AI in Dentistry: Reading Intraoral Radiographs

OVERFIT: AI, Machine Learning, and Deep Learning Made Simple

This article describes a clinical validation study that investigates the effectiveness of a deep learning algorithm for detecting dental anomalies in intraoral radiographs. The algorithm is trained to detect six common anomaly types and is compared to the performance of dentists who evaluate the images without algorithmic assistance. The study utilizes a paired data approach where each image is evaluated twice by the same dentist, once with and once without the algorithm. The researchers employ statistical analysis, including McNemar's test and the binomial hypothesis test, to assess the algorithm's impact on sensitivity and specificity. The results demonstrate a significant increase in sensitivity and a slight decrease in specificity when the deep learning algorithm is used for diagnostic guidance. Additionally, the area under the localization ROC curve (AUC) also shows a significant increase, further supporting the algorithm's effectiveness. The study concludes that the deep learning algorithm significantly enhances the detection of dental anomalies, providing valuable diagnostic assistance for dentists.

Read more: https://arxiv.org/abs/2402.14022v1