Position Weighted Convolutional Neural Network for Unbalanced Children Caries Diagnosis

Panoramic radiograph is one of the most widely used inspection tools for dentists making caries diagnosis, especially teeth that are hard to be diagnosed through visual inspection. Recently, several deep learning methods, e.g., based on convolutional neural network (CNN) or transformer network, have been proposed automatic diagnosis dental panoramic radiographs, and promising results achieved. However, current approaches use all equally when training their models, which in performance degeneration because unbalanced classification difficulties different tooth positions. The objective this study introduce a position weighted CNN alleviate above problem more accurate diagnosis. module evaluates revises output specially designed incorporate information. In addition, novel data augmentation method balance with uneven decayed normal teeth, reasons leading difficulty. To verify method, children database collected labeled than 6,000 teeth. approach outperforms state-of-the-art methods accuracy, precision, recall, F1 area-under-the-curve being 0.8859, 0.8875, 0.8932, 0.8903 0.9315, respectively. Specially, model displays higher compared two attending doctors five-year clinical experience but patterns, showing potential tool assisting dentists.