TL;DR
This scoping review maps the landscape of Artificial Intelligence (AI) in dental image analysis for clinical decision-making, revealing a rapidly growing field primarily focused on diagnosis and detection. The findings highlight the widespread adoption of Convolutional Neural Networks (CNNs) and the potential for AI to enhance diagnostic accuracy and efficiency for dentists.
What Problem Does This Paper Address?
The integration of Artificial Intelligence (AI) into healthcare is accelerating, and dentistry is no exception. However, the sheer volume and diversity of AI applications in dental image analysis, particularly for clinical decision-making, can be overwhelming. This paper addresses the need for a comprehensive overview of this burgeoning field. It aims to map the current state of AI in dental image analysis, identify the most common techniques and applications, highlight emerging trends, and pinpoint research gaps. This understanding is crucial for researchers, clinicians, and developers to navigate the field effectively and guide future advancements.
Methods
This study employed a scoping review methodology, adhering to the PRISMA-ScR guidelines. The researchers conducted a systematic electronic literature search across two major databases: PubMed and Scopus. They screened titles and abstracts, followed by a full-text review based on predefined inclusion criteria. Data extraction focused on the application of AI in clinical decision-making, the specific dental fields involved, the imaging modalities used, and the AI algorithms employed. The review synthesized qualitative data from 276 included articles.
Datasets
The paper does not introduce new datasets. Instead, it analyzes the datasets used in the 276 reviewed studies. The sizes of these datasets varied significantly, ranging from as few as 10 images to as many as 83,676 images, with a total of 601,122 images analyzed across all included studies.
Key Insights and Ideas
- Explosive Growth: The field of AI in dental image analysis has seen a significant surge in publications, particularly since 2019, indicating its rapid evolution.
- Dominance of CNNs: Convolutional Neural Networks (CNNs) are the overwhelmingly preferred AI architecture, reflecting their effectiveness in image feature extraction, a critical aspect of dental diagnostics.
- Diagnosis is King: The primary application of AI in dentistry is for diagnosis, detection, and classification, aiming to improve accuracy and efficiency in identifying oral health issues.
- Radiographs Reign Supreme: Orthopantomograms (OPGs) and intraoral radiographs are the most commonly used imaging modalities, likely due to their widespread availability and familiarity among clinicians.
- Multidisciplinary Impact: AI applications span across various dental specialties, with oral medicine, oral surgery, and orthodontics being the most prominent areas of research.
Key Results
The review identified 276 primary studies, with the majority published from 2019 onwards. The total number of images analyzed across these studies was substantial, exceeding 600,000.
- AI Application: 84% of the articles discussed AI directly influencing clinical decision-making, with 84.4% focusing on diagnosis, detection, or classification.
- Dental Fields: Oral medicine (19%), Oral and Maxillofacial Surgery (OMFS) (16%), and orthodontics (12%) were the most frequently studied dental fields.
- Imaging Modalities: OPGs (33%) and intraoral radiographs (20%) were the most utilized imaging modalities.
- AI Algorithms: Convolutional Neural Networks (CNNs) were the most common AI models, used in 79% of the studies. ResNet was the most frequently utilized CNN architecture.
Limitations and Future Work
- Limited Clinical Integration: Despite promising results, the application of AI in clinical practice remains limited, with many studies representing proof-of-concept work.
- Data Challenges: Barriers to widespread adoption include the need for large, high-quality, and consistently annotated training datasets. Inconsistencies in data annotation and a lack of diverse population demographics in training data are significant concerns.
- Geographic Bias: The majority of studies originated from Asian countries, raising questions about the generalizability of findings to other populations, such as those in Australia.
- Ethical Considerations: Future work needs to address ethical issues surrounding patient data privacy, security, and the liability associated with AI-driven diagnoses.
- Standardization Needed: The paper calls for more systematic reviews and meta-analyses to critically appraise the quality and validity of AI applications. Standardization of training protocols and future clinical studies in diverse international settings are also recommended to address existing research gaps.
Summary updated