Pearl is a dental artificial intelligence company whose flagship product, called Second Opinion, analyzes dental X-rays in real time and highlights potential problems a dentist might otherwise overlook. The software uses deep learning to scan radiographs for conditions like cavities, bone loss, and periapical lesions, then overlays color-coded annotations directly on the image so the clinician can review them chairside. It received FDA clearance as a diagnostic aid, making it one of the first commercially available AI tools specifically designed for everyday dental practice rather than specialty research settings. The technology sits at the intersection of computer vision and clinical dentistry, and its growing adoption raises practical questions about accuracy, workflow, liability, and what it means for patients.
What Pearl’s Software Actually Does
At its core, Pearl’s Second Opinion product ingests a standard dental radiograph, whether it is a bitewing, periapical, or panoramic image, and runs it through a trained neural network that has learned to recognize visual patterns associated with dental pathology. Within seconds, the software returns the same image with colored outlines around areas it flags as suspicious. Each flag comes with a label identifying the type of finding and, in many cases, a confidence score indicating how certain the algorithm is about the detection.
The range of conditions the software targets is broad for a single tool. Validation studies of deep learning algorithms built for this purpose have described detection of caries, apical lesions, root canal treatment defects, marginal defects at crown restorations, periodontal bone loss, and calculus on intraoral radiographs.1arXiv. Statistical validation of a deep learning algorithm for dental anomaly detection in intraoral radiographs using paired data Separate research evaluating AI software on periapical radiographs has assessed its accuracy across a similar set of parameters including marginal bone loss, periapical radiolucency, caries, type of restorative material, type of crown retainer material, and detection of open crown margins.2PubMed Central. Assessment of the Diagnostic Accuracy of Artificial Intelligence Software in Identifying Common Periodontal and Restorative Dental Conditions In practical terms, this means the software is trying to catch not just the obvious cavity but also subtler issues like early bone loss around teeth or a gap under an existing crown.
How Deep Learning Reads an X-Ray
Pearl’s technology belongs to the deep learning branch of artificial intelligence, which differs from older image-processing approaches in that it learns directly from large collections of labeled images rather than following hand-written rules about what a cavity looks like. Research into dental X-ray analysis has historically fallen into three broad categories: traditional image processing, classical machine learning, and deep learning, with deep learning now representing the dominant approach because of its ability to handle variation in image quality, tooth anatomy, and pathology presentation.3PubMed Central. Descriptive analysis of dental X-ray images using various practical methods: A review
The basic idea is that the system is shown millions of dental X-rays where experts have already drawn boundaries around each problem area and labeled what the problem is. Over many rounds of training, the algorithm adjusts its internal parameters until it can reliably reproduce those expert labels on images it has never seen before. The architecture most commonly used for this kind of work in medical imaging is a type of convolutional neural network, which excels at recognizing spatial patterns like edges, textures, and shapes within an image. Pearl trains its models on what it describes as one of the largest proprietary datasets of annotated dental radiographs in the industry, though the exact size of the dataset is not publicly audited by an independent body.
Why Annotation Quality Matters So Much
The accuracy of any AI system like Pearl’s depends heavily on the quality of the labels it was trained on, and this is where dental AI gets complicated. A review of publicly available dental image datasets found that while about three-quarters contained annotations, the methods used to create those labels were often unclear and inconsistent.4PubMed Central. Publicly Available Dental Image Datasets for Artificial Intelligence If one dentist draws the boundary of a cavity slightly differently than another, or if two experts disagree about whether a shadow on a radiograph is an early lesion or an artifact, the training data carries that ambiguity forward into the model’s predictions.
Research has directly tested how different annotation strategies affect AI accuracy for caries detection. In one study, models were trained using labels from individual dentists, aggregated strategies like majority voting and consensus meetings, and even micro-CT-based ground truth, which uses three-dimensional imaging to verify what is actually there.5PubMed Central. Accuracy of deep learning-based AI models for early caries lesion detection: the influence of annotation quality and reference choice The takeaway is that how you define “correct” during training has real consequences for what the finished product flags as a problem. Pearl addresses this by using panels of specialists to annotate its training data, though the company’s internal annotation protocols are proprietary.
How Accurate Is It Compared to a Dentist?
