Bridging Sight and Science: Employing Vision-Language Models (VLMs) as Next-Generation Visual Assistants in Digital Dentistry

ABSTRACT

1. Background and Purpose

Recent breakthroughs in multimodal Artificial Intelligence have introduced Vision-Language Models (VLMs)—such as GPT-4o, Claude 3.5 Sonnet, and open-source medical AI models—which can simultaneously process clinical text and interpret complex medical images. While traditional AI requires deep programming expertise to analyze radiographs or clinical photos, VLMs allow clinicians to query visual data using natural language.
The purpose of this research initiative is to evaluate how effectively these state-of-the-art general and specialized VLMs can assist in diagnostic reasoning, surgical planning, and post-operative monitoring within digital dentistry and oral maxillofacial surgery (OMFS). Undergraduate researchers will curate distinct clinical datasets (which may include panoramic radiographs, 3D surface scans, or intraoral clinical photographs), design advanced prompt-engineering protocols, and benchmark AI visual diagnostic accuracy against human clinical experts. This project aims to yield a peer-reviewed publication in a dental science journal while equipping students with cutting-edge skills in dental informatics and digital health.

2. Research Questions

Students joining this project will investigate core questions tailored to their specific clinical data focus, including:
1) To what degree of accuracy can a VLM correctly identify, classify, and report critical anatomical landmarks, pathologies, or surgical metrics directly from dental imaging compared to an expert oral surgeon?
2) Which prompt-engineering strategies (e.g., zero-shot, few-shot, or Chain-of-Thought reasoning) yield the highest reliability and lowest rates of visual “hallucination” when interpreting dental visual data?
3) How effectively can a VLM translate its visual findings into both structured surgical checklists for residents and empathetic, easily understood risk-communication summaries for patients?

SKILLS & EXPERIENCE REQUIRED

1) Highly motivated undergraduate dental, engineering, or computer science students.

2) For dental students, no prior coding or computer science background is required. Strong organizational skills, attention to clinical detail, and familiarity with basic dental anatomy/radiology are highly valued.