There are two major categories of artificial intelligence: narrow AI, which is designed for specific functions, and general-purpose AI for everyday use, such as ChatGPT. We have developed a specialized narrow AI system for screening osteoporosis using panoramic X-ray images. This system, called PanoSCOPE, has been approved and implemented as the first medical AI device for dentistry in Japan. At the same time, there is increasing interest in how general-purpose AI can be incorporated into dental practice. PanoSCOPE is a software-as-a-medical-device that evaluates the mandibular inferior cortex and automatically determines the mandibular cortical index (MCI). It replicates the diagnostic workflow of expert oral and maxillofacial radiologists for visual MCI assessment. The system was trained on expert-annotated image datasets and was validated using images independently evaluated by other experts for MCI classification. The explainable nature of the system—based on predefined regions of interest and explicit diagnostic criteria—played an important role in obtaining regulatory approval. Moreover, combining traditional rule-based processing with AI-driven functions appears necessary for practical clinical devices. In contrast, general-purpose AI systems, such as ChatGPT, are pretrained on large-scale internet-derived datasets and can be used simply by prompting. Because general-purpose models evolve rapidly, their potential medical applications may outpace traditional research and development cycles. For example, when DICOM files are submitted to ChatGPT, tag information that was unreadable in earlier versions has become interpretable in newer versions. Such systems can also assist with evaluating intraoral photographs and evaluating tongue coating according to the Oral Health Assessment Tool (OHAT) criteria. Furthermore, by providing radiographs with marked target lesions along with clinical symptoms, they can generate structured imaging reports that are clinically convincing.
However, the responses generated by general-purpose AI are not always correct. Medical professionals should therefore use such systems cautiously and avoid placing unwarranted trust in non-validated outputs.