Deep learning prediction of tooth extraction decisions from limited intraoral and extraoral image data

Decision making in orthodontic treatment, especially regarding irreversible tooth extractions, is a complex and controversial challenge due to its impact on facial aesthetics and long-term functional stability. Artificial Intelligence and Machine Learning are advancing dentistry, aiding in tooth extraction decisions. Here we propose a Deep Learning solution to the Dental Extraction decision problem. We compare Intraoral and Extraoral image from a Dataset, which is an important contribution of this paper, with 1720 images from 215 patients, a small number considering the typical Deep Learning data sets. We compare five well known pretrained Convolutional Neural Network applying several balancing and optimization techniques. We significantly improved the baseline, achieving an F1-Score of 89.52% and an AUC of 95.21% using VGG19, one of the best-performing architectures. Oversampling of the minority class proved to be the most effective balancing strategy. These results are promising considering the small dataset. Models trained with intraoral and extraoral views showed superior performance. The main outcome of this paper is that Deep Learning with Occlusal images have a great potential to be an excellent orthodontic clinical decision support system.