Cranio-maxillofacial post-operative face prediction by deep spatial multiband VGG-NET CNN
Rizwan Ali,Rui Lei,Haifei Shi,Jinghong Xu
2022-01-01
American journal of translational research
Abstract:Current plastic and reconstructive surgery computational techniques are not precise and take a longtime to perform. Therefore, these limitations reduced the adoption of computational techniques. Although computer-aided surgical preparation systems may help to enhance clinical results, minimize operating time and costs, they are too complicated and require detailed manual information, which restricts their usage in doctor-patient communication and clinical decision-making. In order to obtain the optimal aesthetic and reconstruction treatment results, these techniques must be designed and implemented carefully. Computer-aided modeling, planning, and simulation techniques enable the preoperational evaluation of various therapeutic strategies based on the 3D patient models. We offer the new deep-learning architecture for diagnostics, risk stratification, and post-operative simulation for face prediction. Initially, preprocessing was done by using the weighted adaptive median filter and Laplacian partial differential equation-based histogram equalization. Then the target area was converted to 3D for clear visualization by using the Smart restorative frustum model. Finally, the post-operative face prediction was constructed by using the deep spatial Multiband VGG NET CNN. We obtained a face dataset of 313,318 CT and their clinical records from different centers. The algorithms were developed by 21,095 scans (Qure25k data set). In addition, CQ500 datasets from various centers were compiled in two batches, B1 and B2, to validate the algorithms clinically. Four hundred ninety-one scans used the CQ500 dataset. Initially, we reconstructed the input image and then devised the post-operative face computationally. The suggested deep spatial Multiband VGG NET CNN showed the high range of post-operative face prediction accuracy. Therefore, successful metrics such as the Jaccard and dice scores have shown accurate outcomes compared to other traditional methods. MATLAB was used to obtain the output of proposed work. With the help of the suggested classifier, the prediction accuracy was 93.7%, sensitivity was 99.9%, and specificity was 99.8%, all of which were higher than traditional approaches. Here, the suggested method provides better results for post-operative face prediction to the applied dataset than any other existing mechanisms. It is a generalized attempt that can apply to other similar datasets as well.