Deep Learning-Based Microcatheter Localization in the Brain for Pretreatment Planning of Arteriovenous Malformation Embolization

Endovascular catheter use for medical interventions has increased due to shorter post-procedural recovery times and a relatively low complication rate. Endovascular interventions are one of the preferred treatments for arteriovenous malformations (AVMs). An AVM is an abnormal vascular structure that irregularly connects arteries and veins, occurring most commonly in the brain. However, navigating a catheter close to the nidus of an AVM is often challenging because of limited visualization, complex vessel structure, and limited manoeuvrability of the guidewire[1]. Locating the catheter within 3D vascular maps is still difficult in current AVM procedures[2]. Therefore, this research focused on providing a detection of the microcatheter inside the brain using 2D X-ray angiography (XRA) images obtained during interventions. The 2D catheter localization can then be mapped into the 3D segmented brain vessels, helping clinicians better navigate the catheter to the target. A deep learning-based approach is proposed to localize the microcatheter. Small patches within vessel structures were extracted from the XRA images, as the vessels represent the region of interest (ROI) for microcatheter presence. The ROI is selected using Digital Subtraction Angiography (DSA) images, showing contrast-enhanced blood vessel structures during the procedure. The patches were then classified using a convolutional neural network to determine the presence or absence of the microcatheter, based on visual features extracted from XRA images. For now, the model was trained on XRA data from 17 patients and evaluated on a different test set of two patients. A post-processing algorithm was implemented to reduce false positives based on the connectivity of positive patches, ensuring that isolated detections are removed. Our proposed approach demonstrates potential in detecting the microcatheter, even with limited data. Further optimization and validation within brain vascular maps is needed for precise 3D catheter localization. [1]Schimmel, K. et al.Int J Mol Sci.2021 [2]Liu, C. et al.Front Neurol.2023