Irregular Motion Recognition of Guidewire in Vascular Interventional Surgery

Abstract

With the increasing incidence and prevalence of cardiovascular and cerebrovascular diseases, cardiovascular and cerebrovascular diseases have become one of the biggest causes of human death, and the diagnosis and treatment of cardiovascular and cerebrovascular diseases are mainly completed by minimally invasive interventional surgery. The operation requires high safety, high control precision, and high level of experience of surgeons. However, in the actual cardio-cerebrovascular interventional procedure, due to the complexity of the intravascular environment and doctors’ operation of the guidewire with reference to the 2D contrast image, the guidewire might present some irregular motion with risks, such as bounce, which is very dangerous for cerebral blood vessels. Therefore, it is very critical to locate the bounce motion accurately. In this paper, transfer learning on VGG16 is used to train the bouncing motion data collected by the vascular interventional surgery robot. The problems of less bouncing motions and limited data volume affect the recognition of guidewire bouncing motions. Therefore, by processing, refining and integrating the guidewire bounce data, the accuracy rate of the guidewire bounce action classification in the test set is increased to 93.2%, which realizes the accurate recognition of the guidewire bounce action.