Automated Tooth Segmentation in X-ray Images using Attention Integrated U-Net++ Model

Oral functioning, the ability to perform various oral functions, pain-free and without being ill, is one of the indicators of overall well-being and life satisfaction. Even though these diseases are mostly preventable, a majority of individuals all over the world suffer from periodontal disease, dental caries, and oral malignancies to an estimated of 5 billion people worldwide. However, there are still use of traditional radiographic techniques like periapical and panoramic radiography that assist in diagnosis. A better diagnosis is enhanced through the use of modern techniques in imaging such as MRI, ultrasonography and CBCT imaging. Deep learning-based solutions enhance diagnostic capabilities through dental segmentation in images since they learn from unmodified data in contrast to traditional machine learning and deep learning techniques which produce inferior results when dealing with massive datasets. Therefore, in an effort to solve the problem of efficient dental segmentation, anencoder-decoder basedtooth segmentation model based on U-Net++ architecture modified using Efficient Channel Attention Network is proposed that aims at enhancing the accuracy of dental diagnosis. The paper also focus of the performance of different state-of-art CNN techniques namely, U-Net, FCNs, ENet, U-Net++, U-Net 3+, SwiftNet on the standard Tuffs Dental dataset for model selection and comparisons. The proposed model outperforms other segmentation models and achieves the best mean average precision of 96.4% and a dice coefficient of 90%.