated segmentation maps. These benefits not only simplify the segmentation process in dental imaging but also improve its reliability. As a result, our automated algorithm makes it easier and more trustworthy for practitioners. However, practitioners should be aware that the effectiveness of our method relies on the assumption of consistency between labeled and unlabeled data. When applying this method, it is crucial to carefully consider the characteristics of the unlabeled dataset. Significant differences in image quality, patient demographics, or acquisition parameters between labeled and unlabeled data might affect model performance.

Machine learning research on medical images is becoming popular as advanced imaging technologies and equipment in medicine become more and more available. Dental Cone-beam Computed Tomography (Dental CBCT), a frequently-used visualization tool for oral diagnosis, provides valuable three-dimensional information, whose development for automation of Dental CBCT analysis, on the other hand, is relatively preliminary. Generally, there are three important characteristics for analyzing Dental CBCT with noisy labels and limited labeled sample size, and availability of oral medicine knowledge. Based on those characteristics, we develop an image segmentation method for Dental CBCT by integrating domain knowledge into deep U-Net for the 3D segmentation. Finally, depending on whether the knowledge can be decomposed into each pixel, the knowledge constraints are classified into two types: separable and non-separable constraints. All knowledge constraints can be represented as a posterior regularization term and solved in different ways in accordance with related types. For separable constraints, the mean-field theory is employed to solve an optimization problem with the independence assumption about the distributions of output variables on each pixel. For non-separable constraints, we propose to combine the importance sampling based approach and the stochastic optimization algorithm. Finally, we propose to formulate the domain knowledge to the learning stage to improve the accuracy and efficiency of automation of Dental CBCT segmentation. Finally, we will apply the proposed methods into the real datasets collected and manually labeled by the doctors at the University of Pennsylvania.

Compared with the rapidly growing artificial intelligence (AI) research in other branches of healthcare, the pace of developing AI capacities in dental care is relatively slow. Dental care automation, especially the automated capability for dental cone beam computed tomography (CBCT) segmentation and lesion detection, is highly needed. CBCT is an important imaging modality that is experiencing ever-growing utilization in various dental specialties. However, little research has been done for segmenting different structures, restorative materials, and lesions using deep learning. This is due to multifold challenges such as content-rich oral cavity and significant within-label variation on each CBCT image as well as the inherent difficulty of obtaining many high-quality labeled images for training. On the other hand, oral-anatomical knowledge exists in dentistry, which shall be leveraged and integrated into the deep learning design. In this article, we propose a novel anatomically constrained Dense U-Net for integrating oral-anatomical knowledge with data-driven Dense U-Net. The proposed algorithm is formulated as a regularized or constrained optimization and solved using mean-field variational approximation to achieve computational efficiency. Mathematical encoding for transforming descriptive knowledge into a quantitative form is also proposed. Our experiment demonstrates that the proposed algorithm outperforms the standard Dense U-Net in both lesion detection accuracy and dice coefficient (DICE) indices in multilabel segmentation. Benefited from the integration with anatomical domain knowledge, our algorithm performs well with data from a small number of patients included in the training. Note to Practitioners-This article proposes a novel deep learning algorithm to enable the automated capability for cone beam computed tomography (CBCT) segmentation and lesion detection. Despite the growing adoption of CBCT in various dental specialties, such capability is currently lacking. The proposed work will provide tools to help reduce subjectivity and human errors, as well as streamline and expedite the clinical workflow. This will greatly facilitate dental care automation. Furthermore, due to the capacity of integrating oral-anatomical knowledge into the deep learning design, the proposed algorithm does not require many high-quality labeled images to train. The algorithm can provide good accuracy under limited training samples. This ability is highly desirable for practitioners by saving labor-intensive, costly labeling efforts, and enjoying the benefits provided by AI.