A Neural Network-Based Framework of Real-Time Heat Map Generation for Surgical Hand-Held Drilling

Abstract

Understanding heat generation can help improve one’s surgical drilling skill to avoid thermal injury. Surgical drilling is mostly done manually, so it can be time-consuming to create personalized thermal models to assess each drilling. For this reason, this paper presents a framework for 2D real-time heat map generation for a moving, varying heat source problem based on neural networks (NN) and linear time-invariant system (LTI). In this framework, several location-specific heat maps and their temporal responses are calculated by finite element analysis (FEA) and trained through NN to build a surrogate model. The total heat map of any given moving heat source can be generated by the superposition of a series of location-specific heat maps along the moving path. The NN training shows a correlation over 99%, indicating a highly representative surrogate model. The validation study of comparing two FEA-based moving heat source problems with the framework predicted results show overall good agreement. Error sources and improvement methods are discussed in this paper.