Bone Drilling Vibration Signal Classification Using Convolutional Neural Network to Determine Bone Layers

In orthopedic surgery, the bone drilling task is the main factor for the successful completion of the surgery. The bone drilling task depends on a high level of dexterity and experience of the orthopedist and surgeon. It is because the bone drilling resistance is relatively large and sometimes violent vibrations might cause difficulties in grasping the hand-piece. In a worse case, it might even break the slender drill. The objective of this paper is to gain an understanding of the frequency properties of bone in order to improve visualization and training. These properties can be detected using different imaging methods or techniques for processing signals. The experimental setup includes the robotic arm to provide an accurate thickness layer and consistent penetration of the drilling. Three-axis accelerometer equipped with National Instrument data acquisition (DAQ) was used in the experiment to acquire the vibration signal on different bone layers. This study proposes a successful approach to categorize bone drilling levels using a Convolutional Neural Network (CNN) with a customized architecture designed for this purpose. The CNN is utilized to classify the raw vibration signal into three distinct labels or layers, namely periosteum, first cortical, and spongy. The results of the study indicate that the CNN can accurately classify the three bone layers, with a higher degree of accuracy for the periosteum and first cortical layers, achieving over 98% accuracy, and a 100% accuracy for the spongy layer. This is due to the unique vibration signal of the spongy layer, which differs from the other two layers.