Artificial intelligence (AI) has significantly impacted numerous industries, including health care, dentistry, and specifically prosthodontics. This review focuses on AI’s role in prosthodontics, detailing its use in diagnosis, design, and manufacturing. AI-driven systems analyze intraoral scans, improve prosthetic planning, and aid in robotic procedures. Emerging technologies, such as generative AI for prosthetic design and AI-driven material innovation, are discussed alongside the ethical and regulatory challenges facing broader adoption. The review highlights AI’s potential to transform prosthodontic workflows, facilitating more accurate, efficient, and personalized care, while also pointing to future developments such as real-time monitoring and enhanced collaboration platforms.
Key points
- •
Artificial intelligence and its subfields, machine and deep learning, are gaining traction in prosthodontics.
- •
Applications range from diagnostics (mainly on images) to treatment planning (including predictive prosthodontics) to conduct (using augmented and virtual reality).
- •
A range of limitations around bias, generalizability, explainability, accountability, and implementation remain.
- •
Regulators worldwide, along with policymakers and other stakeholders, strive to make dental AI more robust, reliable, and performant.
Introduction
Artificial intelligence (AI) is the simulation of human intelligence in machines designed to perform tasks such as learning, reasoning, problem-solving, and decision-making. The concept of AI originated in the mid-twentieth century with pioneers like Alan Turing, who proposed the idea of machines that could simulate any human task. Today, AI applications are widespread across numerous industries, like finance (AI algorithms detecting fraudulent activities, automating trading, and providing personalized financial advice), manufacturing (AI optimizing supply chain management, improving quality control, and predicting equipment maintenance needs), and health care.
In medicine, a range of applications for AI have been developed: 1. Diagnostics; mainly assisting image analysis, in which AI systems analyze radiographs or other imagery to identify abnormalities, often surpassing human experts in accuracy; 2. Predictive analytics; AI models predict disease progression and patient outcomes, supporting decision-making; 3. Personalized medicine; AI processes various biological or social data to tailor treatments to individual patients’ profiles in fields like oncology; and 4. AI accelerates drug discovery by predicting interactions between drugs and biological targets, reducing the time and cost required to bring new drugs to market.
In dentistry, applications mirror those in medicine. AI analyzes dental images to detect dental caries, apical lesions, and periodontal bone loss. Similarly, AI supports landmark detection in orthodontics, or segmentation of anatomic structures. AI is also used to support dental robotics, or for virtual reality (VR) applications in dental education and treatment support. Last, AI fosters self-dentistry and tele-dentistry, allowing patients to receive expert advice and routine check-ups without visiting a clinic. At the core of most AI systems in medicine and dentistry are machine learning (ML) and its subfield, deep learning (DL).
To address the evolving landscape of prosthodontics, this review on AI in prosthodontics is both timely and essential. The aims of the present study are as follows: (I) provide an updated overview of AI methodologies relevant to prosthodontics, including ML, DL, and computer vision; (II) examine key applications in prosthodontic diagnostics, treatment planning, and patient management; and (III) discuss ethical, regulatory, and technical barriers that may limit broader adoption of AI in prosthodontics. This review also covers emerging trends, such as the use of generative AI in prosthetic design and AI-driven material innovation, with the goal of facilitating informed decision-making among practitioners and researchers as they navigate this rapidly transforming field.
Machine learning
ML is a branch of AI that focuses on developing algorithms that enable computers to learn from and make decisions based on data. Unlike traditional programming, where explicit instructions dictate operations, ML identifies patterns and relationships within data to make predictions or decisions. The core concept of ML involves training models on a dataset, allowing them to improve their performance over time without being explicitly programmed for specific tasks. A range of steps are involved in such training ( Fig. 1 ).
- 1.
Data collection and preparation: The first step in ML involves gathering and preprocessing data. These data can come from various sources, including databases (eg, electronic health records), sensors (eg, imagery), or user inputs (eg, prospectively collected data in a clinical trial). Preprocessing involves cleaning the data, handling missing values, and transforming it into a format suitable for analysis.
- 2.
Choosing a model: Depending on the problem, different types of ML models are selected. Common models include linear regression for predicting continuous values, classification algorithms like decision trees or support vector machines for categorizing data, and clustering algorithms like k-means for grouping similar data points. For analyzing images and speech, DL models (see below) are commonly used.
- 3.
