Artificial Intelligence Methods in the Detection of Oral Diseases on Pantomographic Images-A Systematic Narrative Review.

Artificial intelligence (AI) is playing an increasingly important role in everyday dental practice and diagnosis, especially in the area of analysing digital pantomographic images. Through the use of innovative and modern algorithms, clinicians can more quickly and accurately identify pathological changes contained in digital pantomographic images, such as caries, periapical lesions, cysts, and tumours. It should be noted that pantomographic images are one of the most commonly used imaging modalities in dentistry, and their digital analysis enables the construction of AI models to support diagnosis. This paper presents a systematic narrative review of studies included in scientific articles from 2020 to 2025, focusing on three main diagnostic areas: detection of caries, periapical lesions, and cysts and tumours. The results show that neural network models, such as U-Net, Swin Transformer, and CNN, are most commonly used in caries diagnosis and have achieved high performance in lesion identification. In the case of periapical lesions, AI models such as U-Net and Decision Tree also showed high performance, surpassing the performance of young dentists in assessing radiographs in some cases. The studies cited in this review show that the diagnosis of cysts and tumours, on the other hand, relies on more advanced models such as YOLO v8, DCNN, and EfficientDet, which in many cases achieved more than 95% accuracy in the detection of this pathology. The cited studies were conducted at various universities and institutions around the world, and the databases (case databases) analysed in this work ranged from tens to thousands of images. The main conclusion of the literature analysis is that, thanks to its accessibility, speed, and accuracy, AI can significantly assist the work of physicians by reducing the time needed to analyse images. However, despite the promising results, AI should only be considered as an enabling tool and not as a replacement for the knowledge of doctors and their long experience. There is still a need for further improvement of algorithms and further training of the network, especially in identifying more complex clinical cases.