Diagnostic Accuracy of Artificial Intelligence Analysis Using Intraoral Photographs Versus Clinical Examination in White Spot Lesions Detection, Identification, And Scoring.
Diagnostic Accuracy of Artificial Intelligence Analysis Using Intraoral Photographs Versus Clinical Examination in White Spot Lesions Detection, Identification, And Scoring.
The goal of this observational study is to compare the diagnostic accuracy of Clinical examination as a standard for detection, identification and scoring of White Spot Lesions Versus Artificial intelligence analysis of intraoral photographs. The photographs are examined by experienced dental professionals to maintain diagnostic accuracy. Machine learning models YOLO and Mask-RCNN will analyze these images in three phases: pre-analytical, analytical and post-analytical. A dataset of 329 labelled photographs, annotated by experts, is used to train these models. Data augmentation methods enhance model performance, and accuracy is assessed against clinical examination results to confirm reliability.
The main question it aims to answer is:
- Is artificial intelligence analysis of intraoral photographs as accurate as clinical assessment in the detection, identification, and scoring of white spot lesions among adult Egyptian patients attending Cairo University Dental Hospital?
Dentists frequently encounter various dental hard tissue anomalies that present both diagnostic challenges and require careful treatment planning. A common example is white spot lesions or whitish discolorations of the teeth that can arise from multiple etiologies. These may be carious in nature, such as post-orthodontic incipient caries, or represent developmental defects like dental fluorosis or molar-incisor hypo-mineralization (MIH). Distinguishing between these conditions is essential for appropriate clinical management. The diagnosis of WSLs primarily relies on visual and photographic examination, which evaluates the morphology, size, color, and location of the lesions. Depth assessment is particularly critical, as it serves as a key determinant in selecting the most appropriate treatment approach. Detection and scoring of WSLs can be accomplished through clinical visual inspection alone or enhanced by adjunctive diagnostic technologies, including laser fluorescence, quantitative light-induced fluorescence (QLF), and electrical impedance spectroscopy.
The clinical characteristics of WSLs can vary considerably, making differential diagnosis challenging. While specialist clinicians demonstrate high validity and reliability in distinguishing between various white spot lesions, general dental practitioners exhibit lower diagnostic accuracy. Conventional diagnostic approaches for white spot lesions often lack precision and consistency. Visual assessment presents several inherent limitations, including the potential for misclassification due to overlapping clinical presentations among different etiologies. A systematic review and meta-analysis examining the efficacy of detection methods for incipient caries reported that photographic visual inspection achieved a sensitivity of only 67% and specificity of 79%, highlighting the need for more reliable diagnostic approaches.
Recent advances in computing power, data accessibility, and processing capabilities have accelerated the development of artificial intelligence (AI) applications, transforming contemporary healthcare research. Dentistry has similarly benefited from this technological evolution, with AI demonstrating considerable potential across various clinical applications. Machine learning (ML), a subfield of AI, represents a powerful approach for computer-aided diagnostic support, with algorithms that identify patterns within datasets during training and apply this knowledge to make predictions on new data. Emerging evidence indicates that these AI advances can improve diagnostic accuracy in caries detection, thereby supporting clinicians in making more precise and reliable assessments. Dental photography, captured using devices such as DSLR cameras or intraoral cameras, serves as a valuable tool for diagnosis and treatment planning. When combined with intelligent image analysis methods, AI can automate the identification and assessment of diagnostic data from photographs, facilitating standalone diagnostic procedures that reduce subjectivity and enhance clinical decision-making. Deep CNNs can detect and distinguish entities of similar but not identical appearance when trained on sufficiently large image datasets. Employing such models as part of an integrated image-analysis software solution would enable rapid classification of existing photographic library data and improve the accuracy and reliability of clinicians' decision-making in treatment planning or referral. Emerging evidence from studies using intraoral photographs and AI algorithms demonstrates promising detection rates, with AI exhibiting high sensitivity, specificity, precision, accuracy, and reliability in diagnostic performance. AI-driven tools are designed to serve as supportive aids for clinicians, strengthening diagnostic accuracy, streamlining workflows, improving cost-efficiency, and enhancing patient care rather than replacing clinical expertise. Simplifying the diagnostic process for white spot lesions is essential to enable their timely detection at an early stage, facilitating prompt intervention and improving preventive outcomes. AI models must be validated using local data from diverse clinical settings, particularly in lower-middle-income countries, with their performance assessed through sensitivity, specificity, and accuracy measures and benchmarked against conventional diagnostic approaches. Therefore, it is crucial to develop and validate white spot lesion detection and classification models using data from Egyptian patients attending Cairo University Hospital to ensure accurate and clinically relevant results, emphasizing the potential improvements in accuracy and reliability that AI can bring to dental diagnostics.
Inclusion Criteria:
Exclusion Criteria:
mohamed_gabr@dentistry.cu.edu.eg+201005660842