Deep learning for guidewire detection in intravascular ultrasound images

Abstract Algorithms for automated analysis of intravascular ultrasound (IVUS) images can be disturbed by guidewires, which are often encountered when treating bifurcations in percutaneous coronary interventions. Detecting guidewires advance therefore help avoiding potential errors. This task is not trivial, since appear rather small compared to other relevant objects IVUS images. We employed CNNs with additional multi-task learning as well different guidewire-specific regularizations enable and improve guidewire detection. In this context, we developed a network block generates heatmaps that highlight without the need localization annotations. The detection results reach values 0.931 terms F1-score 0.996 area under curve (AUC). Comparing thresholded ground truth segmentation masks leads Dice score 23.1 % an average Hausdorff distance 1.45 mm. Guidewire has proven handle quite well. Employing further generation indicate position actual labels.