Convolutional Neural Network Learning Versus Traditional Segmentation for the Approximation of the Degree of Defective Surface in Titanium for Implantable Medical Devices

Abstract

One prevalent option used in the manufacturing of dental and orthopedic medical implants is titanium, since it is a strong, yet light, biocompatible metal. Nevertheless, possible micro-defects due to earlier chemical treatment can alter its surface morphology and lead to less resistance of the material for implantation. The scope of the present paper is to give an estimate of the defectuous area in titanium laminas by analysing microscopic images of the surface. This is done comparatively between traditional segmentation with thresholding and a sliding window classifier based on convolutional neural networks. The results show the supportive role of the proposed means towards a timely recognition of defective titanium sheets in the fabrication process of medical implants.