This is the question most patients and practitioners care about, and the evidence is encouraging but comes with caveats. AI-supported diagnostic studies have reported accuracy figures that are quite strong, with the best-performing systems showing diagnostic accuracy well over 90% and sensitivity and specificity values in the range of 80 to 95%.6PubMed Central. Emerging trends in the early diagnosis of dental caries: a scoping review of artificial intelligence, digital diagnostics, and teledentistry Those numbers are impressive, but they vary depending on the condition being detected, the quality of the radiograph, and the specific dataset used for testing.
A more nuanced picture comes from studies comparing AI head-to-head against practitioners at different experience levels. One study evaluating periodontal parameters found that AI and senior specialists consistently achieved the highest performance in detecting attachment loss and alveolar bone loss. The AI scored a mean of 6.12 in identifying teeth with attachment loss, compared to 5.43 for senior specialists, 4.58 for other specialists, and 3.65 for general dentists.7PubMed Central. AI Efficiency in Dentistry: Comparing Artificial Intelligence Systems with Human Practitioners in Assessing Several Periodontal Parameters The pattern here is telling: AI’s consistency in identifying subtle conditions was comparable to that of senior specialists, while general dentists showed greater variability. For a solo practitioner without a periodontist down the hall, that kind of second set of eyes has obvious value.
That said, accuracy in a controlled validation study is not the same as accuracy in a busy clinic with suboptimal X-ray technique, unusual anatomy, or artifacts from existing dental work. The real-world performance gap is something the field is still working to quantify across different practice settings.
FDA Clearance and What It Means
Pearl’s Second Opinion received FDA clearance through the 510(k) pathway, which is the regulatory route for medical devices that are substantially equivalent to something already on the market. This is not the same as FDA “approval” in the way that term is used for drugs; it means the agency reviewed the company’s evidence and agreed the device is safe and effective enough for its intended use as a diagnostic aid. A narrative review of FDA-authorized dental AI imaging devices searched regulatory databases including 510(k), De Novo, and PMA pathways through mid-2025 to catalog the growing landscape of cleared products.8PubMed Central. FDA-Approved AI Solutions in Dental Imaging: A Narrative Review of Applications, Evidence, and Outlook
An important distinction: FDA clearance for Pearl’s software means it is authorized as a clinical decision-support tool, not as an autonomous diagnostic device that replaces the dentist’s judgment. The dentist still reviews the flagged image, decides whether the AI’s findings are clinically meaningful, and makes the treatment recommendation. The software is designed to reduce the chance of something being missed, not to generate a diagnosis on its own.
How It Fits Into a Dental Visit
From the patient’s perspective, the experience is mostly invisible. Your X-rays are taken the same way they always were. The difference is that the images are automatically routed through Pearl’s cloud-based platform before or as the dentist reviews them. By the time the dentist pulls up your radiographs on their screen, the AI’s annotations are already overlaid. The dentist can toggle them on or off, click on individual findings for more detail, and use the markups as a starting point for their own assessment.
For the practice, integration typically happens through the existing imaging software. Pearl connects to most major dental imaging and practice management systems, so adopting it does not require replacing hardware or changing the way images are captured. The processing happens in the cloud, which means the practice needs a stable internet connection, and the images are transmitted to Pearl’s servers for analysis. That cloud dependency raises the data privacy question that recurs across all health-tech AI.
Patient Trust and Privacy Concerns
Patients are generally open to the idea of AI-assisted dental diagnosis, but their enthusiasm comes with reservations. A survey of over 500 participants examining attitudes toward AI in dental diagnosis found that about 60% trusted AI dental diagnosis, while 55% expressed concerns about data privacy and accuracy.9PubMed Central. Patients’ Acceptance and Intentions on Using Artificial Intelligence in Dental Diagnosis: Insights From Unified Theory of Acceptance and Use of Technology 2 Model Social influence, meaning whether friends, family, or the dentist endorsed the technology, was the strongest predictor of a patient’s willingness to accept AI in their care. That finding suggests the dentist’s framing matters a lot: patients who hear their dentist explain and endorse the tool are much more likely to feel comfortable with it than patients who discover it on their own.
The privacy concern is not trivial. When your dental X-rays are sent to a cloud server for analysis, they become subject to whatever data handling and storage policies the AI vendor maintains. Pearl states it is HIPAA-compliant, which means it follows the same privacy framework that governs your dental records generally. But HIPAA compliance is a floor, not a ceiling, and the question of whether de-identified dental images might be used to train future models or shared with third parties is one that the broader dental AI industry has not fully resolved in a way patients can easily verify.
