赵阳1, 李俊诚1, 成博栋2, 牛娜君3, 王龙光4, 高广谓5, 施俊1(1.上海大学;2.西安电子科技大学;3.南京医科大学口腔医学院;4.空军航空大学;5.南京邮电大学)
Applications and challenges of deep learning in dental imaging
Dental imaging is an essential tool for the detection, screening, diagnosis, and therapeutic evaluation of clinical oral diseases, with accurate analysis of the images being vital to the development of subsequent treatment plans. Deep learning can automatically learn and obtain superior feature expressions from large sample data to improve the efficiency and performance of various machine learning tasks, which is currently widely used in many fields such as machine translation, speech recognition, and computer vision. With the integration of artificial intelligence and various fields, smart healthcare has become an important application area of deep learning, which provides an effective way to solve the following clinical problems: 1) The shortage of experienced radiologists in the field of dentistry cannot meet the rapidly growing medical demand; 2) With sufficient medical resources, experienced physicians cannot meet the rapidly growing medical demand; 3) Different physicians have different interpretations of the same oral image, which are easily influenced by subjectivity. Deep learning-based dental image processing is a topical research topic at present. Due to the inherent specificity and complexity of the medical field, as well as the problem of insufficient dental image data samples, brings new challenges to applying deep learning methods in relevant learning tasks and scenarios. This paper mainly reviews the applications of deep learning methods in various applications using three major dental imaging (i.e., two-dimensional oral X-ray images, three-dimensional tooth point cloud/mesh images, cone beam computed tomography (CBCT)). These applications include tooth segmentation, caries detection, tumor detection, etc. The reviews on two-dimensional oral X-ray images focus on bitewings, periapical and panoramic X-ray based on deep learning methods. Bitewing X-rays usually show the contact surface from the distal end of the canine to the most distal molar. They are mainly used to diagnose proximal caries, assess the extent of caries, identify secondary caries under existing restorations, etc. For caries detection using bitewing X-rays, we mainly review deep learning methods using convolutional neural network architectures, such as full convolutional neural networks and U-Net architectures. Detection of periodontitis and caries based on periapical X-rays primarily introduces methods on the basis of convolutional neural networks and backward propagation neural network. In contrast to bitewing and periapical X-rays, panoramic x-rays show not only teeth and gums, but also jaw, skull, spine and other bones. We provide a focused review of the application of deep learning methods in panoramic radiographic images from three categories of directions. Namely 1) tooth detection and numbering. 2) tooth segmentation. 3) non-dental disease detection. The three-dimensional tooth point cloud/mesh image is a digital 3D oral model obtained by scanning and reconstructing the patient"s mouth in real-time using an intraoral scanner. The reviews on three-dimensional tooth point cloud/mesh image focus on tooth segmentation based on deep learning. Deep learning methods for tooth segmentation can be divided into two categories: 1) fully supervised methods. 2) non-full supervision methods. For fully supervised methods, we mainly introduce the hierarchical network architecture model and the end-to-end network architecture model with a large amount of labeled data and annotation. While for non-full supervision methods, we primarily review self-supervised learning and semi-supervised learning methods that require only partially annotated data, as well as methods that utilize weakly annotated ideas. Currently, cone beam computed tomography imaging is a non-invasive, low-radiation technique widely used in dental diagnosis and treatment. This paper summarizes deep learning methods on CBCT images focusing on three major areas: 1) tooth segmentation. 2) dental implants. 3) oral and maxillofacial surgery. Nevertheless, from the current application of deep learning methods, it is still not yet easy for the deep learning methods already proposed to analyze all aspects of the patient"s oral health and systemic health status together in order to develop a more personalized and high-level treatment plan for oral diseases. Deep learning has made some progress in the field of dental image processing, but still faces some serious challenges. Small sample size has been a serious problem in the field of medical image analysis, which can be effectively alleviated by non-fully supervised deep learning methods such as weakly supervised learning, self-supervised learning, along with machine learning methods including migration learning, sample less learning, and incremental learning. In addition, annotation of dental medical images is a time-consuming and laborious task that relies heavily on the experience of the practitioner, which is one of the barriers limiting the widespread and deep application of deep learning. Therefore, research focusing on automatic data annotation must be extensively conducted. At Present, the development of deep learning applications in dental imaging is still in a relatively early stage. The development of this field cannot be achieved without the cooperation of computer scientists and clinicians as well as experts in imaging equipment and software development to solve the problem of how to deploy lightweight deep learning networks into convenient medical devices. In conclusion, the combination of deep learning and dental image analysis has been a major trend, with significant results in various analysis tasks, yet further in-depth research is still needed to lead the development of dental image analysis into a new phase.