Prediction of Geometric Features in 3d Soft Tissue Images using Pre-Trained Network Model

Abstract

Recently, a deep learning method for predicting three-dimensional facial soft tissue landmarks, an essential tool in dentistry, was developed. In its most basic form, deep learning depends on converting 3D models into 2D maps, which compromises accuracy and information. This research presents a 3D face soft tissue model-based region cut-off and feature analysis with an image scaling index method (ISIM)that can recognize landmarks correctly. An object detection network's initial step is to ascertain the range of each organ. Second, the prediction networks acquire landmarks by using the 3D models of various organs. The experimental results are compared to other learning algorithms or geometric information methods. However, the proposed model gives a Mean error in local trials of 1.8, which is lower. Additionally, 100% of the test data's mean error falls within this range, and over 72% of it does so within 2.5. Furthermore, our method predicts 32 landmarks, more than any previous machine learning-based method up to this point. Based on the observations, the results demonstrate the practicality of using 3D models directly for prediction by showing that the recommended approach can properly predict many 3D facial soft tissue landmarks.