Overview

AI-powered innovation in healthcare is accelerating. From improving diagnostic accuracy to increasing treatment efficiency and enhancing patient management, AI is being adopted across a wide range of medical applications. Dentistry is no exception. AI is now being actively used across the full dental care workflow, from diagnosing oral diseases to planning treatment.

According to Knowledge Sourcing Intelligence’s global AI in dental imaging market forecast, the market is expected to grow from approximately $270 million in 2021 to about $2.1 billion by 2028, representing a compound annual growth rate (CAGR) of 34.07%. This growth is being driven by rising demand for AI solutions that can improve the accuracy and efficiency of dental care.

Professor Heekyung Kim of Ajou University Dental Hospital reached out to Superb AI while exploring the convergence of telemedicine and AI in dentistry. This customer success story looks at how Professor Kim and Superb AI collaborated to develop a smartphone-based cavity diagnosis AI model, and how the team solved key challenges along the way.

Challenge: How to Build an AI Model with Sensitive and Limited Data

“Cavities are one of the most common dental conditions. Patients always want to know whether they have cavities or not. But many people are afraid of visiting the dentist, and access to dental care is often limited for people living in remote or underserved areas, as well as for people with disabilities. I thought that if anyone could diagnose cavities using a smartphone, it could help improve oral health for the entire population. However, medical diagnosis involves both objective and subjective judgment, which makes it extremely difficult to define the correct answers for AI training.”

— Professor Heekyung Kim, Ajou University Dental Hospital

The first challenge in developing a cavity diagnosis AI model was the sensitivity of medical data. Because the project involved images taken inside a person’s mouth, strict data security and anonymization were essential.

Another challenge was data scarcity and imbalance. To develop a multi-class model that could classify cavities by severity level, sufficient data was needed for each class. However, as more people have become attentive to oral hygiene, it has become difficult to collect enough data from severe cavity cases. When only a small amount of data is available and class imbalance is significant, AI model performance is likely to decline.

The data collection method also introduced variability. Because patients captured images of their own mouths using smartphones, image quality varied widely. Differences in device type, shooting distance, lighting, and shadows resulted in non-standardized data. Creating reliable ground-truth data for AI training, based on consistent diagnoses from experts, was also a complex process.

Solution: Completing the First Model in Three Weeks with Strong Security and Automated Labeling

“We chose Superb AI because the company already had experience developing other medical AI models. The project manager actively incorporated my feedback while learning dental terminology such as prosthetics, natural teeth, mandible, and maxilla, and helped improve the AI model. Since AI was new to me as well, I kept asking about model evaluation metrics such as precision and recall. I appreciated that the collaboration allowed both sides to bring their expertise together without conflict.”

— Professor Heekyung Kim, Ajou University Dental Hospital

 Superb AI prioritized data security from the very beginning and worked closely with the researcher throughout the project. The key solutions included the following:

1. Data Security and Rigorous Labeling Review: Superb AI Project Manager Eunseo Baek used Superb Platform to manage data labeling, AI model development, and model improvement in a single environment. The platform’s advanced security framework enabled secure handling of medical data, while the team established data collection guidelines and continuously collected additional data to address shortages. Professor Kim proposed classifying and comparing cavity severity across categories such as mild, moderate, and severe. Based on this feedback, the Superb AI team quickly developed a model and supported its use in the research process.
2. Rapid Initial Model Development and Continuous Improvement: One of Superb AI’s strengths is its efficient AI model development process, which includes data augmentation and standardization. This helped reduce the impact of variability in smartphone image capture conditions. Within just three weeks of project kickoff, the team completed an initial model capable of classifying cavity conditions into multiple classes. Because the limited amount of data created unfavorable conditions for model training, the team continued improving model performance after building the initial model.
3. Responsive Communication and Iterative Feedback: Through continuous feedback with Superb AI, Professor Kim suggested areas for model improvement, which were reflected in the model development process. The model had to learn from diagnoses made by four dental specialists selected to avoid gender and experience bias. Despite multiple rounds of feedback, Professor Kim noted that Superb AI quickly incorporated customer requirements and delivered a satisfying collaboration experience. 

(Example image by cavity diagnosis stage)

Benefit: AI Model Performance Metrics Suitable for Medical Research, Available in Real Time

“I was excited and impressed to receive a platform where I could upload data and check results in real time. It did not stop at AI model development. I could use it directly in classes and lectures, and because I could immediately evaluate the model’s diagnostic capability, it was extremely useful. Smooth collaboration between dental experts and AI experts is key to improving AI, and Superb AI supported that process very well.”

— Professor Heekyung Kim, Ajou University Dental Hospital

(Superb Platform with an AI model capable of real-time cavity diagnosis)

Through the collaboration with Superb AI, Professor Kim was able to produce research outcomes in a much shorter period than would have been possible with conventional experiment-based research. She noted that if data is well standardized and a sufficient amount of data is secured, Superb Platform makes it possible to build AI models quickly. The project delivered the following results:

1. Rapid Model Development and Performance Validation: In traditional dental materials or biology research, experiments often continue for an entire year. Even after data is collected, statistical issues can sometimes require researchers to restart the experiment from the beginning. With Superb AI, Professor Kim was able to build the first AI model protocol in just three weeks and continue improving it, significantly accelerating the research timeline. For enterprises, this also demonstrates how quickly a proof of concept (PoC) can be tested before launching a large-scale Vision AI project.
2. AI Model and Performance Metrics Suitable for Medical Research: Key AI performance metrics for research papers can be checked directly within the platform and compared at a glance through the dashboard. Professor Kim evaluated and validated the accuracy of smartphone-based cavity diagnosis using key metrics provided by Superb AI, including F1 score, mAP, Precision, and Recall. (* Featured post: Understanding Model Diagnosis Metrics and Applications)Superb Platform provides objective and accurate performance metrics that are suitable for use in medical research papers.
3. Real-Time Data Utilization: Through the endpoint platform provided by Superb AI, Professor Kim was able to input research data in real time and immediately check the results. Superb AI’s intuitive and user-friendly platform can be used not only for real-time cavity diagnosis, but also in patient health education sessions and lectures for dental professionals.

(AI model performance metrics available in Superb Platform)

Conclusion: New Possibilities for Medical AI Through Expert Collaboration

This project laid the foundation for developing a smartphone-based cavity diagnosis AI model and expanded the possibilities of medical AI technology. The collaboration between Superb AI and Professor Heekyung Kim went beyond technical development. It demonstrated the importance of mutual learning and close collaboration between researchers and technology companies.

Superb AI looks forward to contributing to the advancement of digital healthcare through its innovative technology and platform.