How Dental Chains Are Using AI to Catch What Dentists Miss

1. The Problem: What Gets Missed on X-Rays

A dentist reviews hundreds of X-rays per week. Each bitewing or periapical image contains dozens of potential findings: early caries, marginal bone loss, periapical pathology, cracked roots, failing restorations. Published research consistently shows that human-only radiographic diagnosis misses a meaningful percentage of pathology. Studies in the Journal of Dental Research report that dentists miss between 25% and 40% of interproximal caries on bitewing radiographs, depending on experience level and workload.

This is not a criticism of dentists. It is a recognition that radiographic interpretation is cognitively demanding, repetitive work performed under time pressure. A busy practice might see 30 to 40 patients per day. Every X-ray requires focused attention. Fatigue, interruptions, and time constraints all contribute to missed findings. The consequences range from delayed treatment (which becomes more invasive and expensive) to patient trust erosion when a cavity "suddenly appears" at the next visit.

AI diagnostic tools address this gap by providing a consistent second opinion on every single image, every single time, without fatigue or distraction. Three companies have emerged as market leaders in this space: Pearl, Overjet, and Diagnocat. All three hold FDA clearances. All three are deployed in production across thousands of practices. And all three have published performance data that shows measurable improvement over human-only diagnosis.

2. The Numbers: AI Detection vs. Human-Only Diagnosis

The performance gap between AI-assisted and human-only radiographic diagnosis is well documented. Pearl reports that practices using their Second Opinion product detect 37% more pathology compared to baseline diagnostics without AI assistance. This is not a laboratory result. It comes from production data across their network of over 23,000 practices in more than 120 countries.

Diagnocat publishes condition-specific accuracy rates based on their analysis of both 2D and 3D imaging. Their system detects over 60 tooth conditions in 3D CBCT images and 40+ conditions in 2D radiographs. The published accuracy figures are notable:

• Impaction detection: 100% accuracy
• Filling detection: 100% accuracy
• Endo-treated tooth identification: 96% accuracy
• Periodontal bone loss (horizontal, mild): 87% accuracy
• Missed canal detection: 74% accuracy

That last number deserves attention. A missed canal is one of the most consequential diagnostic errors in endodontics, often leading to persistent infection, retreatment, or extraction. At 74% detection accuracy for a condition that is notoriously difficult to identify on radiographs, AI provides a meaningful safety net that did not previously exist in routine clinical workflow.

Overjet, which has raised more than $130 million in funding, takes a different measurement approach. Their Vision AI system is FDA-cleared for comprehensive diagnosis and patient education. Rather than publishing aggregate accuracy percentages, Overjet emphasizes case acceptance improvements: when patients see AI-annotated images showing exactly where disease exists, they are more likely to accept treatment recommendations.

3. Case Studies: Who Is Deploying and What They Found

The following case studies are drawn from published data by the companies involved. These are not pilot programs or research projects. They are production deployments in real clinical environments.

Large DSO Deployments

Single-Practice and Small-Group Results

4. The AI Tools: Pearl, Overjet, Diagnocat Compared

5. How Diagnostic AI Actually Works in a Dental Chair

Understanding the clinical workflow removes much of the uncertainty about adopting these tools. Here is what a typical AI-assisted appointment looks like in practice.

Step 1: Image Capture (No Change)

The dental assistant takes radiographs using the existing sensor and imaging equipment. Pearl, Overjet, and Diagnocat all integrate with existing hardware. There is no need to replace sensors, X-ray units, or imaging software. The AI connects to your practice management system or imaging software through an integration layer.

Step 2: Automatic Analysis (2 to 5 Seconds)

As soon as the image appears in the software, the AI analyzes it. Pearl overlays colored annotations highlighting detected pathology. Overjet outlines disease and anatomy with visual markers. Diagnocat generates a structured report listing all detected conditions with confidence scores. This happens in seconds, before the dentist even looks at the image.

Step 3: Clinician Review (Enhanced, Not Replaced)

The dentist reviews the image with AI annotations visible. They can agree with the AI findings, dismiss them, or add their own observations. The AI serves as a checklist and a second opinion. If it highlights an area the dentist might have overlooked, that is a caught miss. If the dentist identifies something the AI did not flag, they document it normally. The clinical judgment remains entirely with the dentist.

Step 4: Patient Communication (Transformed)

This is where case acceptance improvements come from. Instead of pointing at a grainy X-ray and saying "there is a small cavity here," the dentist can show the patient an annotated image where the AI has clearly marked the area with a visual overlay. The patient sees what the dentist sees. Multiple case studies report that this single change drives 30% to 48% improvements in treatment acceptance.

