Computational Dentistry With 3d Point Cloud Segmentation

Republished By Plato

Followers: 0

Modern dentistry has undergone noteworthy changes owing to the technical advancements in the field of machine learning-backed AI models. Upgraded imaging methods have now been incorporated in dentistry to ensure heightened efficiency levels and providing a reliable experience to customers. Significant improvements can be seen with treatment planning and diagnostic changes using computation dentistry that encompasses intra-oral and extra-oral optical imaging; herein, the use of data is also evidential for the machine learning-backed AI models enabled with point cloud labeling data

The intra-oral scanners imaging devices use light for capturing the surface of the anatomical structure of a patient’s mouth and the project pattern of the mouth is measured by imaging sensors; creating an accurate 3D point cloud. Obtained 3D point cloud shows the geometrical profile of tooth and gingiva in high spatial resolution (30-80 points per mm2) and equally high spatial accuracy. For AI implementation, this 3D point cloud data is further used for orthodontic planning and treatment planning in modern dentistry. This also enables providing a detailed view of the anatomical structure of the clinical dental application.

AI and Deep neural networks in modern dentistry

The same methodologies have been used so far for the segmentation of individual teeth using imaging sensors and 3D point information. Such a methodology is IOS Segmentation methods and deep learning approaches are also used in the implementation of the same. Under the OS Segmentation methods, the projecting the 3D IOS mesh on 2D plain and applying computer vision algorithm on the obtained data, which is used for projection in 3D space. On the other hand, the deep learning approaches include two methodologies for segmentation.

Each methodology is then viewed as a separate set of approaches to teeth segmentation. In the first approach, the teeth are taken as a multi-class segmentation problem and every tooth is viewed as a separate semantic class. While in a second way, the tooth is segmentation is done as per the semantic instance segmentation problem. Simultaneously, a deep learning model for instance segmentation is also used for teeth segmentation, wherein a hybrid framework is used. This framework used 2D image data for detection and project 3D point cloud information in the preceding stage.

Furthermore, a new instance of segmentation model based on deep learning Mask MCNet is applied to irregular 3D point cloud information to predict 3D bounding box for object instances. Mask MCNet is an end-to-end deep learning framework, assuring maximum accuracy in computational dentistry with diverse clinical applications.

Along with this, in another instance, deep neural networks were utilized for detecting caries lesions on annotated data of single tooth segments. Converting the annotation in binary class level, CNNs – Resnet18, Resnext50 models were trained to predict the neural networks. Both the models successfully detected areas on teeth affected by caries lesions.

Computational dentistry: What’s ahead

Dentistry, as a branch of medicine, is changing form in the 21st century. More than detecting common dental deformities or decay issues, in the past few years, the increased application of Artificial intelligence has accelerated research and development of advanced procedures for patients. Machine learning approaches like computer vision are being endorsed widely for predicting treatments and active diagnosis of various dental diseases.

Driven by imaging data, deep neural networks like Mask MCNet are also proving efficient in predicting diseases and related treatment with precision. The scope for investigation and finding out results with relation to dental conditions in dental subfields like dentistry, cariology, endodontics, periodontology, orthodontics, and forensic dentistry. A wave of augmented intelligence, courtesy progress in AI, and Deep Learning is making use of computational processes to build models which can provide results and improve dentistry as a practice, in the coming years. Such

progression will also be evident in handling the patient load, writing elaborative observations on paper, and performing a multitude of tasks on suggestive dental treatments