Designing a Subsystem for Creating a Three-dimensional Model of an Orthopedic Insole Based on Data from a Laser 3D Scanning of the Patient's Feet

• Published in: 2020 International Conference on Engineering Management of Communication and Technology (EMCTECH) pp. 1 - 6
• Main Authors: Voronov, V. I., Dovgolevskiy, P. A.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 20.10.2020
• Subjects: artificial intelligence methods; data acquisition; data preprocessing; dimensionality reduction; GAN; Generative adversarial networks; generative adversarial neural network; Generators; laser 3D scanning; Measurement by laser beam; medical computing; modified generative adversarial network; neural nets; neural network modeling; optical scanners; orthopedic insole; orthopedic insoles; orthopedic structures; orthopedics; orthotics; Pix2Pix; software architecture; software module-data loader; software package; Software packages; Solid modeling; solid modelling; specialized 3D scanner; Three-dimensional displays; three-dimensional model; Training
• DOI: 10.1109/EMCTECH49634.2020.9261530

The article discusses the use of artificial intelligence methods for the design of orthopedic structures intended for the therapy and treatment of various pathologies of the human skeleton. The development of a software package for creating a three-dimensional model of an orthopedic insole based on an image of a patient's foot obtained using a specialized 3D scanner is described. The model was built using a modified generative adversarial network (GAN) Pix2Pix. This type of networks is used for the first time to obtain medically significant results in the field of orthopedics. The problems associated with obtaining a dataset suitable for neural network modeling are considered. Methods are described that make it possible to bring this set to a form suitable for training and further use in neural network modeling. A software module-data loader has been designed and implemented, which allows converting a three-dimensional model into a numpy array of a depth map with subsequent use as a training set. For the preprocessing of the original images, the authors used a wide range of methods, including dimensionality reduction, noise suppression (Gaussian filter). The results of neural network modeling of an orthopedic insole are presented. The presence of a three-dimensional model of the insole will allow it to be made on an industrial milling machine or printed on a printer using specialized medical material.