78 AUSTRALASIAN DENTIST FEATURE: ARTIFICIAL INTELLIGENCE In 2026, we are no longer asking if artificial intelligence will enter dental practice ; we are witnessing its impact every day. The more important question is whether we truly understand what type of AI we are adopting, and what it is actually doing. Across Australia, dental practices are implementing AI tools at multiple layers of workflow: reception automation, clinical notetaking, radiographic interpretation, CBCT analysis and patient communication. Administrative AI systems primarily improve efficiency. However, clinically significant AI systems now influence imaging interpretation, treatment planning and patient communication. This article compares Pearl, Diagnocat, CoTreat and Eyes of AI from a technical, governance and cost perspective. Core AI architectures Most dental AI systems rely on one or more of three architectures: Convolutional Neural Networks (CNNs) for pattern detection, Semantic Segmentation models for voxel-level anatomical mapping, and Large Language Models (LLMs) for generating patient-friendly explanations. Each architecture serves a different purpose and fails differently. AI in Dentistry 2026: Beyond reception bots: Understanding the architecture behind the tools By Dr Jenny Liu, Principal Dentist, Happy Dentistry, Melbourne Platform Core Architecture Primary Focus Approximate Cost Pearl CNN detection 2D radiographs ~AUD 450/month + setup Eyes of AI CNN detection 2D radiographs TBC Diagnocat CNN + Segmentation CBCT & planning ~AUD 5,000–7,500/year CoTreat CNN + LLM hybrid Communication & follow-up ~AUD 299–699/month Feature Pearl Eyes of AI Diagnocat CoTreat 2D Detection Yes Yes Yes Partial CBCT Segmentation No No Yes Limited Implant Planning No No Beta No Patient-Friendly Summaries Limited TBC Limited Yes Feature Comparison Pricing Overview By Dr Jenny Liu Diagnocat. Pano and Intraoral images taken in a single appointment are automatically combined into a convenient report. Diagnocat. Cephalometry (right side) Platform analysis Pearl (Second Opinion®) Pearl uses multiple deep learning vision models for 2D radiographic detection. The detection layer is vision-based, with builtin image normalization for robustness. Detections include confidence levels (high, medium, low), and clinicians can adjust sensitivity. Processing is cloud-based. Pearl integrates with Carestream, dentally (Vision), Exact and Planmeca Romexis. Diagnocat 2.0 Diagnocat operates on a unified deep learning pipeline combining CNN detection with voxel-level semantic segmentation for CBCT analysis. Detection and segmentation operate within the same model framework. Clinicians can view probability scores and CAT mask overlays. The orthodontic and airway modules are built on the same segmentation backbone. Processing is fully cloud-based. CoTreat CoTreat employs a hybrid AI architecture combining CNN-based image analysis with Large Language Models. The CNN layer detects visual patterns and treatment signals, while the LLM layer synthesizes structured findings into patient-friendly communication. It integrates with Dentally and emphasizes SMS-based engagement workflows. Eyes of AI Eyes of AI is an emerging Australian 2D radiographic detection platform built on CNN-based computer vision. Validation data and pricing are still evolving. Governance and bias considerations Clinicians should evaluate dataset diversity, bias mitigation processes, explainability features, cloud data processing and regulatory positioning before adopting AI platforms. AI remains adjunctive, and diagnostic responsibility remains with the clinician. Conclusion AI in dentistry is layered. CNNs recognize patterns. Segmentation models map anatomy. LLMs translate structured data into language. The key adoption question is not which AI is most advanced, but which clinical problem is being solved, and whether the chosen architecture is appropriate for that task. Disclosure: The author has participated in product demonstrations with the vendors discussed. No financial compensation was received for this article. u
RkJQdWJsaXNoZXIy MTc3NDk3Mw==