Dr. Michael Glogauer (MG): To start with a big picture perspective, I think there’s this fear in every realm of the economy that AI could replace people. But I don’t think that’s what will actually happen. Instead, people who use AI will replace people who don’t use AI. I’ve witnessed how much the Internet has changed almost every aspect of our daily lives. AI will do the same, but in a larger order of magnitude. AI could be wonderful for dentistry from technological and educational standpoints. Dr. Mohammad Moharrami (MM): As a PhD student, when I started exploring research in the application of AI in dentistry, I observed that dentistry is following in the footsteps of medicine regarding data sources and algorithms, but obviously focusing on different outcomes. Some commercial applications have been built that use AI to help with diagnosis using radiographs, but there wasn’t much research that used oral photographs as a source of data. And oral photographs could be a great source of data if we want to focus on prevention and democratize the use of this technology for both clinicians and the public. Dr. Sonica Singhal (SS): Dental caries is the most prevalent chronic disease globally, resulting in a significant burden. Enhancing the treatment and prevention of this common condition has the potential for improving the quality of life for literally billions of people. We believe that using AI with a large data source of oral photographs could be a very powerful public health tool for oral health care providers. It has the potential to enhance preventive care and improve oral health outcomes on a population scale through early detection, remote assessment, and education through personalized feedback. I use the word “tool” intentionally; while AI cannot replace the work of oral health care providers, it can serve as a valuable resource. Our systematic review aimed to assess the effectiveness of photographs in detecting dental caries. How reliable is it? Does it generate false positives? With AI, the quality of the tool is dependent on the quality of the data it is trained on. MG: In dentistry, there’s a rich source of non-invasive data that’s both professionally procured and made by patients themselves. And if you’re able to harness these photographic images, you can generate tools for screening for dental caries that could be of incredible value, especially for people who can’t access dentistry as readily as others. This kind of tool could be very beneficial to public health. MM: We screened more than 3,000 studies as a part of this review and selected 19 papers that met the criteria of being directly pertinent to using AI for detecting dental caries from oral photographs. Some used smartphone photos as their main source, while others used professional or intraoral cameras. Compared to the annotated images by dentists, the experts who made the diagnosis in the first place, AI models’ performance was acceptable according to the reported metrics. AI that used images from intraoral cameras were better than those from professional cameras, which in turn were better than smartphones. This makes sense because the quality of the photos is simply better. As a dentist, too, we prefer to look at the highest quality image. When it comes to AI algorithms, readers should note the distinction between one-stage and two-stage algorithms. Two-stage models take longer to train and provide inference but are generally more accurate. They also need more computational power and resources, so it isn’t something usually deployed on portable devices such as smartphones, at least at the time of our article publication. Most studies trained AI models using image annotations based solely on dentists’ diagnoses from images, not In dentistry, there’s a rich source of non-invasive data that’s both professionally procured and made by patients themselves. And if you’re able to harness these photographic images, you can generate tools for screening for dental caries that could be of incredible value. 21 Issue 1 | 2025 | Issues and People
RkJQdWJsaXNoZXIy OTE5MTI=