Improved classification of medical orthoses based upon the point clouds algorithms

In medical rehabilitation, point clouds algorithms, as practical tools for 3D model analysis, possess significant advantages in orthoses processing and design. In this paper, an orthosis point clouds classification network with downsampling and data augmentation modules was proposed to classify a large number of the 3D orthoses with complex surfaces before inputting them into an expert template library which uses the previous orthoses to help the customized orthosis design for new patients. Initially, the effects of three types of the basic network were investigated to obtain the optimum basic network. Then, two kinds of data augmentation modules and four kinds of down-sampling modules were respectively added to the optimum basic network in order to obtain the best comprehensive network. The experimental results show that the basic classification with appropriate down sampling and data augmenta3tion methods can effectively address the time-consuming and low accuracy of the existing networks and reduce the orthosis classification time by 12.83% and improves the classification accuracy by 4.29% on average.