CDC Foundation and Georgia Tech Deploy Vision-Language Models to Track "Smart" Vapes

In a consequential advancement at the intersection of public health and artificial intelligence, the CDC Foundation and the Georgia Tech Research Institute (GTRI) have published a pioneering study demonstrating how machine learning can automatically detect and classify emerging e-cigarette technologies. Published on July 9, 2026, in the journal Nicotine and Tobacco Research, the research details a novel computational pipeline utilizing Vision-Language Models (VLMs) to monitor the proliferation of "smart vapes" across digital marketplaces. The clandestine Rise of Smart Vapes E-cigarettes are continually evolving, with manufacturers introducing novel device designs, digital screens, built-in games, and Bluetooth connectivity. These features have the potential to gamify nicotine consumption, making them exceptionally appealing to youth and young adults. According to a CDC Foundation study of a nationally representative cohort, nearly a third of young e-cigarette users now utilize these "smart" devices. However, because these products are predominantly marketed online and change rapidly, they evade traditional surveillance methods, creating a quandary for public health officials. A Dual-Model Machine Learning paradigm To ameliorate this surveillance gap, GTRI scientists engineered a sophisticated, two-tiered AI architecture. The team first trained an AI-based object detection model on an augmented dataset of approximately 7,000 images sourced from open-source repositories and five online tobacco retailers. After rigorous testing on 3,920 additional images to ensure fidelity, the model successfully isolated 2,401 images containing e-cigarettes. Subsequently, the researchers deployed a Vision-Language Model (VLM)—a cutting-edge class of AI that synthesizes large language models with computer vision to process visual and textual data simultaneously. The VLM analyzed the isolated images alongside product text descriptions to automatically ascertain the presence of digital screens. The empirical results were stellar, achieving an accuracy rate exceeding 90 percent. Force Multiplication for Public Health "Monitoring online e-cigarette marketing is like a game of Whack-A-Mole, with so many new products and features popping up," articulated Kristy Marynak, PhD, senior director for Tobacco Control Initiatives at the CDC Foundation. "This study shows how machine learning techniques can shed light on the online e-cigarette marketplace and the vast quantities and types of e-cigarette products available." Dr. Hunter Morera, a GTRI researcher and the study’s lead author, emphasized that traditional manual coding methods simply cannot keep pace with the thousands of products launched monthly. By leveraging natural language processing and deep learning, the new AI pipeline acts as a force multiplier, enabling real-time data capture that can inform policymakers and regulatory agencies with immediate market intelligence. Future applications in Epidemiological Surveillance The implications of this research extend far beyond tobacco control. Charity Hilton, the GTRI research scientist leading the project, noted that while AI development is accelerating, its application to critical public health challenges remains suboptimal. By proving that VLMs can systematically identify novel features of emerging consumer products at scale, this study establishes a template for monitoring other rapidly evolving health threats. The CDC Foundation plans to integrate this AI-based process into its broader e-cigarette monitoring efforts, marking a definitive shift from reactive observation to proactive algorithmic surveillance.