2D to 3D Neurovascular Reconstruction from Biplane View via Deep Learning
Jingyi Zuo · 2021 2nd International Conference on Computing and Data Science (CDS) · 2021
Three-dimensional visualization of vessels from digital X-Ray angiogram expands the current application of angiography technique. Conventionally, such techniques require a high dose of radiation exposure and long processing time; therefore, seeking a safe and instant reconstruction method of vessels is important. This paper aims to propose a framework of reconstructing the neurovascular model with biplane X-Ray angiography as inputs. The model is generated from a well-trained deep learning model by which effectively reduce the time spent during reconstruction. We implement an adversarial network, consisting of a U-Net and a standard convolutional neural network, to collaborate with the X-Ray angiography as label and the ground truth as training output. The result of reconstruction is evaluated on several quantitative metrics for medical imaging segmentation models. Experimental result shows that our model improves the prediction performance while preserving computational efficiency.