Approaches of Deep Learning for Diagnostic use on Cone Beam CT Imaging in Dentistry: A Review

Abstract

Deep learning (DL) has fundamentally transformed the analysis of cone beam computed tomography (CBCT) images in dentistry, a field where its importance is steadily growing for the detection, diagnosis, and surgical treatment of dental pathologies. DL, particularly based on convolutional neural networks (CNNs), is applied to improve the detection and classification of oro-maxillofacial pathologies. Studies show that using CBCT images analyzed with DL can provide superior diagnostic performance for detecting certain odontogenic cystic lesions compared to panoramic radiographs. Precise detection of these lesions is especially critical because they can lead to bone defects, requiring guided bone regeneration (GBR) or bone grafting interventions to fill the void. Beyond pathologies, DL plays a key role in assessing bone quality, particularly after regenerative procedures such as GBR. It notably enables accurate segmentation of critical bone structures like the mandibular canal using CNNs such as U-Net and its 3D variants, improving efficiency while reducing errors compared to semi-automatic methods. Moreover, DL is employed to classify bone density measured in Hounsfield units (HU) from CBCT images, using approaches based on 3D/2D CNNs, offering speed and precision. Evaluating bone density at the implant site is especially essential as it conditions the primary stability of the implant and directly impacts long-term treatment success. These advances mark a major evolution in radiological analysis, enabling a more objective and reliable assessment of bone density, crucial for implant planning. This survey paper presents an overview of several deep learning algorithms currently applied in diagnostic dentistry and implantology, highlighting their visualization of dental and bone structures, thus facilitating inpotential to optimize clinical workflows and improve patient outcomes.