ANALYSIS OF CONTACT POINTS ON GUIDEWIRES IN NEUROINTERVENTIONAL SURGERY USING U-NET-BASED SHAPE RECONSTRUCTION

A study is currently underway to enhance a robotic system tailored for assisting surgeons during neuroendovascular interventions. This system relies on a guidewire to navigate through the vascular artery network. However, due to the guidewire's high flexibility, locating the contact points on it poses a challenge, which will lead to safety concerns and imprecise navigation outcomes for the robotic system. To address this issue, this research proposes a shape-based force estimation method to precisely locate the contact points on the guidewire. This method uses the guidewire's contour in the image as the shape information and the guidewire's mechanical property to establish the relationship between shape change and applied loads. To extract the guidewire's contour from the background, a convolutional neural network, specifically a U-Net model, was employed. Ninety-six images were used for the U-Net model training. The U-Net model achieved a Dice score of 82.8%, indicating its effectiveness in contour extraction. Based on the extracted contour, a thinning method was used to extract the mid-line of the guidewire, which can be used to calculate the curvature of the guidewire. By analyzing the curvature change, the localized maximum value of the curvature indicated the location of the applied load. An experimental validation of the proposed method was conducted. The relative error of the load location estimation was approximately 6.5%, demonstrating the accuracy of the proposed approach.