Feasibility and Validation of an Automated AI-Enabled Deep Ultraviolet (DUV) Cytology Platform for Point-of-Care Oral Cancer Triage

Overview

This research study evaluates an experimental, point-of-care tool that combines Deep Ultraviolet (DUV) light imaging with Artificial Intelligence (AI) to rapidly screen for oral cancer and pre-cancer (dysplasia). The main goal is to determine if this automated tool can accurately deliver immediate, same-visit triage results (categorizing cells as either non-dysplasia or dysplasia/cancer) without needing chemical stains, sample transport, or off-site laboratory processing. The following occurs during a singular clinic visit for participants: 1. Minimally Invasive Brush Test: A clinician gently rotates a soft brush over the mouth tissue to collect surface cells. 2. Stain-Free Imaging \& AI Analysis: The DUV microscope scans the unstained cells in minutes, and the AI analyzes cell features (such as nucleus size and shape) in under 10 seconds. 3. Standard Medical Care: Participants with suspicious spots receive a standard tissue biopsy to ensure a definitive, confirmed diagnosis regardless of the AI result. This is important given that the five-year survival rates for oral cancer exceed 80% when caught early but drop to under 20% if detected late. By delivering rapid results directly at the clinic, this technology aims to eliminate long waiting periods for lab results, accelerate specialist referrals, and improve access to early screening in community dental and medical clinics.

This research project establishes a new diagnostic paradigm by transforming oral brush cytology from a centralized, laboratory-dependent procedure into an automated, AI-assisted point-of-care (POC) triage system. Traditional brush cytology relies on chemical staining, physical sample shipping, and off-site expert review, which can introduce diagnostic turnaround delays. By combining label-free optical imaging with automated hardware and deep learning, this system aims to deliver immediate, same-visit triage of suspicious oral lesions directly within primary dental and outpatient clinics. Optical Physics \& Optomechanical Engineering: The core imaging system utilizes Deep Ultraviolet (DUV) microscopy operating in the 200-280 nm spectral range. At these wavelengths, cellular nucleic acids and proteins exhibit strong intrinsic light absorption, enabling submicron-resolution, high-contrast visualization of nuclear architecture without chemical dyes or fluorescent stains. Transformer-Based AI Framework (CellViT): Cellular analysis and classification are powered by an adapted CellViT deep learning framework, which replaces traditional object-detection architectures like YOLOv7.