Applications of deep learning in periodontal disease diagnosis and management: a systematic review and critical appraisal

Background: Periodontal disease remains a significant global health challenge, with traditional diagnostic methods often limited by subjectivity and time constraints. Recent advances in artificial intelligence (AI) and deep learning technologies have shown promise in various medical applications, potentially offering more objective and efficient approaches to periodontal diagnosis and management. This systematic review aimed to synthesize and critically evaluate the current body of evidence regarding the application of deep learning methodologies in the diagnosis and management of periodontal disease, with a focus on their potential to enhance clinical decision-making and patient outcomes. Methods: A comprehensive literature search was conducted across PubMed, Scopus, and IEEE Xplore databases, encompassing studies published between January 1, 2010, and March 31, 2024. The inclusion criteria encompassed studies that employed deep learning techniques for periodontal disease diagnosis, risk assessment, treatment planning, or prognosis prediction. The methodological quality of the included studies was assessed using the Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) tool. Results: The search yielded thirteen studies that met the predefined inclusion criteria. Deep learning models, particularly convolutional neural networks (CNNs) and hybrid architectures, demonstrated promising performance in the radiographic diagnosis and staging of periodontitis. Accuracies ranged from 73.0% to 98.6%, depending on the specific task and model architecture. Notably, a CNN-based model achieved 81.0% accuracy for premolars and 76.7% for molars in diagnosing periodontally compromised teeth (PCT). A hybrid framework combining deep learning for detection and conventional computer-aided diagnosis (CAD) processing for classification demonstrated high accuracy (dice coefficient of 0.93 for periodontal bone level detection) and excellent reliability (intraclass correlation coefficient of 0.91 with radiologists) in automatically diagnosing periodontal bone loss and staging periodontitis. Model performance exhibited variability contingent upon tooth position, with higher accuracy generally observed for premolars and canines compared to molars and incisors. The integration of clinical data with imaging analysis showed potential for improving diagnostic accuracy and treatment planning. However, challenges in generalizability across different populations and imaging centers were identified, highlighting the need for diverse training datasets and consideration of factors such as dental status in model development. Conclusions: Deep learning methodologies show significant promise in enhancing the diagnosis and management of periodontal disease. However, further research is needed to address challenges in generalizability, integrate diverse data types, and validate these models across various clinical settings to ensure their robustness and applicability in real-world scenarios.