Step 5: Documentation and Billing (Streamlined)

AI findings can pre-populate clinical notes and support insurance claim documentation. Some tools (Pearl's Precheck, Overjet's ReviewPASS) directly assist with insurance verification and pre-authorization. This reduces administrative burden and can accelerate reimbursement timelines.

6. European Context: MDR, GDPR, and CE Marking

European dental practices face a regulatory environment that differs significantly from the United States. Understanding these requirements is essential before evaluating any AI diagnostic tool.

EU Medical Device Regulation (MDR)

Under the EU Medical Device Regulation (2017/745), AI software that performs diagnostic functions is classified as a medical device. This means any AI tool that analyzes dental radiographs and provides diagnostic suggestions must carry a CE mark under MDR, not just the older MDD (Medical Devices Directive). The MDR is more stringent, requiring clinical evidence, post-market surveillance, and a designated Notified Body review for higher-risk classifications.

For dental AI, the classification typically falls under Class IIa (diagnostic decision support) or Class IIb (if the AI directly influences treatment decisions). Key implications for your practice:

• Verify CE marking: Ask vendors to confirm MDR compliance, not just MDD. Some tools may still operate under MDD transitional provisions, which expire in phases through 2028
• Check the Notified Body: The CE certificate should reference a recognized Notified Body. You can verify this on the EU EUDAMED database
• Clinical evidence: MDR requires clinical evaluation demonstrating that the AI performs as claimed in clinical settings. Ask vendors for their clinical evaluation report or published validation studies
• Post-market surveillance: Vendors must actively monitor performance after deployment. This benefits you because it means ongoing validation and improvement of the AI's accuracy

Pearl holds 26+ international regulatory clearances including MDSAP approval, which facilitates multi-market compliance. Diagnocat received FDA clearance in October 2025 and has Health Canada Class II approval. European practices should confirm current CE/MDR status directly with each vendor, as the regulatory landscape is evolving rapidly.

GDPR Compliance for Patient Data

Dental radiographs are classified as health data under GDPR, which places them in the "special categories of personal data" requiring the highest level of protection. Any AI tool processing patient X-rays must comply with these requirements:

• Data processing agreement (DPA): Your AI vendor must sign a DPA specifying how patient imaging data is processed, stored, and protected. This is mandatory, not optional
• Lawful basis: Processing health data for treatment purposes falls under GDPR Article 9(2)(h), but you must ensure your patient consent forms cover AI-assisted analysis. Update your privacy notices to mention that X-rays may be analyzed by AI diagnostic software
• Data residency: Understand where patient images are sent for AI analysis. EU-based processing is strongly preferred. Some vendors process images locally (on-premise or edge computing), which eliminates data transfer concerns entirely
• Data retention: AI vendors should not retain patient images beyond the analysis period unless you explicitly authorize it for model improvement. Clarify data retention policies before signing
• Patient rights: Under GDPR Article 22, patients have the right not to be subject to decisions based solely on automated processing. Since dental AI provides decision support to a licensed professional rather than making autonomous treatment decisions, this typically does not create a compliance issue. Document your workflow to demonstrate that the dentist, not the AI, makes all clinical decisions

Data Processing Location

Where patient images are processed matters. Three models exist in the current market:

• Cloud processing (EU-hosted): Images are sent to EU-based servers for analysis. Compliant with GDPR provided proper DPAs are in place. Latency is typically 2 to 5 seconds
• Cloud processing (US-hosted): Requires EU-US Data Privacy Framework compliance or Standard Contractual Clauses (SCCs). More complex from a compliance standpoint. Some European dental practices and regulatory bodies may object
• Edge/local processing: Images are analyzed on local hardware within your practice. No data leaves your premises. The strongest privacy position but may limit access to cloud-based features and model updates

When evaluating vendors, ask explicitly: "Where are patient images processed, and do you offer EU-hosted or on-premise processing?" This question alone will reveal how seriously a vendor takes European regulatory requirements.

7. ROI: What a Single-Practice Clinic Can Expect

The economics of dental AI diagnostics are unusually straightforward because the revenue impact is directly measurable. Here is a realistic projection based on published data from Pearl and the case studies above.

