Inferring material properties in robotic bone drilling processes

Recent innovations in robotics have enabled the development of automatic bone drilling tools which allows surgeons to improve the precision of their surgical operations. However, these tools still lack valuable tactile information about the material properties of the bone, preventing surgeons from making decisions while operating. The aim of this work is to explore whether robotic drilling tools can infer bone condition on the basis of certain key measures, particularly thrust force.

To infer material properties in robotic bone drilling processes 1) a complete database of experimental operations with an automatic bone drilling tool is implemented and 2) binary logistic regression models are developed to estimate the type of material from the observed values (mainly the central tendency of the thrust force). This work compares three different materials: bovine bone specimens, porcine bone specimens and FullCure 720, which is a general-purpose resin with, a priori, much less feed resistance. The DRIBON automatic bone drilling tool developed at CEIT is used for the experiments.

The classification matrices derived using the logistic models show that it is possible to recognize a bovine bone vs. a porcine bone with a relatively high success rate rate (approximately 90%). In contrast, it is possible to recognize bone material vs. another material (in our case a resin) with a 100% of success. These results are successfully implemented in a new hand-held version of DRIBON.

We propose a method and devise a novel hand-held tool which show that robotic systems can effectively infer bone material properties.