Training the model: The chosen model is usually trained using a labeled dataset, which contains input-output pairs. During training, the model adjusts its internal parameters to minimize the difference between its predictions and the actual outcomes. This process involves using algorithms like gradient descent, which iteratively updates the model parameters to reduce error. Notably, training can also be done using so-called unsupervised learning approaches, where no labeled data are used, that is, ML identifies patterns in data without being specifically tasked with what to learn.
- 4.
Evaluation: After training, the model is evaluated using a separate test dataset. Performance metrics such as accuracy, precision, recall, and F1 score are used to assess how well the model generalizes to new, unseen data. Cross-validation techniques can further be used to display if the training led to robust models.
- 5.
Deployment and prediction: Once tested, the model can be deployed to make predictions on new data. This process involves feeding the model fresh inputs and using its learned parameters to generate outputs.
Deep learning: a subfield of machine learning
DL is a specialized subfield of ML that focuses on neural networks with many layers, known as deep neural networks. These networks are inspired by the human brain’s structure and function, allowing them to model complex patterns and relationships within data.
DL utilizes artificial neural networks, which consist of layers of interconnected nodes (neurons). Each neuron processes input data and passes the result to subsequent layers. The complexity and depth of these networks enable them to capture intricate patterns in data. A typical deep neural network has an input layer, multiple hidden layers, and an output layer. The input layer receives raw data, which is then transformed and processed through hidden layers. Each hidden layer applies various transformations and learns abstract representations of the data. The output layer generates the final prediction.
Training deep neural networks involves a process called backpropagation. During training, the network makes a prediction and calculates the error by comparing it to the actual outcome. This error is then propagated backward through the network, adjusting the weights of the connections between neurons to minimize the error.
Neurons in a neural network apply activation functions to introduce nonlinearity into the model, allowing it to learn complex patterns. Common activation functions include Rectified Linear Unit (ReLU), sigmoid, and tanh.
Specific types of DL networks are designed for different tasks. Convolutional neural networks are highly effective for image and video analysis due to their ability to capture spatial hierarchies. Recurrent neural networks, including long short-term memory networks, are well-suited for sequential data like time series and natural language processing.
At present, particular attention lies on generative AI, which refers to a class of AI models designed to create new content—such as text, images, music, or even code—by learning from datasets. These models, particularly those based on DL architectures like generative adversarial networks or transformers, are trained on extensive data to understand patterns, structures, and relationships within the data. Once trained, they can generate new content that mimics the style, tone, or structure of the original data, often with remarkable creativity and coherence. For instance, a generative AI model trained in text can produce human-like sentences, stories, or essays, while once trained on images can create realistic or abstract visuals.
Deep learning for image analysis
In image analysis, AI fulfills a range of tasks, specifically classification, object detection, regression, various forms of segmentation, and landmark detection ( Fig. 2 A–E).
- 1.
Classification refers to the process of categorizing data or images into predefined labels. In image analysis, each image is assigned a single label (eg, identifying an image as containing a cat or a dog—or dental radiograph showing a carious lesion or periodontal bone loss).
- 2.
Object detection goes beyond classification by locating and identifying multiple objects within an image, usually using bounding boxes identifying both the object class and its location.
- 3.
Regression involves predicting continuous values rather than discrete labels. In image analysis, regression models can be used to predict properties like object size, distance, or image quality, commonly applied in medical imaging for measuring tissue volumes or estimating disease progression.
- 4.
Segmentation is a more detailed analysis task, where the goal is to partition an image into meaningful regions. Semantic (or pixelwise) segmentation labels each pixel in an image according to the class it belongs to, without differentiating between individual instances (eg, identifying all pixels belonging to cars, or highlighting all teeth in a panoramic radiograph without being able to separate individual teeth). In medical and dental applications, instance segmentation is more common. It assigns unique labels to individual objects in an image, also those of the same class, allowing to identify each instance separately (eg, distinguishing between 2 overlapping cars—or showing multiple carious lesions on a bitewing radiograph and allowing to separate each of them).
- 5.
Landmark detection identifies specific points (landmarks) in an object; in dentistry, a typical landmarking task is cephalometric analysis. Landmarking can also be used to allow other tasks, like measurements. To measure periodontal bone loss, landmarks like the cementum-enamel junction or the bone line are often employed in the first step, before measuring any distances or expressing bone loss in %.