Revenue Impact

Pearl reports an average $30,000 monthly production boost per practice. Even discounting this figure significantly for a smaller European practice, the math is compelling. Consider a practice that currently has a 45% treatment acceptance rate:

• A 30% improvement in acceptance (the conservative end of published data) raises acceptance from 45% to 58.5%
• For a practice presenting EUR 50,000 in treatment per month, that 13.5 percentage point increase translates to EUR 6,750 in additional monthly production
• The 37% improvement in disease detection means additional findings that were previously missed, generating treatment plans that simply did not exist before

Time Savings

Pearl reports 20+ hours saved weekly per office. This time comes from several sources: faster radiographic interpretation, reduced need for second opinions from specialists (the AI provides an immediate second opinion), streamlined documentation, and more efficient patient communication. For a practice where a dentist's time is valued at EUR 150 to 250 per hour, 20 hours per week represents EUR 3,000 to 5,000 in recovered productive capacity weekly.

Cost Estimate

Dental AI tools typically price on a per-location monthly subscription model. Based on publicly available information and industry reports, expect:

• Single practice: EUR 300 to 800 per month depending on features and vendor
• Multi-location group (5+ sites): Volume discounts typically reduce per-location cost by 20% to 40%
• Implementation: Integration with existing imaging systems usually takes 1 to 3 days. Most vendors provide onboarding support at no additional cost
• Training: Clinical teams typically become proficient within 1 to 2 weeks. The tools are designed to integrate into existing workflows, not replace them

Payback Period

At a conservative estimate of EUR 500 per month for the tool and even a modest EUR 3,000 per month in additional production from improved case acceptance plus time savings, the payback period is effectively immediate. The first month of use covers the annual subscription cost many times over. This is why adoption is accelerating: the ROI case does not require a leap of faith.

8. Next Steps for European Dental Practices

AI-assisted dental diagnostics is not experimental technology. It is deployed across more than 23,000 practices globally. The tools are FDA-cleared. The results are published and repeatable. European practices now have reference deployments like mydentist in the UK to validate both clinical and regulatory feasibility.

For a single dental practice or small group, the implementation path is straightforward:

1. Week 1: Request demos from Pearl, Overjet, and Diagnocat. Evaluate based on your imaging equipment, practice management system, and primary clinical focus (general dentistry, endodontics, implants)
2. Week 2 to 3: Verify regulatory compliance. Confirm CE/MDR status, GDPR compliance, data processing location, and DPA availability. Ask about the mydentist UK deployment as a compliance reference
3. Week 3 to 4: Run a trial. Most vendors offer evaluation periods. Test with your actual images, your actual patients, your actual workflow. Measure treatment acceptance rates before and during the trial
4. Month 2: Deploy and measure. Track case acceptance, additional findings per 100 images, time saved on documentation, and patient feedback. These metrics will tell you within 30 days whether the tool delivers value

The practices that move on this now will have a measurable clinical and competitive advantage over those that wait. Every X-ray analyzed without AI assistance is an X-ray where pathology might be missed, a treatment plan might not be presented, and a patient might walk out without the care they need.

Related Resources

Sources

• Pearl (2026). Product documentation and published metrics: 37% disease detection improvement, 30% treatment acceptance increase, 23,000+ practices, 120+ countries. hellopearl.com
• Pearl case studies (2026). Meloria Dental (UK): 48% treatment acceptance increase. Clyde Munro Dentistry (Scotland): 2+ year deployment. Onsite Dental, Rand Center for Dentistry, Blue Court Dental, P4D. hellopearl.com/case-studies-guides
• Overjet (2026). Company data: $130M+ funding, FDA-cleared Vision AI and CBCT 3D. Clients include NADG, mydentist UK, Dental Care Alliance, Jefferson Dental. overjet.com
• Overjet blog (2025-2026). NADG: 216-location Overjet Voice deployment (Dec 2025). mydentist UK: largest dental AI rollout in British dentistry (Feb 2026). Swish Dental: 21 Texas locations for insurance verification (Jan 2026). overjet.com/blog
• Diagnocat (2026). Published accuracy rates: impaction 100%, filling 100%, endo-treated tooth 96%, periodontal bone loss 87%, missed canal 74%. FDA cleared October 2025. 60+ conditions in 3D, 40+ in 2D. 50+ countries. diagnocat.com
• European Commission (2017). Regulation (EU) 2017/745 on Medical Devices (MDR). Classification rules for AI diagnostic software.
• European Commission (2016). General Data Protection Regulation (GDPR). Article 9 (special categories), Article 22 (automated decision-making), Article 28 (data processors).