The False Positive Problem
One underappreciated risk with any diagnostic aid, AI-powered or otherwise, is that improving detection does not always improve outcomes. Catching more problems sounds like an unambiguous good, but when a tool flags something that turns out not to be a real problem, the result is unnecessary treatment. Research on radiographic caries detection, even without AI, has shown that the use of radiographs in diagnostic strategies for caries detection in children brought more harms than benefits in some cases, due to false positives, overdiagnosis, and lead-time bias.10PubMed Central. Consequences of radiographic assessment in the diagnosis of caries lesions in deciduous molars: a secondary analysis of a randomized clinical trial
This is not a problem unique to AI; human dentists disagree with each other about borderline findings all the time. But an AI tool that flags early-stage shadows aggressively could shift the balance toward more interventions if the dentist defers to the technology rather than using it as one input among many. The framing of AI as a “second opinion” is deliberate for this reason: the software is supposed to surface possibilities, not dictate treatment plans. Whether every dentist uses it that way in practice is another question.
Legal Liability When AI Gets Involved
A persistent question in dental AI adoption is who bears responsibility when the software contributes to a wrong call. If Pearl flags a shadow as a cavity and the dentist drills into a healthy tooth, is the dentist liable? What if Pearl fails to flag something and the dentist misses it because they were relying on the AI to catch it? The integration of AI into clinical dentistry raises challenges around safeguarding patient autonomy and data privacy, addressing algorithmic bias, ensuring transparency, and maintaining professional accountability.11International Journal of Dentistry. Artificial Intelligence–Driven Dentistry: A Systematic Review of Ethical and Legal Challenges
Under current legal frameworks in most jurisdictions, the dentist retains responsibility for clinical decisions regardless of what tools they use. The AI is classified as a decision-support device, which means it is legally analogous to any other diagnostic instrument: the clinician interprets the output and owns the decision. But the practical reality is murkier. If AI-assisted diagnosis becomes the standard of care, a dentist who chooses not to use it and misses something the software would have caught could face malpractice claims for falling below the standard. The technology is new enough that these questions are still being worked out in regulatory and legal circles.
Public Health Triage and Underserved Populations
Beyond private practice, dental AI tools have potential in public health settings where specialist access is limited. A pilot study tested an AI-driven triage tool in Australian public oral healthcare and found an overall triage decision accuracy of 98.3%, with perfect sensitivity and a specificity of 97.5%.12Value in Health. Face Validation of an Artificial Intelligence Driven Tool for Clinical Triaging in Australian Public Oral Healthcare: A Pilot Study That study also revealed that the population using public dental services had higher rates of dental caries and moderate-to-severe periodontitis than national averages, underscoring the point that the people who stand to benefit most from AI-assisted screening are often those with the least access to specialist care.
Pearl itself has explored partnerships aimed at underserved communities, though most of its commercial footprint remains in private dental practices and dental service organizations. The broader promise is that cloud-based AI could allow a general dentist in a rural area to get analysis comparable to what a specialist in a major city would provide, flattening some of the geographic inequality in dental care quality. Whether that promise materializes at scale depends on cost, internet infrastructure, and whether insurance reimbursement models adapt to cover AI-assisted diagnostics.
Where the Technology Is Heading
Pearl currently works with two-dimensional radiographs, but the field is moving toward three-dimensional imaging. AI-enhanced analysis of cone-beam computed tomography scans has shown promise in improving diagnostic speed and accuracy for periodontal bone defects, and in some studies has outperformed conventional interpretation.13International Journal of Research and Review. A Comprehensive Review on Artificial Intelligence Assisted CBCT Analysis for 3D Alveolar Bone Morphometry and Periodontal Diagnosis Future development in dental AI is expected to focus on multimodal data fusion, meaning combining imaging data with clinical measurements and patient history, as well as building explainable AI systems that can show the dentist why the algorithm flagged something, not just that it did.
The broader trajectory for AI in dentistry points toward more personalized and preventive care. Rather than just catching problems after they develop, future systems may predict which teeth are at highest risk for disease progression and recommend targeted interventions before damage occurs.14PubMed. Artificial intelligence models for clinical usage in dentistry with a focus on dentomaxillofacial CBCT: a systematic review Pearl has signaled interest in expanding beyond detection into treatment planning assistance, though the regulatory bar for tools that recommend specific treatments is considerably higher than for tools that simply flag findings on an image. For now, the technology is best understood as a highly consistent second reader, one that never gets tired, never rushes through the last patient of the day, and processes every pixel of every image with the same attention. What it cannot do is replace the clinical judgment, patient relationship, and tactile examination that make dentistry a human profession.