New study advances AI diagnosis of MIH

Dental Tribune International

Mon. 23. September 2024

save

MUNICH, Germany: Molar incisor hypomineralisation (MIH) constitutes a global oral health issue, affecting around 14% of the world’s population. Like with other areas of oral health, the detection of MIH is being revolutionised through the utilisation of artificial intelligence (AI) models. Building on a groundbreaking study that introduced an open-access AI model for the detection of MIH using digital photographs, a new study by a group of researchers based at the Department of Conservative Dentistry and Periodontology at LMU Munich has externally validated the model, thereby bolstering confidence in its clinical efficacy.

The model was first developed on the basis of 18,719 photographs of teeth with 34,710 pathological findings. The images were then analysed by trained dentists according to internationally accepted classification frameworks. As described in the initial paper, the model yielded remarkable high internal validity, but the authors cautiously emphasised that the model needed to be further improved and externally validated.

Taking this lead, the validating researchers tested the model using images freely accessible from internet searches. A total of 455 images were evaluated by a working group of five dentists, and they determined that 277 images showed teeth with MIH on occlusal or smooth surfaces and 178 images showed teeth without MIH. When these images were subsequently subjected to analysis by the AI model, the results were highly illuminating: the model achieved an overall accuracy of 94.3% for image-based detection of MIH.

Speaking to Dental Tribune International, co-author Prof. Jan Kühnisch, head of the section of paediatric dentistry at the department, commented on the significance of the study. “Dental photographs—which have to be understood as the digital and machine-readable equivalent to the clinical examination—can be automatically evaluated by AI methods and may potentially contribute to accurate diagnostic evaluations in the future. When considering that images can be captured by various intra-oral cameras, semi-professional cameras or even mobile phones, it’s a small step to be analysed by AI algorithms and receive a fast and potentially cheap dental diagnosis.”

This naturally raises the question as to how such AI-assisted diagnosis stands to transform the dental profession. Prof. Kühnisch continued, “I think that such AI models will be integrated in dental software apps step by step and will support professional work. It’s similar to other dental developments, such as self-etching adhesive. There was a vision long ago, but it took one or two decades to make the performance sufficiently stable so that it could be used in daily dental practice.”

Like other areas of dentistry being transformed by the uptake of AI, a fundamental issue will remain how scientifically accurate these models are and also how they can be integrated meaningfully alongside human evaluations and judgements.

The study, titled “External validation of an artificial intelligence-based method for the detection and classification of molar incisor hypomineralisation in dental photographs”, was published in the September 2024 issue of the Journal of Dentistry.